In short
- A cohort table groups customers by first-order month and source, then tracks repeat purchase by age. Compare creator, paid and organic cohorts at the same age.
- Gaps between sources are often within noise. With 640 creator customers and 2,100 paid ones, a 4-point gap in 180-day repeat rate has a 95% range of about ±4 points.
- Compare within the first product, not across the mix. In our made-up case creators repeat more in each product segment and less overall.
- Reading a 25% repeat rate within ±5 points takes about 290 customers. Pool creators by tier, language or format until each cell is that large.
Influencer customer cohort analysis groups customers who placed their first order in the same month through the same source, then follows how many reorder as time passes. It answers a question averages hide: do creator-acquired customers come back as often as customers from paid ads or search? The answer depends on the product, the offer and the creators, and we found no published India evidence for either direction.
Our guide to influencer marketing LTV and CAC works one cohort through to payback. This post goes deeper on the table itself: how to build it from clean data, how to read the shape, how much noise sits in each cell and how a mix trap can reverse the conclusion. It belongs beside the complete guide to influencer marketing in India and the definitions in how to measure influencer marketing ROI. The stakes are real, because 72% of surveyed Indian brands prefer long-term creator partnerships, per a 2025 Kantar-backed study covered by Campaign India, and cohort data is how you find out which long-term deals deserve to exist.
What a cohort table shows
Rows are groups of customers, such as "March first orders, from creators". Columns are age, such as 30, 60, 90 and 180 days after the first order. Each cell holds a rate or an amount, such as the share who have reordered or the cumulative contribution per customer. The rule that matters: compare at the same age, never at the same calendar date.
| Measure | What it tells you | How to compute it |
|---|---|---|
| Repeat rate at age X | Share of the cohort who reordered by X days | Customers with a second order by X ÷ cohort size |
| Orders per customer at age X | Depth of repeat | Orders after the first, by X ÷ cohort size |
| Median days to second order | The rhythm of repeat | Median gap among those who reorder |
| Second orders placed with a code | How much repeat is discount-led | Second orders with a code ÷ second orders |
| First-order return rate | Whether the product matched the promise | Returned first orders ÷ first orders |
| Cumulative contribution per customer | The money view, as in the LTV post | Sum of contribution after variable costs |
Get the data right first
Cohorts fail on identity, not on arithmetic. Customers order under a second phone number, a different email or a guest checkout. If you do not merge them, returning customers show up as new, which inflates acquisition and deflates repeat.
- Merge in a fixed order: phone number after cleaning country codes and spaces, then email, then delivery address.
- Freeze the source rule: code first, link second, survey third, as in the LTV and CAC guide, and never change it mid-year.
- Store the first product, the first discount and the creator ID at the first order, because you cannot rebuild them later.
- Accept the blind spot: marketplaces rarely share customer-level data, so their buyers are invisible. Our guide to measuring influencer impact on marketplace sales covers that side.
- Treat the table as a floor on creator-acquired customers. Only verified source tags count, as our attribution guide explains.
A worked cohort table (made-up numbers)
A hypothetical consumable D2C brand takes all customers whose first order fell in one month and splits them by source.
| Cohort | Customers | Repeat by day 30 | Day 60 | Day 90 | Day 180 | 95% range at day 180 |
|---|---|---|---|---|---|---|
| Creators | 640 | 8% | 17% | 24% | 33% | ±3.6 points |
| Paid social | 2,100 | 7% | 15% | 21% | 29% | ±1.9 points |
| Organic and other | 1,300 | 10% | 20% | 28% | 38% | ±2.6 points |
The range is 1.96 × √(p × (1 − p) ÷ n) for each cohort. Now compare creators with paid social at day 180. The gap is 4 points, and the 95% range on the gap is about ±4.1 points. That puts zero just inside the range: creators may repeat more, but this table does not prove it. Organic and other buyers include people who already knew the brand, which is one reason they may repeat more, though that is a reading, not a finding.
Build it in a spreadsheet
- Export every order with customer ID, order date, product, discount and source tag.
- For each customer, keep the first order's date, source, product and discount.
- Add the date of the second order, if any, and the days between the two.
- For each cohort, count customers whose second order falls inside 30, 60, 90 and 180 days, using only customers old enough to have reached that age.
- Divide by the eligible customers and add the 95% range, 1.96 × √(p × (1 − p) ÷ n).
A pivot table handles steps 4 and 5. Rebuild the table each month and keep dated copies, so you can watch a cohort move.
Read the shape, not only the level
Two cohorts can end at the same repeat rate by different routes. Plot each as a curve and ask whether it bends where the product says it should. A 60-day consumable should show a rise in reorders around days 45 to 75. A cohort with no such rise is repeating for other reasons, such as a reminder message or a sale, not because the product ran out. A cohort that repeats early and then flattens may have been bought through a discount.
Check the median days to second order against the product's real cycle. A cohort that reorders much later than the cycle suggests customers are not using the product as intended, a brief or onboarding problem, not a creator problem.
The mix trap
Customers differ by what they bought first, and creators may bring a different mix from paid ads. A blended comparison can then point the wrong way. A made-up example:
| First product | Creator customers | Creator repeat at day 180 | Paid customers | Paid repeat at day 180 |
|---|---|---|---|---|
| Refill-type product | 200 | 45% | 600 | 42% |
| One-off gift set | 600 | 15% | 200 | 12% |
| All customers | 800 | 22.5% | 800 | 34.5% |
Creators repeat more in both segments, by three points each, and look twelve points worse overall, because their mix leans toward the gift set, which few people reorder. The blended figure answers "who repeats more per customer acquired", which can be a fair question, but it says nothing about whether creator customers are better or worse buyers of the same thing.
Always split by first product, and by first-order discount as the LTV post shows, before comparing sources. Our post on tracking sales with discount codes covers discount depth and how it affects the customers you attract.
Age, season and promotion
- Compare at the same age. A cohort only 60 days old has no day-90 figure. Exclude customers younger than the age you are reading, or the cohort's repeat rate will look low.
- Compare like seasons. Customers acquired in a sale season or a festival differ from those acquired in a quiet month. Compare each with the same season last year, not with the month before.
- Do not project from the youngest cohort. Early repeat can mislead when purchase cycles differ by product. Our LTV post shows how to scale an older curve and label the result "projected".
How big a cohort must be
Repeat rates are shares, so the noise falls with the square root of the cohort size.
| Repeat rate | Customers for ±5 points (95%) | Customers for ±10 points |
|---|---|---|
| 10% | about 140 | about 35 |
| 25% | about 290 | about 75 |
| 40% | about 370 | about 95 |
A single creator with 60 customers and a 25% repeat rate carries a range of about ±11 points, so one creator's cohort rarely supports a ranking. Pool by tier, language or format until each cell holds a few hundred customers, and note that comparing two cohorts needs more than reading one.
From the table to a decision
| What the table shows | What to do |
|---|---|
| Creator cohort repeats at or above paid after pooling and splitting | Make the case for longer deals; our creator ambassador programmes are built for it |
| Repeat is concentrated in customers who used no code | Test a smaller discount in one wave |
| One tier or language cohort repeats more | Weight it in the creator scorecard and cast more of it |
| All cohorts repeat poorly | Look at the product and the post-purchase journey, not the creators |
| First-order returns are high in the creator cohort | Check that briefs describe the product accurately |
| The creator cohort repeats less | Compare the first-product mix before concluding anything |
In a 2026 vendor survey, 62.1% of brands agreed that long-term creator deals give better ROI. That is opinion. Your cohorts are the evidence for your brand, and our post on influencer marketing incrementality testing shows how to check how many of the customers in a cohort the creators caused.
What to report
Put one small table in the quarterly page: cohort, customers, repeat at day 90, the 95% range and the sentence "compared at the same age and split by first product". Keep the definitions identical from quarter to quarter, as our guide to reporting influencer marketing to leadership advises.
Mistakes to avoid
- Leaving duplicate customers unmerged.
- Comparing cohorts of different ages.
- Comparing sources without splitting by first product.
- Reading a single creator's cohort of 60 customers.
- Treating a gap inside the noise as a finding.
- Projecting a year of repeat from one month of data.
- Putting customer-level data in shared reports.
If you want cohort tracking built into your next creator programme, send us a brief. Our D2C brands page covers programmes built around CAC and repeat purchase, and the rates guide covers the fees behind acquisition cost.
Sources
- 69% of Indian consumers turn to influencers for information and 63% for product discovery (consumer and brand survey coverage), Campaign India (12 Jun 2025)
- India creator-economy research report page: brand behaviour survey, Q1 2026 (long-term creator deals and ROI; vendor estimate), Influencer-platform vendor research page (accessed 31 Oct 2026)
- DPDP Rules 2025 notified, with staged commencement, The Week (PTI) (14 Nov 2025)
Frequently asked questions
What is cohort analysis in influencer marketing?
It groups customers who made their first order in the same period and through the same source, such as creators, then follows how many reorder over the following weeks and months. It shows whether creator-acquired customers come back as often as customers from other channels, which a single average cannot show.
Do influencer-acquired customers repeat more than paid-ad customers?
We found no published India evidence either way. It depends on the product, the offer and the creators, so measure your own cohorts. Compare them at the same age, split by first product and discount, and check the gap against its noise before acting on it.
How do I tag customers by creator?
Store an acquisition source and creator ID on the customer record at the first order, using a fixed rule such as code first, link second, survey third. Merge duplicate customers by phone number and email. Our LTV and CAC guide has the field list.
How big does a cohort need to be?
For a 25% repeat rate, about 290 customers gives a 95% range of ±5 points and about 70 gives ±10 points. Comparing two cohorts needs more. A single creator with 60 customers cannot be ranked, so pool by tier, language or format first.
How long should I wait before reading repeat purchase?
Read at ages that cover at least one full purchase cycle for the product, and only for customers old enough to have reached that age. A 60-day consumable needs customers at least 90 days old. Young cohorts understate repeat, so never compare them with older ones.
Can I run cohort analysis on marketplace customers?
Usually not, because marketplaces do not share customer-level data with sellers. Run cohorts on your own-site buyers and use the marketplace methods in our guide to measuring influencer impact on marketplace sales for the rest. The own-site cohort then covers customers who bought directly, which may differ from marketplace buyers, so say which group the table describes.