In short
- A library is a set of graded learning cards, each with a scoped claim, evidence, a grade and a review date. A list of campaign notes is not one.
- Grade every card: A for a replicated controlled test, B for one test or a three-campaign pattern, C for one observation. A and B set defaults; C seeds tests.
- Capture three to five cards at each day-30 review, including tests that found nothing, and keep a ledger of tests run against tests won.
- Make retrieval a step in every brief, and write into the agency contract that the brand owns the data and the library.
An influencer campaign learnings library is a searchable set of graded learning cards, each holding one scoped claim, the evidence behind it, a confidence grade and a review date. It turns what a repeat programme discovers into something the next brief can use. Without it, each campaign relearns the last one, and a long-term creator programme becomes a series of short ones.
Our post-campaign analysis checklist ends each campaign with a learnings log. This post covers the layer above: how those rows become cards, how to grade them, how to find them at brief time and how to keep them current. It sits inside the plan in the complete guide to influencer marketing in India, and it feeds on the numbers defined in how to measure influencer marketing ROI. The memory gap is common: 46.4% of surveyed Indian brands judge performance campaign by campaign, per a 2026 vendor survey covered by Campaign India, and 72% prefer long-term creator partnerships, per a 2025 Kantar-backed study also covered by Campaign India. The two statistics sit uneasily together.
A log is not a library
| Learnings log | Library | |
|---|---|---|
| Unit | A row in one campaign's file | A graded card |
| Scope | The campaign it came from | Stated: category, platform, tier, season |
| Evidence | What the team remembers | Campaign IDs, sample sizes and the design |
| Finding it | Scroll and hope | Search by tag at brief time |
| Lifespan | Forever | Reviewed, and retired when conditions change |
| Failures | Sometimes recorded | Always recorded |
The learning card
| Field | What goes in it | Rule |
|---|---|---|
| ID | L-014 | Sequential, never reused |
| Claim | "If we do X, then Y happens, for Z" | One claim per card |
| Evidence | Campaign IDs, metric, sample, window | Links to the locked files |
| Design | Audience split, matched pairs, holdout, or observation | Decides the grade |
| Grade | A, B or C | By the rule below, not by seniority |
| Scope | Category, platform, tier, language, price point, season | The conditions it was seen under |
| Review date | Six months out by default, twelve for A | Nothing lives forever |
| Status | Hypothesis, default or retired | Defaults feed the playbook |
| Action and owner | What changes, and who changes it | A card without an action is a note |
A card should be sharper than the advice it replaces. "Short videos work better" fails on every field: no scope, no evidence, no test. A usable version reads: "For Hindi micro-creator Reels for a skincare serum, Reels under 30 seconds held more viewers to the end than Reels over 45 seconds, in three campaigns, with no control." That is a grade B card with a scope, an evidence trail and an obvious next test.
Grades, and what each may be used for
| Grade | Meaning | Use |
|---|---|---|
| A | A controlled test that has been replicated, or is large and clear | May set defaults in briefs and budgets |
| B | One controlled test, or the same pattern across at least three campaigns without a control | Guides briefs; verify before large spend |
| C | One observation, or an opinion | Seeds a test; never a rule |
Controlled means an audience split, matched creator pairs or a holdout, as set out in our guides to A/B testing and incrementality testing. Grade by the design on the card, and let someone other than the person who ran the test apply the rule.
Tags that make cards findable
A library is only as good as the tags on it. Use controlled lists with about eight values each and one owner.
| Facet | Example values |
|---|---|
| Goal | Awareness, traffic, sales, installs, leads, footfall |
| Lever | Hook, offer, creator tier, language, format, timing, amplification, landing page, tracking |
| Platform and format | Instagram Reel, Story set, YouTube integration, Short, LinkedIn post |
| Creator tier | Nano, micro, mid, macro |
| Language or region | Hindi, Tamil, metro, tier 2 |
| Offer | Code depth, free gift, none |
| Season | Festive, sale, exam, wedding, ordinary |
| Funnel stage | View, click, purchase, repeat |
Every card carries at least a goal, a lever and a platform. Add new values by rule, at the quarterly review, so that "Reel" and "reels" and "short video" do not become three tags.
Four cards, filled in (made-up)
| ID | Claim | Evidence and design | Grade | Scope | Action |
|---|---|---|---|---|---|
| L-007 | Problem-first openings beat demo-first openings on click-through | Two audience-split tests of the same creator video, 1,20,000 impressions per arm, gaps beyond chance | A | Skincare serum, Instagram Reels, Hindi, ordinary season | Default to problem-first; re-test in a new category |
| L-011 | Regional-language Reels converted at lower cost per order than Hindi Reels | ₹610 against ₹790, 9 and 8 posts, no control | C | Haircare, one southern cluster | Seed a matched-pair test |
| L-014 | Creators with under 40% of their audience in the target market under-delivered on cost per result | The same pattern in three campaigns, observational | B | D2C consumables, micro tier | Add an audience-fit check to vetting |
| L-019 | Cutting the code discount from 15% to 10% did not change conversion by a detectable amount | Landing-page split, 9,000 visitors per arm | B | The same category | Use 10% next wave; test 5% |
Card L-019 is a null result, and it needs its limits written on it. With 9,000 visitors per arm, a 2% conversion rate could only have shown a change of about 30% or more, as the sample-size table in our A/B guide shows. The honest claim is that no change that large appeared, not that nothing changed. Cards that say "no effect" without the sample behind them mislead the next team.
Capture: three to five cards at the day-30 review
Hold a 60-minute session after the data lock described in the checklist. Ask three questions.
- What surprised us?
- What did we change mid-flight, and what happened?
- What would we do differently, and what evidence do we have?
Write at most five cards, name the design on each, and include at least one test that found nothing. Also keep a test ledger: every test run, with its design and outcome, including the ones that showed no difference. A made-up year might show 14 tests, of which 5 found a clear difference, 4 were ties and 5 were inconclusive. That ratio keeps expectations realistic and stops the library from reading as a series of wins.
Retrieval: make it a step in the brief
A library nobody searches is an archive. Before each brief, filter by goal, platform and tier, and read every A and B card that matches. Cite card IDs in the brief, for example "Uses L-007 and L-014", and list matching C cards under "to test". The default cards become a one-page playbook that a new strategist can read in ten minutes. Our guide to briefing an influencer marketing agency shows what the brief holds, and the first 30 days with an agency is the right moment to agree the card format and the tag list.
A search, worked through
A skincare brand is briefing a sales campaign of Hindi micro-creator Reels in an ordinary season. The strategist filters the library by goal "sales", platform "Reel" and tier "micro". Three cards return. L-007, the grade A card on problem-first openings, matches the scope exactly, so the brief tells creators to open with the problem and cites the card. L-014, the grade B card on audience fit, matches the tier and the D2C category, so the shortlist rule gains a 40% audience-fit check. L-011 comes from haircare in one cluster, so it is only a seed: the plan reserves two matched pairs for a regional-language test. Twenty minutes of searching changed three lines of the brief, and each change has a reason on file.
Use the low grades to plan the test queue. Where a lever takes a large share of spend and its only evidence is grade C, test it first. The forecast guide shows how test results become the rates you plan with.
Review, promote and retire
| Event | Action |
|---|---|
| A second campaign confirms a C card | Promote to B |
| A controlled test replicates a B card | Promote to A |
| A result contradicts a card | Keep the card, add the new evidence, drop one grade and schedule a test |
| A platform changes a format or rule | Mark every card for that platform "review now" |
| The review date arrives | Re-check, renew or retire |
| The season turns | Check seasonal scope on every default |
Hold a 45-minute review each quarter, as part of the monthly and quarterly rhythm. Retired cards stay in an archive with the reason, because the reason is often the next test.
Ownership and portability
The library has value only if it stays with the brand. Agree these in the contract and in the first month:
- Raw data: post-level exports, creator IDs, links, codes and tracking set-up, handed over at each lock and on exit.
- Locked files and test designs, in an agreed format such as a sheet or CSV.
- The library itself, in the card format and tag list you chose.
- A handover step: if you change agency, the library is the briefing document for the new one.
Our guide to the influencer marketing agency contract covers the clauses. This is general information, not legal advice.
Where to keep it
A shared spreadsheet with filters and fixed drop-down lists is enough for the first couple of hundred cards. Switch on version history, name one owner and protect the tag lists from casual edits. Do not spend a month choosing software before you have ten good cards.
Mistakes to avoid
- Writing learnings as vague advice such as "short videos work better".
- Leaving out the scope, so a result from one category is applied to every other.
- Grading by who argued loudest.
- Recording only wins.
- Never searching the library, so briefs do not change.
- Letting cards live forever after platforms and seasons change.
- Keeping the library with the agency.
If you want a library and a test ledger set up as part of your next creator programme, send us a brief. Our influencer marketing strategy service covers audits and the plan behind them, and the rates guide covers the fees behind every campaign the library remembers. The creator scorecard is the companion record for individual creators.
Sources
- Coverage of a 2026 India creator-economy report: brand survey on revenue targets and campaign-by-campaign judging, Campaign India (21 May 2026)
- 69% of Indian consumers turn to influencers for information and 63% for product discovery (consumer and brand survey coverage), Campaign India (12 Jun 2025)
Frequently asked questions
What is a campaign learnings library?
It is a structured, searchable record of what your influencer campaigns have taught you, kept as cards that each state a scoped claim, the evidence and design behind it, a confidence grade and a review date. It lets a new team member or agency find what is already known before planning the next campaign.
What should a learning card contain?
An ID, one claim written as if-then, the evidence with campaign IDs, metric, sample and window, the design, a grade from A to C, the scope in which it was observed, a review date, a status and the action it changes. Keep it to one claim per card so that it can be graded and retired on its own.
How do I grade how reliable a learning is?
Grade A for a controlled test that has been replicated or is large and clear. Grade B for one controlled test, or the same pattern across at least three campaigns without a control. Grade C for a single observation or an opinion. Use C to plan tests, never to set rules. See our A/B testing guide.
How many learnings should we record after each campaign?
Three to five. More than that dilutes attention and fills the library with weak cards. Choose the ones that would change a decision next time, include at least one test that found nothing, and note the design on every card so the grade can be checked.
Where should a learnings library be kept?
A shared spreadsheet with filters and fixed drop-down lists is enough for the first couple of hundred cards. What matters is one owner, a controlled tag list, version history and a step in every brief that searches it. Move to a heavier tool only when search becomes slow.
Who owns the learnings if I change agency?
Write into the contract that the brand owns its campaign data, creator IDs, tracking set-up, locked files and the library, and that the agency hands them over in an agreed format at each lock and on exit. Our agency contract template guide covers the clauses. This is general information, not legal advice.