Definition
Cohort analysis is a method of grouping buyers by a common attribute (usually the date of the first purchase) and tracking key metrics for those groups over time to understand behavior and retention. The goal is to reveal how the value and activity of each cohort change after the first interaction.
How cohort analysis works
A cohort is typically formed by the date of the first order, but you can use other attributes: registration date, traffic source, promo-code channel, or the item of the first purchase. For each cohort you then build time series of metrics: the share of users who made a repeat purchase after N days, average order value, LTV, churn rate.
- Step 1 — forming the cohort: for example, all buyers who made their first purchase in January 2025.
- Step 2 — choosing metrics and time intervals: retention at 7/30/90 days, average order value, % returns, ARPU, LTV.
- Step 3 — visualization: a table where rows are cohorts (Jan, Feb, Mar), columns are days/weeks/months after the first purchase, and cells show retention percentage or average order value.
- Step 4 — comparison and hypotheses: compare cohorts to each other to identify the impact of promos, changes in the product card, or seasonality.
Practical analysis frequency: for fast-moving products — daily/weekly cohorts; for products with a purchase cycle of 3–6 months — monthly cohorts.
Which metrics to watch and how to calculate them
Key metrics in cohort analysis for a Kaspi.kz seller:
- Retention rate: the share of cohort customers who made at least one purchase in the selected period. Formula: retained / cohort_size × 100%.
- Repeat purchase rate: % of customers who made a second purchase in a given interval (30/90 days).
- Average order value (AOV): the cohort's average check across periods.
- LTV (lifetime value): cumulative revenue from the cohort over N days. For Kaspi evaluation, LTV30, LTV90, LTV365 are common.
- CAC (customer acquisition cost): if you have advertising data; compare CAC with LTV to assess payback.
Example: the January cohort has 1,000 buyers. After 30 days, 180 people made a repeat purchase. Retention30 = 180 / 1000 = 18%. The average repeat purchase check for this cohort is 6 500 тг, initial AOV was 8 000 тг. LTV30 = 8 000 + 6 500 × (180/1000) ≈ 9 170 тг.
Why sellers on Kaspi.kz need it
Cohort analysis provides actionable insights rather than abstract numbers:
- Understand promo response: if you ran a Kaspi sale in March and March cohorts show retention30 = 12% while cohorts without the promo have 18%, cohort analysis helps quantify the short-term lift and long-term cannibalization.
- Measure the impact of changes to the product card: updates to photos, descriptions, or delivery terms may change conversion and repeat purchase behavior; cohorts let you see when and how those changes affect customer value.
- Detect quality issues: a cohort with high returns or complaints will show lower LTV and retention — this signals problems with SKU quality, packing, or logistics.
- Optimize marketing spend: by comparing CAC to cohort LTV you can decide which channels bring valuable customers and which only drive one-off buyers.
- Prioritise retention tactics: cohorts highlight windows where interventions (SMS, push, coupons) are most effective for driving a second purchase.
Examples of scenarios on Kaspi.kz with numbers
Scenario 1 — Promo that increases acquisition but lowers retention:
- In March you ran a 20% off promo and acquired 3,000 new customers (March cohort). Retention30 = 12%, AOV first order = 6 000 тг, repeat AOV = 5 500 тг.
- In January (no promo) you acquired 1,000 customers, Retention30 = 18%, first AOV = 8 000 тг, repeat AOV = 6 500 тг.
- Interpretation: promo brought volume but attracted more price-sensitive buyers with lower repeat rates and smaller checks. Compare CAC and LTV30 to evaluate if the promo is profitable long-term.
Scenario 2 — improving repeat purchases via bundled offers:
- April cohort (after introducing an accessory bundle) shows Retention30 up from 15% to 21% versus March; the average repeat check grows from 5 500 тг to 6 300 тг.
- If the number of repeat buyers increased by 40% and repeat AOV increased by 14%, LTV30 can rise substantially — calculate the incremental revenue per original cohort member to estimate ROI of the bundling initiative.
Scenario 3 — long-cycle product (3–6 months):
- For a product with a typical repurchase at ~120 days, compare retention at 90 and 180 days. If Retention90 = 8% and Retention180 = 12% (cumulative additional buys), the real purchasing behavior may be distributed over months and immediate 30-day metrics would underestimate value.
Practical tips for implementation and tools
- Choose cohort granularity by business needs: use daily/weekly cohorts for fast-moving SKUs, monthly or quarterly for low-volume or long-cycle goods.
- Persist the first purchase date: store an immutable first_order_date per buyer_id so cohorts don’t shift as you update data.
- Count net transactions: exclude fully returned or cancelled orders when counting repeats (see next section for details).
- Automation: set up a daily ETL that pulls Kaspi Seller exports (buyer_id, order_date, order_value, return_status) and updates the cohort matrix in your BI.
- Tools: simple cohorts can be built in Excel or Google Sheets; for scale use BI tools like Metabase, Looker, Power BI, or your in-house dashboards. Ensure links to Kaspi Seller exports are stable and document field definitions.
- Quality controls: add checks for duplicates, timezone alignment, missing buyer_ids, and consistent return statuses.
- Segment cohorts: split by traffic source, promo tag, SKU category, or first item to gain deeper insight into which customer types bring long-term value.
Typical mistakes and limitations
- Using too small cohorts: weekly cohorts for a low-volume shop produce noisy results. Merge periods or use larger windows to reach statistical significance.
- Counting gross instead of net: including returned or fully cancelled first purchases inflates cohort size and distorts retention and LTV.
- Ignoring seasonality: comparing winter and summer cohorts without season adjustment can mislead — always compare similar seasons or use seasonally adjusted baselines.
- Attributing everything to promos: changes in assortment, pricing, or marketplace policy can also affect cohorts; check multiple hypotheses before drawing conclusions.
- Confusing correlation with causation: cohorts show correlations — confirm drivers with A/B tests or controlled experiments where possible.
Conclusion
Cohort analysis is a practical, interpretable method to understand customer retention and lifetime value on Kaspi.kz. For sellers, it helps measure the real effect of promos, product-page changes, and retention tactics — and informs which segments deserve marketing investment. Start simple: pick a cohort granularity that fits your volume, persist first-purchase dates, exclude returns from net metrics, and build an automated report that updates regularly. Over time, cohorts will become a key tool to grow repeat purchases and improve unit economics.
Часто задаваемые вопросы
- How to form cohorts on Kaspi.kz for a shop with irregular sales and low order volume?
- Use larger time windows — for example, monthly or quarterly cohorts instead of weekly ones to obtain statistically meaningful samples. You can merge adjacent periods (rolling cohorts) or group cohorts by product type/category. Also set a minimum cohort size threshold and mark small cohorts as "insufficient data" for reliable interpretation.
- Which metrics are most important in cohort analysis for products with a 3–6 month purchase cycle?
- Look at retention at 90 and 180 days, repeat purchase rate for the same intervals, and cumulative LTV over 6–12 months. Analyze average order value and purchase frequency per user, as well as the return rate — these strongly affect a cohort’s true value. With long cycles, account for cohort size and seasonality when interpreting results.
- What does a sharp drop in retention for a specific cohort in the third month mean and what initial checks should be done?
- A sharp drop can indicate seasonality, the fading of a promo effect, stock issues, or an increase in returns/cancellations. Compare the promo schedule, prices, stock availability and reviews for that cohort versus previous ones; break down by traffic source and SKU. If needed, run hypotheses: test a re-engagement campaign (email/SMS) and analyze return reasons for quick diagnosis.
- How to correctly account for returns and cancellations in cohort analysis on Kaspi.kz?
- It's better to count retention and repeat purchases using net transactions, excluding fully returned or cancelled orders from repeat purchases. Adjust the initial cohort if the first purchase was fully returned — such buyers should not be included in the cohort size. Separately calculate the % returns per cohort to understand their impact on LTV and retention.
- How to automate updating cohort reports when integrating with Kaspi Seller data exports?
- Set up a daily ETL that pulls orders with fields buyer_id, order_date, order_value and return status; persist an immutable first purchase date for each buyer. In your BI tool, automate building the cohort matrix (rows — cohorts by first purchase, columns — periods after purchase) and schedule regular updates. Include data quality checks: duplicate detection, timezone consistency and verification of return statuses.
