Site, CRM and receipts: abandoned carts, the delivery threshold, payments, message frequency, the lapsed threshold, day indices and pairs of items
The document gathers seven small calculations that answer the questions of an online shop and a customer base: whether the abandoned cart chain pays back, where to set the free shipping threshold, what payments and instalments cost, how often to write to subscribers, when a customer is considered lapsed, which days are strong and which items are bought together. The decision it supports: which mechanics to keep and how to test them.
Why this document
Each of these mechanics looks profitable on average numbers and turns out disputable by the distribution: the delivery threshold by the average check, frequency by opens, a lapsed customer by a round 90 days. The document teaches to look at the distribution, the incremental effect and significance. It is read by the online shop manager, the CRM manager, the category manager and the analyst. There are no norms: you take thresholds and parameters from your own data and check them with a test.
At a glance
- What is this document and why is it needed?
- The document gathers seven small calculations that answer the questions of an online shop and a customer base: whether the abandoned cart chain pays back, where to set the free shipping threshold, what payments and instalments cost, how often to write to subscribers, when a customer is considered lapsed, which days are strong and which items are bought together. The decision it supports: which mechanics to keep and how to test them.
- Who owns the document and how often is it rebuilt?
- Brand and retail: the online shop and CRM manager runs the calculations, the analyst checks the tests, the finance director approves budgets; recalculate once a season.
- What mistakes are the most common?
- Not subtracting orders the customer would have made on their own → the chain looks more profitable → a test with a control group. Choosing the delivery threshold by the average check → the distribution is hidden → count by bands. Counting the instalment fee from margin → costs are understated → count from turnover.
For whom
- Brand
- A brand with its own site wins more often from the delivery threshold and the referral reward and builds bundles from pairs of items from the receipts of its own shop.
- Retail
- Retail relies on shop receipts: day indices and co-purchases are counted from them, and the delivery threshold and carts from the site.
What it consists of
Abandoned carts: the payback of the chain
- Abandoned = sessions with a cart × share abandoned; reachable = abandoned × share with a contact and consent; returned = reachable × response; incremental = returned × (1 − the share of those who would have ordered anyway). Profit = incremental × check × margin × redemption − chain cost.
- The break-even response = cost ÷ (reachable × check × margin × redemption) shows at which response the chain breaks even. One may remind only with the customer’s consent; the share “would have ordered anyway” is determined by a test with a control group without reminders.
The free shipping threshold by the distribution of checks
- Checks are split into bands, the threshold coincides with a band boundary. Free at the threshold = the share of checks not below the threshold; at the new threshold a catch-up is added: a share of buyers from the band right under the threshold add an item to get free delivery.
- The change in profit per order = delivery saving + catch-up margin − losses. The average check hides the distribution, so a threshold by the average is often wrong; the model is simple, and the threshold is tested on part of the traffic, comparing with the competitors’ threshold.
Payments and instalments
- The average fee = the sum of (share of the method × rate of the method). Take rates from contracts on the date: they change, fast payments are usually cheaper than cards. The instalment fee is charged on the whole order and paid by the shop.
- The break-even conversion growth with an instalment = margin ÷ ((1 + check growth) × (margin − instalment fee)) − 1. If the fee is not below the margin, the instalment loses money on every order. Count the fee from turnover and not from margin and account for returns.
Message frequency and the lapsed threshold
- The return per subscriber grows ever more slowly with frequency while losses grow linearly: the optimum f* = k × ln(ceiling ÷ (k × (sending cost + unsubscribe share × future margin))). Send only with consent, the unsubscribe works at once; count channels separately.
- The “lapsed” threshold is taken as a quantile of the gaps between repeat purchases (position (n − 1) × share of repeats, linear interpolation). A customer is sleeping after the threshold and lost after the threshold times a multiplier. Count by category and season and do not take a common 90 or 180 days.
Day-of-week indices and pairs of items in receipts
- Day index = the average sales of the weekday ÷ the average over all days; count on ordinary weeks and untick holiday ones. An index above the threshold marks strong days for promotions and mailings, a low one shows days that need pulling up.
- For a pair of items support = receipts with both ÷ all receipts, confidence A → B = support of the pair ÷ support of A, lift = support of the pair ÷ (support of A × support of B), leverage = support of the pair − support of A × support of B. Pairs with support below the set level (for example 0.5%) are not ranked. A lift above one shows co-occurrence but not a cause; receipt data tied to a customer is personal, use aggregates.
Where the data comes from
| Data source | Fields |
|---|---|
| The site and analytics | sessions with a cart, abandoned, contacts with consent, the distribution of checks by band, delivery cost |
| Payment contracts | shares and rates of payment methods, the instalment fee |
| CRM and mailings | the base size, return and unsubscribes by frequency, gaps between purchases, recency of the last purchase |
| Tills and receipts | sales by weekday and hour, the number of receipts, receipts with each item and with a pair |
How to build it step by step
- Step 1: Count the abandoned cart chain and the break-even response, determine the share “would have ordered anyway” by a test.
- Step 2: Lay out checks in bands and compare delivery thresholds at band boundaries.
- Step 3: Enter the shares and rates of payment methods from contracts and find the average fee.
- Step 4: Compare the break-even conversion growth for the instalment with the expected one and check with a test.
- Step 5: Find the optimal message frequency and check it on groups of subscribers.
- Step 6: Derive the lapsed threshold from gaps by category and launch the win-back for sleeping customers.
- Step 7: Build day-of-week indices on ordinary weeks and place promotions on strong or weak days.
- Step 8: Count pairs of items and test a bundle or adjacent placement in part of the shops.
Formulas
| Measure | Formulas | Note |
|---|---|---|
| Change in profit from the delivery threshold | (free before − after) × delivery + share catching up × band share × added amount × margin − losses | Losses = share lost × share starting to pay × order margin. |
| Lift of a pair | support of the pair ÷ (support of A × support of B) | One means a chance match. |
| Optimal message frequency | k × ln(ceiling ÷ (k × (cost + unsubscribe share × future margin))) | Zero if the ceiling does not cover the loss of the first message. |
Worked example
Illustrative data of an online shop and a subscriber base
| Metric | Value |
|---|---|
| Orders returned from carts | 392 |
| Chain profit, RUB | 471,787 |
| Change in profit from a 4,000 threshold instead of 3,000, RUB per order | 53.7 |
| Optimal frequency, messages a month | 4.76 |
| The “sleeping” threshold, days | 248 |
| Lift of the best pair | 2.22 |
The average payment fee is 2.03%, the break-even conversion growth with an instalment 3.3%. All numbers are illustrative.
Figures
Data as a table
| Measure | Change per order |
|---|---|
| 2,000 | −77.7 |
| 3,000 | 0 |
| 4,000 | 53.7 |
| 5,000 | 83.2 |
Change in profit per order by delivery threshold, RUB. A profitable threshold is where the delivery saving exceeds the lost margin.
Data as a table
| Measure | Index |
|---|---|
| Mon | 0.71 |
| Tue | 0.78 |
| Wed | 0.74 |
| Thu | 0.85 |
| Fri | 1.14 |
| Sat | 1.49 |
| Sun | 1.29 |
Day-of-week index. Weekends are noticeably stronger than weekdays.
Specifics by category
- Apparel
- Apparel: many sizes and seasonality, so count redemption, purchase gaps and day indices by season.
- Footwear
- Footwear: redemption is lower because of fitting, and reverse logistics noticeably changes the cost of a redeemed unit.
- Accessories
- Accessories: the check is small, pairs with apparel give a lift; the delivery threshold and bundles work stronger.
- Kids
- Kids: frequent repurchases by height, gaps between purchases are shorter, the lapsed threshold is lower.
- Men
- Men: a steady core range, less frequent purchases, a longer lapsed threshold.
- Women
- Women: more novelties and seasonal peaks; count day indices and pair lift separately by season.
Common mistakes
- Not subtracting orders the customer would have made on their own → the chain looks more profitable → a test with a control group.
- Choosing the delivery threshold by the average check → the distribution is hidden → count by bands.
- Counting the instalment fee from margin → costs are understated → count from turnover.
- Counting only email opens → the future margin of those who unsubscribed is lost → count profit per subscriber.
- Taking fixed 90 or 180 days → false lapsed → a threshold from your own gaps by category.
- Mixing holiday and ordinary weeks → indices are distorted → untick holiday weeks.
- Taking pairs with support below 0.5% → the lift is large by chance → set a minimum support.
Russian specifics
- Consent to messages (the personal data law and the advertising law) is mandatory for mailings and reminders; receipt data is personal when tied to a customer. Take payment fee rates from contracts on the date.