mbfinotti/partnerships-skillsチェック済み
SKILL DETAIL
influencer-measurement-framework
mbfinotti/partnerships-skills/influencer-measurement-framework
Design the pre-launch measurement plan for one influencer or creator campaign - KPIs matched to the objective (awareness, engagement, conversion, affiliate-linked, always-on ambassador), an attribution method and data source behind each KPI, reporting cadence per stakeholder, and declared blind spots. Delivered as a measurement plan document before launch, for B2B and B2C alike. Use whenever the user mentions influencer KPIs, creator campaign attribution, campaign reporting, or asks how to measure a creator partnership, even if they never say measurement. Do NOT use for an ongoing affiliate program dashboard - use mbfinotti/partnerships-skills@affiliate-performance-dashboard instead.
インストール · 498出典を見る
Installation
npx skills add https://github.com/mbfinotti/partnerships-skills --skill influencer-measurement-framework
スキルファイル
SKILL.md
最終同期 · 2026/10/23
evals/evals.json›
{
"skill_name": "influencer-measurement-framework",
"evals": [
{
"id": 1,
"prompt": "Our agency just sent the reporting section of their proposal for the spring flight and I have to take it to our CFO on Friday. Brand is Halden & Co, DTC skincare, this is the first time we've ever worked with creators. The flight is 8 micro-creators on TikTok and Reels, $40K, six weeks. What they've put down is: EMV target of $250K, projected 5.78x return based on industry data, and a 4.5% engagement rate target which they say is the industry benchmark. Can you turn this into the actual measurement plan we'd run?",
"expected_output": "A pre-launch measurement plan document that strips the EMV figure, the 5.78x projection and the borrowed 4.5% ER target, replaces them with a recomputable glossary and a first-flight baseline-setting posture, and does not invent tracking capabilities the brand has not confirmed.",
"files": [],
"expectations": [
"Refuses to carry the $250K EMV figure into the plan as a financial, revenue, or ROI value",
"States that EMV and MIV are AVE derivatives whose only legitimate use is relative, directional comparison inside one vendor's own methodology",
"Cites the AMEC Barcelona Principles, specifically that AVEs are not the value of communication, as the basis for rejecting EMV as value",
"Rejects the 5.78x figure as a planning input and identifies it as a vendor benchmark with weak or inconsistent provenance",
"Rejects the 4.5% engagement-rate figure as a target because vendor benchmarks are calibration context, not targets",
"Sets no numeric performance target and labels this a first, baseline-setting flight because the brand has no trailing history of its own",
"Names which engagement-rate denominator the plan will use (reach, impressions, followers, or views) and commits to holding it constant for the whole campaign",
"Notes that the same post reads very differently depending on which engagement-rate denominator is chosen",
"Delivers a measurement plan document containing at minimum glossary, KPI tree, attribution, baseline, test design, cadence, data sources and blind spots sections",
"Gives each KPI a definition that states numerator, denominator and window",
"Does not assume any tracking capability the user has not confirmed: either asks which of codes, UTM links, affiliate network, pixel, post-purchase survey or CRM field exist, or records them as unconfirmed open items",
"Labels the provenance of figures it carries forward (standards-body, platform-published, vendor-sourced, practitioner, or the brand's own history)"
]
},
{
"id": 2,
"prompt": "Northwind Audio is launching a new pair of over-ear headphones. This is a pure brand push - 3 YouTube tech reviewers and 2 podcast hosts, $180K, eight weeks, and the goal is simply that more people in our category have heard of us. Two constraints from above: our VP of Growth says the headline number in the readout has to be ROAS so it sits next to our paid search dashboard, and he wants the YouTube brand lift study because it's the cheapest one to run. The readout goes to the board, who will use it to decide whether to double our category spend next year. Put the measurement plan together.",
"expected_output": "A measurement plan that refuses ROAS as the primary KPI for an awareness objective, substitutes awareness-stage KPIs, makes pre-launch baseline capture mandatory, and escalates the brand-lift choice above a platform-run study because the board decision must survive the seller grading its own work.",
"files": [],
"expectations": [
"Refuses ROAS as the primary KPI and states the reason: the primary KPI's funnel stage must match the campaign objective's funnel stage",
"Proposes an awareness-stage primary KPI set instead, drawn from reach, impressions, CPM, video views, or brand lift",
"Explains the refusal to the user rather than silently complying with the ROAS instruction",
"Makes pre-launch baseline capture mandatory and names trailing branded-search volume and direct traffic as what gets captured",
"States the consequence of launching without that baseline: post-campaign lift claims will be unsupported",
"Removes geo-lift and platform conversion-lift holdout from the test-design menu on the grounds that an awareness campaign with no conversion volume cannot use them",
"Identifies brand lift as the only instrument that measures this campaign's stated objective",
"Flags the self-measurement conflict in a platform-run brand-lift study: the platform that sold the media grades its own work",
"Recommends independent-panel brand lift, or names the condition requiring it, because the readout drives a budget-size board decision",
"Notes that brand-lift studies need roughly 1-4 weeks of in-flight surveying and that short flights return inconclusive results",
"Does not equate a platform view count with an MRC-viewable impression, and states the MRC bar (50% of pixels for 1 second display, 2 seconds video)",
"States that podcast placements have no native brand-side analytics access and must be measured through codes, vanity URLs and publisher-reported numbers"
]
},
{
"id": 3,
"prompt": "Cedarline is a meal-kit startup. We ran one creator flight last quarter - 5 Instagram creators, roughly 380 attributed orders, $55K spend. Finance now thinks the whole creator line item is wasted money. To buy time I told them we'd run a geo-lift test this quarter across our 6 shipping regions to prove incrementality, and the result has to be in the board pack in five weeks. Write the measurement plan so I can show them the methodology.",
"expected_output": "A measurement plan that runs the MDE power check first, concludes the program cannot power a geo-lift test at this volume and on this timeline, writes directional-only into the plan, and substitutes the baseline-anchored branded-search series with correlational framing.",
"files": [],
"expectations": [
"Runs, or requires as the first step, a Minimum Detectable Effect power check before any commitment to the geo-lift test",
"States the decision rule: if the computed MDE exceeds the lift the business would actually act on, the test is not run",
"Concludes that a program at this conversion volume cannot power the test and writes \"directional only\" into the plan instead of promising incrementality",
"Explains that an underpowered test returns \"no significant lift\" for a channel that genuinely works, and a working channel gets cut on that result",
"References the practitioner MDE target of roughly 5-10% over a 4-8 week window",
"Notes that the five-week deadline alone rules out geo-lift, which needs roughly a quarter of calendar time to return an answer",
"Keeps branded-search and direct-traffic movement against the pre-launch baseline in the plan as the series that survives when the test does not",
"Labels that branded-search comparison as correlational rather than causal",
"Names concrete geo-lift or synthetic-control tooling (Meta's GeoLift package or Google's CausalImpact) rather than leaving the method unspecified",
"Does not hand the board an incrementality or causal claim the design cannot support",
"States that incrementality is realistic at always-on or aggregated portfolio level and never per creator",
"Given the hard date, prioritises the near-zero-setup rungs - unique promo codes, UTM-tagged links, and the branded-search proxy"
]
},
{
"id": 4,
"prompt": "Arcwell is a B2B observability platform, average deal around $38K ACV, sales cycle runs about seven months. We've just put four people on retainer indefinitely - two newsletter writers and two podcast hosts - at $12K a month total. Our VP Marketing wants a monthly report showing revenue from the program. We already have UTM links and a \"how did you hear about us\" dropdown on the demo form; right now about 12% of closed-won deals pick anything trackable in that dropdown. Build the measurement plan.",
"expected_output": "A B2B always-on measurement plan whose revenue object is influenced pipeline rather than revenue, that promotes the self-reported channel field to primary because trackable attribution is below the threshold, limits the stack to what newsletter and podcast placements actually allow, and declines incrementality on power grounds.",
"files": [],
"expectations": [
"Replaces revenue with influenced pipeline or opportunities created as the primary revenue object, because purchase rarely falls inside the flight window in B2B",
"States that a roughly seven-month cycle makes per-flight conversion KPIs undercount, and either stretches the attribution window or shifts to pipeline-stage metrics",
"Promotes the \"how did you hear about us\" field to the primary conversion source because fewer than about 20% of closed-won deals name a trackable channel",
"States that the self-reported channel field is analyzed at cohort level and never per deal",
"Uses trailing branded-search volume as the corroborating proxy alongside the self-reported field",
"Notes that newsletter and podcast placements give the brand no native analytics access",
"Limits the attribution options for those placements to unique codes and per-creator vanity URLs",
"Asks whether the publishers' download numbers are certified against the IAB Tech Lab Podcast Measurement Technical Guidelines v2.2",
"States that podcast downloads prove delivery, not response",
"Declines an incrementality test because B2B conversion volume at this size cannot power one, and writes that into the plan",
"Flags publisher-reported figures as publisher-sourced wherever they appear in reporting",
"Sets reporting cadence per stakeholder, with each row naming the decision that cadence drives",
"Declares dark social as a blind spot, including that the person filling the demo form may not be the person the creator influenced"
]
},
{
"id": 5,
"prompt": "Tavolo is a DTC cookware brand and we're finally doing creator marketing properly this year - around 20 creators across three flights. My plan was to have engineering build the TikTok and Meta pixel and shopping integration first, since that's the most accurate source we'd have, then layer everything else on top; eng quotes about six weeks including the consent work. Context: we already run an affiliate network for our coupon-site partners, and checkout is a Shopify build we can change freely. This is meant to be a capability we reuse every flight, not a one-off. What order should we build the measurement stack in, and write me the plan.",
"expected_output": "A measurement plan that explicitly ranks the attribution stack by coverage and confidence bought per hour of setup, puts pixel last rather than first, treats codes plus UTM links as the floor, promotes the already-live affiliate network and the post-purchase survey, and names which user answer moved each rung.",
"files": [],
"expectations": [
"Ranks the attribution stack explicitly by coverage and confidence bought per hour of setup, not by each method's apparent rigour",
"Places platform pixel and shopping integration last in the build order rather than first",
"Gives the reason pixel is last: engineering plus consent work, for a lossy signal the platform grades itself",
"Places unique promo codes and UTM-tagged links first, as the floor of the stack",
"States that codes and tagged links are kept together because they cover different slices - a code survives being screenshotted or read aloud, a tagged link does not",
"Promotes the already-running affiliate network to the top rung at near-zero incremental effort, and names that existing platform as the answer that moved it",
"Recommends the one-question post-purchase survey and justifies it by the reuse-every-flight mandate, noting it is the only method that sees dark social",
"States that a single-method stack is not triangulated",
"Writes each method's known failure mode next to it in the plan",
"Includes a code-leakage policy and a leakage monitor alongside the promo-code rung",
"Notes that pixel-based counts are a floor rather than a total, because of mobile tracking opt-outs and blocked third-party cookies"
]
},
{
"id": 6,
"prompt": "Kestrel Supply runs an affiliate-linked creator program - 30 creators, each with their own discount code. Our links are split across two networks because we migrated halfway through the year and the old contracts don't expire until March. Last month redemptions on four of the codes jumped about 40% but the click numbers on those creators didn't move at all. Finance also wants me to reconcile what each creator is owed for Q1 in the same document. Draft the measurement plan.",
"expected_output": "A measurement plan that reads the redemption-without-clicks pattern as code leakage rather than performance, installs the leakage guardrail and remediation ladder, records the two networks' differing cookie windows without summing across them, and refuses to do payout reconciliation in this document.",
"files": [],
"expectations": [
"Identifies redemptions rising with no matching link clicks as promo-code leakage rather than genuine creator performance",
"Names coupon or cashback browser extensions inserting themselves as the last click as the mechanism behind it",
"Makes code-leakage rate a mandatory guardrail KPI for this affiliate-linked objective rather than an optional extra",
"Defines the leakage monitor concretely as redemptions with zero matching link clicks",
"Recommends immediate deactivation of leaked codes",
"Recommends switching to single-use codes once leakage passes a few percent of redemptions",
"Adds a creator-onboarding policy restricting code distribution to the creator's own channels rather than coupon sites",
"States that the two affiliate networks' cookie windows differ and records each window in the glossary",
"Refuses to sum or compare conversions across the two networks as if they were on the same basis",
"Declines to compute what each creator is owed in this document and states that payout and commission reconciliation is a separate exercise",
"States that unmonitored leakage inflates the creator's credit and misallocates spend",
"Fixes one attribution window for the campaign and labels it in the glossary"
]
},
{
"id": 7,
"prompt": "Quick one. Bramble Home does about 12 creator collabs a quarter across TikTok, Instagram and YouTube. Current process: each creator DMs us screenshots of their insights a week after posting, we add the view counts up into one \"total views\" figure for the quarterly deck, and we compare that against the same quarter last year to show growth. Our new CMO says our numbers look soft next to what she saw at her last company. I want to tighten this up before Q3 kicks off - what should the measurement plan say?",
"expected_output": "A measurement plan that replaces screenshot sourcing with contracted analytics access, refuses the cross-platform view total, splits the year-over-year series at the platform view-definition change, date-stamps definitions in the glossary, and rejects the CMO's previous-company numbers as a target.",
"files": [],
"expectations": [
"Flags creator-supplied screenshots as trivially falsifiable and a documented fraud vector",
"Requires API or account-level analytics access to be written into the creator agreement rather than requested ad hoc",
"States that any figure that remains screenshot-sourced must be labelled as such in every report that carries it",
"Refuses to sum view counts across platforms into a single \"total views\" number",
"Gives the reason: each platform defines a view differently, so the totals are not additive",
"States that platforms redefined or relabelled views during 2025-2026 and that the year-over-year comparison crosses a definition change",
"Requires the reporting series to be split at the definition-change date and annotated there",
"Records each platform's view definition in the glossary with a date stamp",
"Rejects the CMO's numbers from a previous company as a target and sets targets from Bramble Home's own trailing median by platform instead",
"Distinguishes impressions from reach, noting impressions count repeat displays and are always greater than or equal to reach",
"Does not equate a platform view with an MRC-viewable impression",
"Delivers the fix as a measurement plan starting with the glossary, not as an ad-hoc correction to the existing deck"
]
},
{
"id": 8,
"prompt": "Sable & Finch, apparel brand. We have a dashboard wired to our creator data that refreshes hourly and everyone in the company has the link - me, the two founders, the paid social lead, the merch team, finance. Nobody looks at it. Our head of brand now wants me to add sentiment and share of voice to it as well. We're about to start a 10-creator autumn flight and I'd like to get the reporting right this time. What should the measurement plan say about reporting?",
"expected_output": "A measurement plan whose cadence section is a per-stakeholder table tying each audience to a frequency, a metric set and a named decision, that identifies the everyone-sees-everything live dashboard as the defect, and that declines to add metrics no decision depends on.",
"files": [],
"expectations": [
"Builds the cadence as rows, one per stakeholder, each naming who, how often, which metrics they see, and the decision it drives",
"States that a report driving no decision gets cut from the plan",
"Identifies the always-on dashboard shared with everyone and owned by no decision as the defect, rather than a shortage of metrics",
"Declines to add sentiment and share of voice unless a named stakeholder decision depends on them",
"Asks which decision each named stakeholder makes from the report before assigning them metrics",
"Demotes vanity counts such as reach, likes and follower counts from headline numbers to context",
"Ties the primary KPI both to the campaign objective's funnel stage and to the decision the report exists to drive",
"Points out that an hourly refresh serves no stated decision cadence",
"Produces a measurement plan for the autumn flight rather than dashboard advice alone",
"Declares the blind spots the reporting structurally cannot see"
]
},
{
"id": 9,
"prompt": "Meridian Pet Co. Our autumn creator flight is $92K all-in: $61K in creator fees, $14K for a shared studio production day that all nine creators used, $9K agency retainer, and $8K in whitelisted paid amplification that we only ran behind four of the nine. I want to report CPA per creator weekly and pause anyone above $45 CPA at the two-week mark. Write the measurement section.",
"expected_output": "A measurement section that makes CPA per creator recomputable by writing the cost-allocation rule into the glossary first, assigns creator-specific costs directly, decides explicitly how shared costs are handled, and blocks the pause decision until the rule and the attribution window are settled.",
"files": [],
"expectations": [
"Requires a written cost-allocation rule in the glossary before CPA per creator is reported at all",
"Assigns creator-specific costs directly to the creators who incurred them, including the $8K amplification that ran behind only four of the nine",
"Decides explicitly whether the $14K shared production day and $9K agency retainer are excluded from the per-creator numerator or allocated pro-rata by fee share, and states which was chosen",
"States that per-creator CPAs will not sum to flight-level CPA, and that this is noted on every report carrying both numbers",
"States that without the allocation rule the metric is not recomputable and must not drive a pause, budget-shift, or rebook decision",
"Flags the conflict between a pause decision taken at two weeks and the attribution window the plan fixes",
"Fixes and names a single attribution window for the campaign",
"Defines CPA with its numerator, denominator and window",
"Does not import an external or industry CPA benchmark to validate or replace the $45 threshold",
"Delivers the section in measurement-plan shape, naming the data source behind each KPI"
]
}
],
"trigger_queries": [
{
"query": "We're paying 8 TikTok micro-influencers for a skincare launch next month - how do we measure it?",
"should_trigger": true
},
{
"query": "Build the measurement plan for our creator campaign before it launches",
"should_trigger": true
},
{ "query": "What KPIs should we use for an influencer campaign?", "should_trigger": true },
{
"query": "Design the KPI and attribution plan for our B2B ambassador program with 4 newsletter writers",
"should_trigger": true
},
{ "query": "How do we attribute sales to individual creators?", "should_trigger": true },
{
"query": "Our agency wants to report EMV as our return. Is that legitimate?",
"should_trigger": true
},
{
"query": "We have promo codes and Google Analytics, nothing else. What can we honestly measure on this creator campaign?",
"should_trigger": true
},
{
"query": "Finance wants proof the creator line item works before we renew it",
"should_trigger": true
},
{
"query": "Set up the reporting for our influencer program before the flight starts",
"should_trigger": true
},
{
"query": "Which engagement rate formula should we standardise on across creators?",
"should_trigger": true
},
{ "query": "Can we run an incrementality test on our creator spend?", "should_trigger": true },
{
"query": "How do I know if my creator campaign actually caused anything?",
"should_trigger": true
},
{
"query": "What should we capture before we launch so we can prove lift afterwards?",
"should_trigger": true
},
{
"query": "Write the measurement framework for a YouTube sponsorship flight",
"should_trigger": true
},
{
"query": "We sponsor 3 newsletters and 2 podcasts. How do we track whether any of it works?",
"should_trigger": true
},
{
"query": "Our creators send screenshots of their analytics. Is that good enough for reporting?",
"should_trigger": true
},
{ "query": "Define what counts as a view for our creator reporting", "should_trigger": true },
{
"query": "Draft the pre-launch measurement plan for our autumn creator flight",
"should_trigger": true
},
{
"query": "How often should I report on our influencer program, and to whom?",
"should_trigger": true
},
{
"query": "My CMO wants one total views number across TikTok, Instagram and YouTube. Thoughts?",
"should_trigger": true
},
{
"query": "We need to decide which creators to rebook. What do we need to be measuring now?",
"should_trigger": true
},
{
"query": "Marketing wants an awareness push with 3 YouTube creators and finance wants proof it works",
"should_trigger": true
},
{
"query": "What's a realistic engagement rate target for our first creator campaign?",
"should_trigger": true
},
{
"query": "Can you help me structure the reporting deck for a creator flight, before we start it?",
"should_trigger": true
},
{ "query": "How do we measure a podcast sponsorship?", "should_trigger": true },
{
"query": "Our influencer promo codes are getting scraped by coupon extensions. How should the reporting handle that?",
"should_trigger": true
},
{
"query": "Do we have enough volume to run a geo holdout on our creator channel?",
"should_trigger": true
},
{
"query": "How should we define CPA per creator when a lot of the cost is shared across the whole flight?",
"should_trigger": true
},
{
"query": "B2B creator program, seven-month sales cycle. What do we actually report monthly?",
"should_trigger": true
},
{
"query": "Our attribution shows creators driving almost nothing but prospects keep telling sales they heard about us on a podcast. How do we reconcile that?",
"should_trigger": true
},
{
"query": "What should go in the glossary for our creator campaign reporting?",
"should_trigger": true
},
{
"query": "Is the $5.78 per dollar figure a reasonable planning assumption for influencer marketing?",
"should_trigger": true
},
{
"query": "We never captured a baseline before launching. What can we still claim?",
"should_trigger": true
},
{ "query": "Set up brand lift measurement for our creator campaign", "should_trigger": true },
{
"query": "What are the blind spots we should be declaring in our creator reporting?",
"should_trigger": true
},
{
"query": "How do I stop our influencer reports from being argued about after every flight?",
"should_trigger": true
},
{
"query": "Put together the measurement approach for a 20-creator always-on program",
"should_trigger": true
},
{
"query": "Which tracking methods should we build first for creator marketing, given limited engineering time?",
"should_trigger": true
},
{
"query": "Our board asked whether the platform's own lift study is trustworthy. What do we do?",
"should_trigger": true
},
{
"query": "Plan how we'll evaluate our creator partnerships at the end of the quarter, starting now",
"should_trigger": true
},
{
"query": "We're about to sign 6 creators. What tracking needs to be in place first?",
"should_trigger": true
},
{
"query": "How do we compare creator performance when each one is on a different platform?",
"should_trigger": true
},
{
"query": "What's the right way to measure an always-on ambassador program?",
"should_trigger": true
},
{ "query": "Should we use engagement rate by followers or by reach?", "should_trigger": true },
{
"query": "We migrated affiliate networks mid-year and the creator numbers don't line up. How should reporting handle it?",
"should_trigger": true
},
{
"query": "I need a document that fixes how we'll judge this creator campaign before anyone spends money",
"should_trigger": true
},
{
"query": "Design reporting for our creator program that our CFO will actually accept",
"should_trigger": true
},
{
"query": "What's the honest version of influencer ROI we can put in front of the board?",
"should_trigger": true
},
{
"query": "Our influencer reporting is all reach and likes. What should it be instead?",
"should_trigger": true
},
{ "query": "Write the creative brief for the 6 creators we just signed", "should_trigger": false },
{
"query": "What deliverable specs should I put in our creator brief - video length, aspect ratio, hook timing?",
"should_trigger": false
},
{
"query": "Draft the key messages and brand guardrails for an influencer campaign brief",
"should_trigger": false
},
{
"query": "How many revision rounds should I allow a creator in the brief?",
"should_trigger": false
},
{ "query": "Help me shortlist creators for a fitness apparel launch", "should_trigger": false },
{
"query": "What criteria should I use to vet influencers for brand safety?",
"should_trigger": false
},
{
"query": "Build a scoring template for ranking candidate creators against our audience fit",
"should_trigger": false
},
{
"query": "Find me B2B creators in the observability space worth approaching",
"should_trigger": false
},
{
"query": "Write the first DM I send to a creator we want to work with",
"should_trigger": false
},
{
"query": "Draft a follow-up sequence for creators who haven't replied to our pitch",
"should_trigger": false
},
{
"query": "How much should I pay a creator with 80k followers on Instagram?",
"should_trigger": false
},
{ "query": "What usage rights should I ask for in a creator contract?", "should_trigger": false },
{
"query": "The creator came back asking for category exclusivity. What's my counter?",
"should_trigger": false
},
{
"query": "Set my walk-away threshold for a podcast sponsorship negotiation",
"should_trigger": false
},
{ "query": "Spec the dashboard for our ongoing affiliate program", "should_trigger": false },
{
"query": "What metrics belong on an affiliate program dashboard for the exec team?",
"should_trigger": false
},
{
"query": "Define the alert thresholds for our affiliate program dashboard",
"should_trigger": false
},
{
"query": "Audit our Q1 affiliate commission payout run before we disburse it",
"should_trigger": false
},
{
"query": "Recompute what each affiliate is owed this month, including clawbacks",
"should_trigger": false
},
{ "query": "What commission rate should we pay our affiliates?", "should_trigger": false },
{
"query": "Should our affiliate program be revenue share or flat CPA?",
"should_trigger": false
},
{ "query": "Does this creator's caption meet FTC disclosure rules?", "should_trigger": false },
{
"query": "Review our influencer contracts against UK CAP disclosure requirements",
"should_trigger": false
},
{
"query": "Build the rules that flag cookie stuffing in our affiliate program",
"should_trigger": false
},
{
"query": "How do I tell real affiliate fraud from a low-incrementality cashback partner?",
"should_trigger": false
},
{
"query": "Design the welcome email series for newly approved affiliates",
"should_trigger": false
},
{
"query": "Write outreach copy to recruit new affiliates into our program",
"should_trigger": false
},
{
"query": "Draft the terms and conditions affiliates accept when they join",
"should_trigger": false
},
{ "query": "Build a per-partner scorecard for our reseller program", "should_trigger": false },
{ "query": "Design the QBR structure for our channel partners", "should_trigger": false },
{
"query": "Model the unit economics of our biggest reseller relationship",
"should_trigger": false
},
{ "query": "Plan a joint webinar campaign with our integration partner", "should_trigger": false },
{ "query": "How much should we pay out for a refer-a-friend referral?", "should_trigger": false },
{
"query": "Set velocity limits so customers can't game our referral program",
"should_trigger": false
},
{
"query": "Rank these five candidate technology alliances by expected value",
"should_trigger": false
},
{
"query": "Write our deal registration policy for partner-sourced deals",
"should_trigger": false
},
{ "query": "How many tiers should our partner program have?", "should_trigger": false },
{
"query": "Should we list on AWS Marketplace or Azure Marketplace first?",
"should_trigger": false
},
{
"query": "Build the measurement plan for our Meta and Google paid search campaigns",
"should_trigger": false
},
{
"query": "Set up GA4 conversion tracking and goals for our website",
"should_trigger": false
},
{ "query": "Design a marketing mix model for our whole media budget", "should_trigger": false },
{
"query": "How do I measure our own company newsletter's open and click rates?",
"should_trigger": false
},
{
"query": "What KPIs should we track for our employee advocacy program on LinkedIn?",
"should_trigger": false
},
{
"query": "Write a social media content calendar for our brand's own TikTok account",
"should_trigger": false
},
{ "query": "Measure the ROI of our booth at next month's trade show", "should_trigger": false },
{
"query": "Build the attribution model for our entire demand gen funnel",
"should_trigger": false
},
{
"query": "Which metrics should we hold our PR agency to for earned media coverage?",
"should_trigger": false
},
{ "query": "Set up a customer satisfaction survey after purchase", "should_trigger": false },
{
"query": "Track our SEO keyword rankings and organic traffic month over month",
"should_trigger": false
}
]
}
references/attribution-and-data-sources.md›
# Attribution methods, data sources, and their failure modes
Every method below fails somewhere. The plan's attribution section lists which methods are in use and, next to each, its known failure - so the readout's gaps are declared before launch, not discovered after.
## Method-by-method
| Method | How it works | Where it breaks |
| ------------------------------------- | ------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Unique promo codes | Creator-specific code at checkout | Leaks to coupon/cashback extensions; over-credits the creator; misses buyers who skip codes |
| UTM-tagged links | Query params read in web analytics | Stripped by shorteners and in-app link handling; lost in dark-social sharing; last-click only |
| Affiliate network links | Cookie or server-to-server tracking, commission on conversion | Cookie windows vary wildly between networks (24 hours to 30+ days), so the same sale is credited differently per network; extensions overwrite the last click |
| Platform pixel / shopping | On-platform conversion events | Mobile privacy opt-outs cause heavy signal loss; walled-garden numbers; the platform grades its own homework |
| Post-purchase / post-signup survey | One "how did you hear about us?" question at conversion | Self-report and recency bias; buyers credit the last-remembered touch |
| Vanity URLs / dedicated landing pages | Per-creator page or redirect | Only captures typed/clicked entries; still last-touch |
| Geo holdout / matched market | Channel on/off by region, compare | Needs scale; contaminated by overlapping campaigns |
**The triangulation rule [practice - near-universal]:** last-click structurally under-credits upper-funnel creator content. Deploy standardized UTMs + unique per-creator codes + a one-question survey together, and cross-check transactions with high code usage but zero link clicks. Report click-based and survey-based attribution side by side; the gap between them is information, not an error to hide.
## Code leakage [verified events]
- Coupon/cashback browser extensions insert themselves as the last click and harvest creator codes. A December 2024 exposé of this practice (13M+ views within days) triggered class-action litigation over "stolen influencer commissions" against several major extensions.
- Plan-level controls:
- Creator onboarding policy ("personal channels only, no coupon sites").
- A leakage monitor (redemption with no matching link click).
- Immediate deactivation of leaked codes.
- Single-use codes once leakage passes a few percent of redemptions.
## Signal loss context
- Mobile app-tracking opt-out rates remain high enough that pixel-based numbers materially undercount; third-party cookies are already blocked by default in several major browsers. Treat pixel/cookie counts as a floor, not a total.
- Cookie-window asymmetry: some affiliate programs credit within 24 hours, others 30 days last-click. When a campaign spans networks, the glossary states each window and the plan never sums across them as if comparable.
## Platform data access - what the brand can actually get
| Platform class | Brand-side access | Screenshot risk |
| ------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------ |
| Image/short-video social | Partnership/branded-content tooling gives ad-manager metrics when the creator opts in; organic story metrics expire fast and live creator-side | High without contracted access |
| Long-form video | Public view counts; watch time and retention only via the creator's analytics | Medium |
| Professional network | Impression definitions include non-follower feed displays; weak marketplace tooling | Medium |
| Live streaming | Platform APIs exist; attribution to commerce is weakest here | High |
| Podcasts / newsletters | No native brand access at all - measure via codes, vanity URLs, and publisher-reported numbers | Highest |
- **Contract the access, don't hope for it:** API or account-level analytics connection belongs in the creator agreement. Screenshots are trivially falsified and a documented fraud vector; any screenshot-sourced figure gets flagged as such in every report.
- **Podcast/newsletter sponsorships [standard - IAB]:** the IAB Tech Lab Podcast Measurement Technical Guidelines v2.2 (May 2, 2024) define server-side, filtered download counting - the only certified standard in this space. Ask publishers whether their numbers are IAB v2.2-certified; pair downloads with unique codes and vanity URLs since downloads prove delivery, not response.
## The view-definition problem [platform - verified changes]
- Major platforms redefined or relabeled "views" during 2025-2026, moving toward counting from the first frame. Any comparison crossing a definition-change date is invalid; split the reporting series at the date and annotate it.
- Consequence for the plan: the glossary records each platform's view definition _as of the plan date_, and cross-platform view totals are never summed into one "views" number.
## B2B: dark social and self-reported attribution
- B2B creator influence travels through DMs, private communities, podcasts, and screenshots - channels that strip every tracking parameter. Click-based attribution then shows "direct" and "branded search" doing the work that creators actually did.
- Primary workaround [practice]: a "How did you hear about us?" (HDYHAU) field on demo/signup forms - dropdown of known channels plus free text - analyzed at cohort level, never per-deal. Corroborate with trailing branded-search volume.
- Provenance caution: the widely quoted dark-social revenue percentages come from individual practitioners' own CRMs and vendor reports [vendor/practice], not audited studies. Use the mechanism (HDYHAU + branded-search proxy), not the quoted percentages, in the plan.
- Decision threshold [practice]: if fewer than ~20% of closed-won deals name a trackable channel, dark social dominates - promote HDYHAU to the primary conversion source and say so in the plan.
references/incrementality-and-brand-lift.md›
# Incrementality testing and brand lift - when they belong in the plan
Incrementality is the only measurement class that answers "did this creator activity _cause_ anything," and most creator campaigns cannot power it. The plan's job is to decide honestly which side of that line the campaign falls on.
## Geo-lift / synthetic control [verified methods]
- Method: withhold (or concentrate) the creator activity in test regions; compare outcomes against a weighted synthetic control built from untreated regions. Privacy-durable - needs no user-level identifiers.
- Open tooling exists: Meta's open-source **GeoLift** package (R; augmented synthetic control) and Google's **CausalImpact** (Bayesian structural time series). Both are free and harness-independent - if you can run R or Python, either works; otherwise hand the design to whoever owns analytics.
- Alternatives when platforms offer them: conversion-lift holdouts and ghost-ad designs - precise, but they require platform ad spend and inherit the self-measurement conflict below.
## The MDE power check - run it before promising a test
1. Compute the Minimum Detectable Effect (MDE) for the available budget, conversion volume, and window (GeoLift ships power-analysis functions; CausalImpact requires a manual pre-period fit).
2. Practitioner target [practice]: a 5-10% MDE over a 4-8 week window.
3. Decision rule: if the computed MDE exceeds the lift you would actually act on (e.g., the test can only detect a 25% lift but a 10% lift would already justify the spend), **do not run the test** - write "directional only" into the plan.
4. Why this is non-negotiable: an underpowered test returns "no significant lift" for a channel that genuinely works, and someone cuts a working channel on the strength of a test that could never have seen the effect.
**Program-size reality [practice]:** single-creator and micro-influencer budgets almost never generate enough conversions to power a test. Incrementality is realistic for always-on programs or creator spend aggregated at portfolio level - never per-creator. B2B conversion volumes rarely power it at all.
## Brand lift studies
- Design: randomized exposed-vs-control groups assigned pre-campaign, surveyed natively on recall, awareness, message association, favorability, consideration, intent.
- Entry costs are statistical-power thresholds, not price lists [vendor/platform - directional]: platform-run studies start around low five figures and run to six figures depending on platform; independent panels cost more but see across platforms. Minimums change - verify current thresholds with the provider (if you can browse the web, check; otherwise mark the figure "unverified, ask provider").
- Duration: typically 1-4 weeks of in-flight surveying; short flights return "inconclusive".
- **The self-measurement conflict [practice - widely voiced]:** platform-run lift studies are the platform "marking their own homework". Note the conflict in the plan; corroborate with an independent panel when budget allows - while remembering the panels selling that argument are also commercially interested.
- Cheap proxy when no study is powered or funded: pre/post movement in trailing branded-search volume and direct traffic against the pre-launch baseline. Correlational, honest about it, and free - which is why the baseline capture step is mandatory for awareness campaigns.
## Decision thresholds that change the plan [practice]
- MDE comes back above ~15-25%: drop incrementality; rely on triangulation + survey; plan says "directional only".
- Code-leakage rate (redemptions with zero matching link clicks) exceeds a few percent: switch to single-use codes.
- Survey self-report for the creator channel exceeds 20-30% while click-based attribution gives it under 5%: the model is wrong, not the survey - reweight toward the survey signal and note it in the readout.
- B2B with under ~20% of closed-won deals naming a trackable channel: promote HDYHAU to primary; branded-search volume becomes the corroborating KPI.
references/kpi-definitions.md›
# KPI definitions, KPI sets per objective, and target-setting rules
Provenance labels used throughout:
- **[standard]**: standards body (ANA/MRC/AMEC/IAB).
- **[platform]**: platform-published.
- **[vendor]**: commercially interested source.
- **[practice]**: practitioner consensus without a governing standard.
- **[internal]**: the brand's own history.
## Exposure metrics [standard - ANA definitions]
- Impressions: total times content was displayed, counting repeat displays to the same person.
- Reach: unique individuals who saw the content at least once. One person seeing a post twice = 2 impressions, 1 reach. Impressions >= reach, always.
- Viewability caveat [standard - MRC]: the MRC bar is 50% of pixels on screen for 1 continuous second (display) or 2 continuous seconds (video). Most organic platform metrics do not meet it - per the ANA guidelines' own footnote. Never equate a platform "view" with an MRC-viewable impression.
## Engagement rate - the denominator problem [practice]
Four formulas circulate; the same post can read 9% on one and 1.2% on another. The plan picks exactly one, states it in the glossary, and holds it constant.
| Formula | Definition | Use when | Distortion |
| ----------------- | ------------------------------- | ---------------------------------------------- | ------------------------------------------------------------- |
| ER by reach (ERR) | engagements / reach x 100 | Judging individual organic posts (most common) | Reach only visible with creator data access |
| ER by impressions | engagements / impressions x 100 | Paid or high-frequency bursts | Dilutes with repeat exposure |
| ER by followers | engagements / followers x 100 | Almost never - legacy habit | Flatters small accounts; meaningless on algorithm-first feeds |
| ER by views | engagements / video views x 100 | Video-first platforms | Inherits each platform's view definition |
## Cost and efficiency metrics [practice; formulas uncontested]
- CPM = (cost / impressions) x 1000 - awareness efficiency.
- CPV = cost / video views - no reliable cross-campaign benchmark, since organic performance drives the denominator.
- CPE = cost / total engagements - consideration efficiency.
- CPA = total campaign cost / attributed conversions. ANA's variant [standard]: program cost / total conversions.
- **CPA per creator** = that creator's allocated cost / conversions attributed to that creator. Recomputable only with a written cost-allocation rule, since the flight-level numerator does not divide by itself:
- Assign creator-specific costs (fee, usage rights, gifted product, creator-specific paid amplification) directly.
- Then either exclude shared costs (platform fees, agency retainer, shared production) or allocate them pro-rata by fee share. State which in the glossary before the first report.
- Without that line, the metric is not recomputable and must not drive a pause, budget-shift, or rebook decision.
- ROAS = attributed revenue / campaign spend.
- ROI = (attributed revenue - cost) / cost x 100. Only defensible when "attributed" is defined in the glossary.
## KPI sets by objective [practice - synthesized consensus, no governing standard]
| Objective | Primary KPIs | Secondary / guardrail | Notes |
| -------------------------- | --------------------------------------------------------------------------- | ------------------------------------------- | ------------------------------------------------------------ |
| Awareness / reach | Reach, impressions, CPM, video views; brand lift if powered | Frequency, view-completion rate, sentiment | Baseline mandatory: trailing branded search + direct traffic |
| Consideration / engagement | ER (chosen denominator), CPE, saves, shares, comments, link CTR | Watch time, completion rate, profile visits | Saves/shares signal intent better than likes |
| Conversion / performance | CPA, ROAS, conversion rate, AOV | CTR, new-customer rate | Requires a working attribution stack, not one method |
| Affiliate-linked | Attributed revenue, EPC, conversion rate | Code-leakage rate, return rate | Leakage rate is a mandatory guardrail, not optional |
| Always-on ambassador | Incremental revenue (if powered), LTV of referred customers, content volume | Post cadence, retention of creators | Only objective where incrementality is usually feasible |
B2B substitution: for the conversion and always-on rows, replace orders/revenue with influenced pipeline, opportunities created, and cohort-level self-reported ("how did you hear about us") channel share - purchase rarely falls inside the flight window, and dark social strips the tracking that would otherwise credit the creator.
## EMV, MIV, AVE - why they never appear as financial value
- EMV (Tribe Dynamics/CreatorIQ) and MIV (Launchmetrics) are proprietary black boxes [vendor]; MIV's own description bases it on "advertising value equivalent" - confirming both are AVE derivatives.
- Barcelona Principle 5 [standard - AMEC]: "AVEs are not the value of communication" (3.0, July 2020; reaffirmed 4.0, June 2025: invalid measures such as AVEs should not be used). The IPR rejected AVE unanimously in 2010. AMEC's "22 reasons to say no to AVEs" (2017) notes there is no agreed methodology and no peer-reviewed multiplier.
- Legitimate use: relative, directional comparison inside one vendor's consistent methodology (creator A vs creator B, this quarter vs last). Never converted to dollars for finance, never called ROI.
## Benchmarks and target setting
- The "$5.78 per $1" influencer ROI figure [vendor - Influencer Marketing Hub] has inconsistent provenance (variously cited as $5.20, misattributed to other sources) and mixes industries, platforms, and self-reported methodologies. Never use it, or any variant, as a planning input.
- Vendor benchmark reports carry survivorship bias (their own customers' successful campaigns) and silently pick flattering ER denominators. Treat all of them as calibration context, provenance-tagged, never as targets.
- Defensible targets are internal [internal]: the brand's own trailing median by platform, objective, and creator tier. First campaign on a platform = baseline-setting flight; the plan says "no target - establishes the baseline" instead of inventing one.
- "Good engagement is 1-5%" and similar ranges are heuristics [practice], not standards; no credible universal benchmark exists for ER, ROI, or CPA in influencer marketing.
references/worked-examples.md›
# Worked examples
Two positive plans (one B2C conversion, one B2B always-on) and one annotated negative example. Fictional brands; every number illustrative, not a benchmark.
## Table of Contents
- [Example 1 - B2C conversion campaign](#example-1---b2c-conversion-campaign)
- [Example 2 - B2B always-on ambassador program](#example-2---b2b-always-on-ambassador-program)
- [Negative example - annotated](#negative-example---annotated)
## Example 1 - B2C conversion campaign
Context: skincare DTC brand, launch flight, 8 micro-creators on short-video platforms, $40K, 6-week flight. Tracking confirmed in interview: unique codes, UTM links, post-purchase survey, platform pixel. Report read by the growth lead; decision: which creators to rebook.
```
MEASUREMENT PLAN - Lumen Skin spring launch, 2026-03-02
Header : Lumen Skin | conversion objective (performance funnel stage) | Mar 16 - Apr 26 |
$40K | 2 short-video platforms | 8 micro-creators
Glossary : CPA = total flight cost / attributed conversions (dedupe: code OR link OR survey,
priority code > link > survey), 30-day window, flight timezone UTC-5.
CPA per creator = that creator's fee + their creator-specific amplification,
divided by conversions attributed to them. Shared costs (platform fees, the
$4K shared production day) are EXCLUDED from the per-creator numerator and
reported only at flight level - so per-creator CPAs do not sum to flight CPA,
and that is stated on every report carrying both.
ER = engagements / reach x 100 (ERR) - reach via contracted creator analytics access.
Conversion = first paid order. View = per-platform definition captured at flight
start, date-stamped; views never summed cross-platform.
KPI tree : PRIMARY CPA per creator (target: none - first flight, baseline-setting)
SECONDARY conversion rate per creator, AOV of attributed orders, link CTR
GUARDRAIL code-leakage rate (<3% of redemptions), return rate, new-customer share
KPI mapping : CPA -> codes + UTMs + survey (triangulated) -> checkout system + analytics + survey tool
ER -> creator analytics API access (contracted) -> per-creator monthly export
Leakage -> redemptions with zero matching link click -> checkout x analytics join
Attribution : codes (leak monitor + single-use fallback) + UTMs (one per creator per placement) +
post-purchase survey ("How did you hear about us?" with creator names listed) + pixel
(floor only, signal loss noted). Click-based and survey-based CPA reported side by side.
Baseline : trailing 8-week branded-search volume and direct traffic captured 2026-03-09;
trailing ER median per creator captured from vetting data.
Test design : DIRECTIONAL ONLY. Power check: ~450 expected conversions across 8 creators cannot
detect a 5-10% MDE at region level. No incrementality claim will be made.
Cadence : growth lead -> weekly -> pacing, CPA by creator, leakage alerts -> pause/shift budget
founder -> end of flight -> CPA + survey-vs-click gap + rebook shortlist -> rebook decision
Data sources: checkout (codes, orders) | analytics (UTM sessions) | survey tool | creator API access |
pixel (flagged: floor). Zero screenshot sources.
Blind spots : dark-social shares strip UTMs; survey recency bias; pixel undercounts mobile opt-outs;
buyers who neither click nor use codes are invisible to everything except the survey.
Open items : confirm survey tool can list all 8 creator names; add analytics-access clause to the
two unsigned creator contracts.
```
Why it passes:
- Every KPI is recomputable, backed by a confirmed capability.
- First-flight target honesty.
- Underpowered test declared instead of promised.
- Leakage guardrail present.
- Blind spots stated.
## Example 2 - B2B always-on ambassador program
Context: devtools SaaS, 4 newsletter/podcast creators on retainer, always-on. Tracking confirmed: UTM links, vanity URLs, HDYHAU field on signup form, CRM. No pixel, no codes (no checkout). Report read by VP Marketing; decision: continue/expand the program at annual budget review.
```
MEASUREMENT PLAN - DevFlow ambassador program, 2026-01-12
Header : DevFlow | always-on ambassador (pipeline objective) | rolling quarters | 4 creators |
newsletters + podcasts
Glossary : Influenced pipeline = open opportunities whose contact cohort selected a creator/
newsletter/podcast HDYHAU option OR entered via creator UTM/vanity URL, 90-day window.
Signup = workspace created. HDYHAU analyzed at cohort level, never per-deal.
Podcast downloads = publisher-reported; IAB v2.2 certification asked, answer pending.
KPI tree : PRIMARY influenced pipeline per quarter (target: trailing 2-quarter median once it exists)
SECONDARY signups via vanity URLs, HDYHAU creator-channel share, branded-search trend
GUARDRAIL creator content cadence vs retainer terms, HDYHAU "other/blank" rate (<40%)
KPI mapping : pipeline -> HDYHAU cohort + UTM/vanity -> CRM + analytics
downloads -> publisher reports (flagged: publisher-sourced, certification unconfirmed)
Attribution : HDYHAU is PRIMARY (dark social dominates: <20% of current closed-won deals name a
trackable channel), UTM/vanity secondary, branded-search volume as corroborating proxy.
Baseline : 12 months of branded-search volume and signup HDYHAU distribution captured pre-program.
Test design : DIRECTIONAL ONLY - B2B conversion volume (~60 opportunities/quarter) cannot power
geo-lift; revisit if the program quintuples. Pre/post branded-search movement reported
against baseline with correlational framing.
Cadence : program manager -> monthly -> cadence, vanity signups, HDYHAU share -> creator check-ins
VP Marketing -> quarterly -> influenced pipeline vs median, branded-search trend ->
continue/expand/cut at annual review
Data sources: CRM (HDYHAU, pipeline) | analytics (UTM/vanity) | publisher reports (flagged) |
search console (branded queries). Zero screenshot sources; publisher numbers flagged.
Blind spots : per-deal creator credit is impossible by design (cohort-level only); podcast downloads
prove delivery, not response; multi-stakeholder buying committees mean the form-filler
may not be the person the creator influenced.
Open items : IAB v2.2 certification answer from both publishers; add "creator/newsletter/podcast -
which one?" follow-up option to the HDYHAU dropdown.
```
Why it passes:
- Revenue object is pipeline, not orders.
- HDYHAU promoted to primary with the threshold stated.
- Publisher-sourced numbers flagged.
- Incrementality honestly declined with the reason.
## Negative example - annotated
The same B2C brand as Example 1, as a plan that fails. Each `<- flaw` annotation names the defect.
```
MEASUREMENT PLAN - Lumen Skin spring launch
Goals : go viral, build buzz <- no funnel stage, no decision named
KPIs : total reach, impressions, likes, followers <- vanity set on a conversion campaign
EMV target: $250K earned media value <- AVE derivative presented as financial value
ER target: 4.5% (industry benchmark) <- vendor benchmark as target; denominator unstated
Tracking : last-click via the affiliate network <- single method; upper funnel structurally
under-credited; no leakage guardrail
creators send monthly screenshots <- falsifiable source, unflagged
ROI : projected 5.78x based on industry data <- the $5.78 vendor figure as a planning input
Testing : we will run a lift test to prove incrementality <- no power check; ~450 conversions cannot
detect the MDE; "no lift" would kill a
possibly-working channel
Baseline : (none) <- launch first, measure later; every "lift"
claim will be unsupported
Reporting: dashboard for everyone, updated live <- no stakeholder, no cadence, no decision
Views : combined views across platforms: 2M target <- sums incompatible view definitions;
YoY comparison crosses a definition change
```
Structural failures beyond the annotations:
- No glossary at all (nothing is recomputable).
- No blind-spot declaration.
- The primary KPI set (reach/likes) cannot inform the rebook decision the campaign exists to make.
SKILL.md›
---
name: influencer-measurement-framework
description: Design the pre-launch measurement plan for one influencer or creator campaign - KPIs matched to the objective (awareness, engagement, conversion, affiliate-linked, always-on ambassador), an attribution method and data source behind each KPI, reporting cadence per stakeholder, and declared blind spots. Delivered as a measurement plan document before launch, for B2B and B2C alike. Use whenever the user mentions influencer KPIs, creator campaign attribution, campaign reporting, or asks how to measure a creator partnership, even if they never say measurement. Do NOT use for an ongoing affiliate program dashboard - use mbfinotti/partnerships-skills@affiliate-performance-dashboard instead.
license: MIT
metadata:
author: Maya-Beth Finotti
version: "1.4.4"
---
# Influencer Measurement Framework
Design how one influencer/creator campaign will be measured - before it launches - and deliver a measurement plan document. The plan, not a dashboard or a report, is the artifact: it fixes KPIs, attribution methods, data sources, cadence, and blind spots so the post-campaign readout is decided in advance, not improvised after.
Influencer measurement has no comprehensive, benchmark-bearing standard behind it. Say this plainly to the user, and never present a branded framework as an industry standard.
The method borrows only genuinely established pieces:
- **ANA Influencer Marketing Measurement Guidelines** (June 2022, MRC-endorsed): standardized definitions only, without benchmark figures.
- **AMEC Barcelona Principles** (3.0 July 2020, reaffirmed 4.0 June 2025): Principle 5, "AVEs are not the value of communication".
- **MRC viewability standard**: 50% of pixels, 1s display / 2s video; most organic platform metrics fail it.
- **IAB Tech Lab Podcast Measurement Technical Guidelines v2.2** (May 2, 2024): for podcast/newsletter sponsorships.
- **Incrementality via geo-lift/synthetic control** (Meta's open-source GeoLift package, Google's CausalImpact): paired with a Minimum Detectable Effect power check.
## Ground Rules
- Match the primary KPI to the campaign objective's funnel stage. Optimizing an awareness campaign on conversion metrics, or the reverse, is the single most-cited practitioner error.
- Give every KPI a one-sentence definition naming numerator, denominator, and window. A metric nobody can recompute is decoration.
- Pick one engagement-rate denominator and one attribution window; hold both constant for the whole campaign and label them in the glossary.
- Never report EMV, MIV, or any AVE derivative as ROI or financial value - relative, same-vendor directional comparison is their only legitimate use.
- Never adopt a vendor benchmark as a target. Targets come from the program's own trailing history, or the plan honestly says "first flight - no target, baseline-setting run."
- Never use the "$5.78 per $1" figure (or its variants) as a planning input - it is a vendor benchmark with weak, inconsistent provenance.
- Triangulate attribution: no single method is reliable alone. Codes leak, links get stripped, pixels lose signal, surveys mis-remember - each covers another's blind spot.
- Provenance-tag every number the plan cites: standards-body definition, platform-published, vendor-sourced, single-practitioner claim, or the brand's own history.
## Interview
Ask before designing anything. One question per message; multiple-choice where possible; skip anything already answered.
- Campaign objective and funnel stage: awareness/reach, consideration/engagement, conversion/performance, affiliate-linked, or always-on ambassador?
- B2B or B2C? (Drives revenue object, window length, and which attribution signals exist at all.)
- Budget and program size: total spend, number of creators, one flight or always-on?
- Platforms in play? (Short video, image/social, long video, live, podcast, newsletter - each has different data access.)
- Deliverable types per creator? (Posts, stories, videos, streams, podcast reads, newsletter placements.)
- Which tracking capabilities actually exist today: unique promo codes, UTM-tagged links, affiliate network, platform pixel/shopping, post-purchase survey, CRM "how did you hear about us" field? (Multi-select; "none" is a valid answer.)
- Does a pre-launch baseline exist - trailing branded-search volume, direct traffic, follower and engagement history? If not, is there time to capture one before launch?
- Who reads the report, and what decision does it drive? (Renew the creator, reallocate budget, prove channel viability to finance, creative learnings.)
- What reporting cadence is expected, and by whom?
- What creator-side analytics access is contractual: API/account connection, platform partnership tooling, or screenshots only?
- By what date must the result land, and what decision is waiting on it? (A hard date promotes the near-zero-setup rungs - codes, UTMs, the branded-search proxy - and rules out geo-lift and panel brand-lift studies, which report after the decision is already made.)
- One-off readout for this flight, or a measurement asset reused every flight? (Compounding promotes the post-purchase survey, the affiliate network and a reusable geo design - each costs its setup once and pays every flight after. One-off keeps the stack at codes + UTMs + baseline proxy.)
- Effort ceiling: analyst hours available, whether checkout can be changed, whether ad spend can be committed to a holdout, and whether the channel can be withheld from whole regions?
- No checkout change demotes the survey and the pixel.
- No willingness to withhold deletes geo-lift and conversion-lift holdouts from the menu.
- No analyst on hand promotes platform-run test designs above self-run ones.
## Workflow
1. Run the Interview; collect every answer the plan depends on.
2. Fix the objective, then select the KPI set for that objective from [references/kpi-definitions.md](references/kpi-definitions.md) - primary, secondary, guardrail. Refuse a primary KPI from the wrong funnel stage; explain why instead of complying.
3. Write the metric glossary before anything else: one-sentence definition per KPI, the chosen engagement-rate denominator, the attribution window, and the view definition per platform. Platform "view" definitions changed materially in 2025-2026, so date-stamp them; if you can browse the web, verify current platform definitions - otherwise flag them as "as of plan date, re-verify".
4. Map each KPI to an attribution method and a data source the Interview confirmed exists, using [references/attribution-and-data-sources.md](references/attribution-and-data-sources.md). Cut or flag any KPI with no feasible source - a KPI backed by a capability the brand lacks is fiction.
5. Assemble the attribution stack in efficiency order: coverage and confidence bought per hour of setup, never per method's apparent rigour. Default order:
a. Unique promo codes and UTM-tagged links together: an hour each, per-creator resolution from day one. The pair is the floor; a single-method stack is not triangulated.
b. A one-question post-purchase/post-signup survey: a week to ship through whoever owns checkout, near-zero effort forever after, the only method that sees dark social.
c. Affiliate-network links: near-zero when a platform is already live, a standing job of contracts and onboarding when it is not.
d. Per-creator vanity URLs: narrow coverage, the only rung that exists on podcast and newsletter placements.
e. Platform pixel/shopping last: engineering plus consent work, for a lossy signal the platform grades itself.
```
effort: pixel > affiliate network > survey > vanity URLs > promo codes == UTM links
coverage: survey > affiliate network > promo codes == UTM links > pixel > vanity URLs
compliance cost: pixel > survey > affiliate network > promo codes == UTM links == vanity URLs
efficiency: promo codes == UTM links > survey > affiliate network > vanity URLs > pixel
```
Codes and links tie because they cover different slices: a code survives being screenshotted or read aloud; a tagged link does not. This order is a default, not a law; it shifts with the brand's existing stack and with who has to build each rung.
Re-rank against the Interview answers, and write in the plan which answer moved which rung:
- An affiliate platform already running jumps to rung (a) at near-zero incremental effort.
- A checkout survey already collecting responses is already rung (a).
- A store that cannot issue unique per-creator codes drops half of (a) and promotes the survey.
- A podcast or newsletter flight has no rungs available but codes and vanity URLs.
Note each method's known failure and the code-leakage policy beside it in the plan.
6. Capture or schedule the pre-launch baseline; if launch precedes capture, write "no baseline - post-campaign lift claims will be unsupported" into the plan rather than hiding it.
7. Decide the test design with [references/incrementality-and-brand-lift.md](references/incrementality-and-brand-lift.md). Run the MDE power check first: an hour of analyst time that gates every rung below it, and the one step here with no alternative. Then take the highest rung the power check _and_ the effort ceiling both allow:
a. Branded-search and direct-traffic movement against the pre-launch baseline: near-zero effort, correlational. It belongs in every plan including the powered ones, since it is the series that survives when the test does not.
b. Geo-lift/synthetic control: the strongest causal evidence available without user-level identifiers, priced in analytics skill, a quarter of calendar time, and the political capital to withhold the channel from real markets.
c. Platform conversion-lift holdout: the platform runs it, so less work than (b), but it needs ad spend on that platform and the platform grades its own homework.
d. Platform-run brand-lift study: attitudes rather than behaviour, a spend minimum, 1-4 weeks of in-flight surveying, same self-measurement conflict.
e. Independent-panel brand lift: cross-platform and conflict-free, and a standing job to commission.
Below power on every rung above (a), write "directional measurement only" into the plan.
```
effort: panel lift > geo-lift > platform brand lift > conversion-lift holdout > search proxy
time to an answer: panel lift == geo-lift > platform brand lift == conversion-lift holdout > search proxy
evidence strength: geo-lift == conversion-lift holdout > panel lift > platform brand lift > search proxy
compliance cost: panel lift > platform brand lift > conversion-lift holdout > geo-lift == search proxy
efficiency: search proxy > geo-lift > conversion-lift holdout > platform brand lift > panel lift
```
Default rung: (a) alone. Geo-lift and panel lift are what this order starves: the first buys the strongest causal evidence available, the second the only conflict-free one, and both lose every round to a proxy series that costs nothing. A plan run on the ratio alone never produces a causal number at all.
Move up a rung on these conditions:
- The power check passes _and_ the decision the report drives is a budget-size one: move up one rung, since proving the channel to finance justifies a quarter of work.
- The claim has to survive the platform that sold the media grading its own work: reach panel lift.
- The decision is a rebook or a creative learning: stay at (a).
This order is a default, not a law. Re-rank on the Interview answers:
- An awareness-primary campaign with no conversion volume skips (b) and (c) entirely, because brand lift is the only instrument that measures its objective at all.
- An in-house analytics team collapses (b)'s effort and promotes it above (c).
- A brand that will not withhold the channel from any region deletes (b) and (c) from the menu rather than demoting them.
8. Set reporting cadence per stakeholder - each row names who, how often, which metrics, and the decision it drives. A report driving no decision gets cut.
9. Declare blind spots explicitly: dark social, view-through, stripped UTMs, code leakage, screenshot-only sources, platform definition changes crossing the flight window.
10. Validate the plan with the user section by section - glossary, then KPI tree, then attribution, then test design, then cadence - before assembling the final document, grounded in a matching worked example from [references/worked-examples.md](references/worked-examples.md).
11. Check the Pass Threshold below; iterate until every criterion holds.
12. If your harness has persistent memory, memorize the agreed metric definitions, chosen denominators, attribution windows, and derived baselines - later campaigns then start from them instead of re-negotiating. Otherwise hand the glossary to the user as the standing reference for the next campaign.
## Output Shape
Deliver every plan as this artifact:
```
MEASUREMENT PLAN - <campaign>, <date>
Header : brand, campaign, objective + funnel stage, flight dates, budget, platforms, creators
Glossary : one-sentence definition per metric (numerator / denominator / window);
chosen ER denominator; attribution window; view definition per platform, date-stamped
KPI tree : primary | secondary | guardrail - per KPI: definition ref, target basis
(internal trailing baseline or "first flight - none"), attribution method, data source
Attribution : the stack in use, known limitation per method, code-leakage policy and monitor
Baseline : pre-launch values captured (branded search, direct traffic, trailing ER median)
or an explicit "no baseline" declaration
Test design : incrementality / brand lift with MDE + power check result, or "directional only" + why
Cadence : stakeholder -> frequency -> metrics shown -> decision it drives
Data sources: per KPI - API / ads manager / affiliate network / CRM / survey / screenshot (flagged)
Blind spots : declared list - what this plan structurally cannot see
Open items : unconfirmed capabilities, contract clauses to add, questions before launch
```
Full worked examples live in [references/worked-examples.md](references/worked-examples.md), never inline.
## Pass Threshold
The plan passes only when the stakeholder named in the Interview could, at campaign end, make their stated decision from it without arguing about definitions. Concretely, all six must hold; iterate until they do:
- Every KPI's definition is recomputable by someone who did not write the plan - numerator, denominator, window all stated.
- The primary KPI's funnel stage matches the campaign objective's funnel stage.
- Every KPI is backed by an attribution method and data source the Interview confirmed exists - zero KPIs resting on wished-for capabilities.
- Baseline status is explicit: captured, scheduled pre-launch, or declared absent with the consequence stated.
- Incrementality or brand lift appears only with a passing power check attached; otherwise "directional only" is written in the plan.
- Zero EMV/MIV/AVE-derived numbers presented as financial value, and zero vendor benchmarks used as targets.
## B2B vs B2C
Identical for both, apply without modification:
- Glossary discipline.
- KPI-to-objective matching.
- Triangulation.
- Baseline capture.
- Provenance tagging.
- The cadence-drives-a-decision rule.
What genuinely diverges:
- **Revenue object.** B2C measures orders, AOV, and revenue inside the flight window. B2B measures influenced pipeline and opportunities - purchase rarely happens during the campaign at all.
- **Cycle length and windows.** B2C attribution windows run days to ~30 days; B2B cycles run months, so per-flight conversion KPIs undercount and windows must stretch or shift to pipeline-stage metrics.
- **Dark social dominance.** B2B creator content spreads through DMs, communities, and podcasts that strip all tracking; self-reported "how did you hear about us" (HDYHAU) on demo/signup forms, analyzed at cohort level, becomes a primary source, not a tiebreaker - with branded-search volume as the corroborating proxy.
- **Audience size and statistical power.** B2B creator audiences are small; conversion volumes almost never power an incrementality test, so B2B plans default to triangulation + HDYHAU and say so.
- **Survey placement.** B2C asks post-purchase at checkout; B2B asks on the demo-request or signup form and again qualitatively on sales calls.
## Failure Modes
| Defect | Consequence | Fix |
| ----------------------------------------------------------------- | ---------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------- |
| Vanity-metric worship (reach, likes, follower counts as headline) | Report looks busy; decision-maker learns nothing | Primary KPI must match objective and drive the named decision; demote counts to context |
| EMV/MIV/AVE reported as ROI | Finance discovers the number is fictional; whole program loses credibility | Barcelona Principle 5; directional same-vendor comparison only, never financial value |
| Last-click as the only conversion lens | Upper-funnel creator content structurally under-credited; program looks worse than it is | Triangulate with survey + codes; report click-based and survey-based attribution side by side |
| No baseline captured before launch | "Lift" claims unsupported; awareness campaigns unmeasurable | Capture trailing branded search, direct traffic, ER median pre-launch - or declare the gap |
| Mixing engagement-rate denominators | Same post reads 9% or 1.2% depending on denominator; trends meaningless | One denominator, chosen in the glossary, labeled on every number |
| Comparing periods across a platform metric-definition change | False trend in either direction | Split reporting at the change date; date-stamp view definitions in the glossary |
| Underpowered incrementality test | "No significant lift" read as failure; a working channel gets cut | MDE power check before committing; below power, declare directional-only |
| Trusting creator screenshots | Trivially falsified numbers enter the report | Contract API/account-level access up front; flag screenshot-sourced data in every output |
| Vendor benchmark adopted as target | Targets miscalibrated; success or failure declared arbitrarily | Internal trailing median by platform/tier/objective; vendor figures are calibration context only |
| Promo-code leakage to coupon extensions | Extension harvests the code; creator's credit inflated, spend misallocated | Monitor redemptions with zero matching link clicks; switch to single-use codes past a few percent |
## Invocation Examples
- "We're paying 8 TikTok micro-influencers for a skincare launch next month - how do we measure it? Build the measurement plan."
- "Design the KPI and attribution plan for our B2B always-on ambassador program: 4 newsletter writers, goal is pipeline."
- "Marketing wants an awareness push with 3 YouTube creators and finance wants proof it works. Write the influencer measurement framework before we sign."
- "We have promo codes and GA, nothing else. What can we honestly measure on this creator campaign, and what should the reporting cadence be?"
## Reference
- Read [references/kpi-definitions.md](references/kpi-definitions.md) when selecting the KPI set and writing the glossary - formulas, KPI sets per objective, ER denominators, the EMV/AVE case, target-setting rules.
- Read [references/attribution-and-data-sources.md](references/attribution-and-data-sources.md) when mapping KPIs to methods and sources - method-by-method failure modes, code leakage, platform data access, view-definition changes, podcast/newsletter measurement, B2B HDYHAU design.
- Read [references/incrementality-and-brand-lift.md](references/incrementality-and-brand-lift.md) when deciding the test design - geo-lift/synthetic control, the MDE power check, brand-lift options and their conflicts, decision thresholds.
- Read [references/worked-examples.md](references/worked-examples.md) when shaping the deliverable - one B2C conversion plan, one B2B always-on plan, one annotated negative example.
- See `mbfinotti/partnerships-skills@influencer-campaign-brief` for stating the campaign goal and deliverables handed to the creator - this skill designs how that goal gets measured.
- See `mbfinotti/partnerships-skills@affiliate-payout-audit` for payout and commission math - never done here, even for affiliate-linked campaigns.
- See `mbfinotti/partnerships-skills@influencer-discovery-brief` for sourcing and vetting the creators being measured.
- See `mbfinotti/partnerships-skills@affiliate-disclosure-compliance` for disclosure obligations - out of scope here.