Payments are experiencing issues due to temporary restrictions in Russia. If your payment does not go through, please submit a support request.Our support team is available 24/7 — we are always here to help with hosting and server issues.We are now accepting requests for dedicated server rental and colocation services in our data center.Reminder: we recommend enabling backups for additional data protection.A new VPS/VDS lineup with NVMe storage and improved performance is now available.Maintenance work on some servers has been completed. All services are operating normally.
Article7 min readViews0

Order Returns: When Automated Reconciliation Is Cheaper Than Manual Processing

The reference model shows at what return volume automation becomes cheaper than manual reconciliation, and which repetitions, exceptions, and failures must be accepted before launch.

Comments 0

A returned package passes through a control ring next to the receipt, payment token, and confirming stamp.
In this article

In the reference model, costs for manual and automated reconciliation are equal at 75 returns per month. From the 76th return onward, automation becomes cheaper for the first time: 16,920 RUB versus 17,100 RUB. This is not a ready-made standard for a store, but a boundary under specific assumptions. If processing time, the share of exceptions, or support costs change, the answer will change.

For the calculation to make sense, the same process must be compared. For a full return, I include four control points: the order, the payment document, the return receipt, and the payment system record. If automation covers only one of these while the manager still manually reconciles the rest, the savings are illusory—a calculator typically does not care that the process is incomplete.

First, align the process scope

In the official 1C-Bitrix lesson the "Refund" action in the payment document is separated from "Cancellation": in the supported scenario, the first sends data to the payment system, while the second only marks the order as cancelled in Bitrix. There, the refund receipt is listed as a separate step. This is important for the estimate: changing the local status does not yet confirm a money refund and does not replace the receipt.

For the YooKassa handler, the refund via payment status is described in the 1C-Bitrix documentation. This scenario requires preliminary configuration and applies to the "Online Store" module version 17.6.0 and above. This condition cannot be automatically transferred to another handler or an older version: before evaluation, you must check the specific payment system and module edition.

  • One return in the model equals one completed full return, not a single button click.

  • Manual and automated variants verify the same four control points.

  • Partial refunds, warehouse receipts, accounting entries, and logistics are included only when they are actually present in both variants.

  • A message or request repetition is considered the same refund if it shares the same persistent operation identifier.

If the scope differs, align it first. For example, comparing manual checks of an order, payment, receipt, and registry against automation of only the payment status tells you nothing about the total process cost.

Assumptions of the reference calculation

Let's calculate one transparent scenario. All figures below are hypothetical: they are not General iT tariffs, not average market prices, and not a promise of return on investment.

Parameter

Value

Why is it needed

Manual verification of a single return

15 minutes

Time for four control points

Employee hourly rate

900 ₽

Total cost of working time in the model

Automation development

180,000 ₽

Distributed over 24 months

Monthly support

6,000 ₽

Fixed portion after launch

Share of exceptions

30%

Returns requiring manual review

Analysis of a single exception

10 minutes

Manual variable work after automation

Manual processing costs 15 × 900 ÷ 60 = 225 RUB per return. With R returns per month, the total is 225 × R.

The fixed part of automation equals 180 000 ÷ 24 + 6 000 = 13 500 RUB per month. The variable part: 30% × 10 minutes × 900 RUB ÷ 60 = 45 RUB on average per return. Total: 13 500 + 45 × R.

Where the decision changes

Equating costs: 225 × R = 13 500 + 45 × R. After rearranging, we get 180 × R = 13 500, so R = 75. At exactly 75 returns, both options cost 16 875 RUB. The first whole volume where automation becomes cheaper is 76 returns.

Returns per month

Manual reconciliation

Automation

Difference

40

9 000 ₽

15 300 ₽

Manual is cheaper by 6 300 ₽

75

16 875 ₽

16 875 ₽

Equality

76

17 100 ₽

16 920 ₽

Automation is 180 ₽ cheaper

100

22 500 ₽

18 000 ₽

Automation is 4 500 ₽ cheaper

200

45 000 ₽

22 500 ₽

Automation is 22 500 ₽ cheaper

Two monthly cost lines intersect at 75 returns and 16 875 rubles
Educational model: A is manual reconciliation at 225 RUB per return, B is automation at 13 500 RUB + 45 RUB per return. The lines intersect at 75 returns and 16 875 RUB per month.

A difference of 180 RUB on the 76th return is too small to base a decision on a single point. We need a buffer for measurement inaccuracies, seasonality, and tasks that were forgotten. Therefore, I would look not only at the intersection of the lines but also at the adverse scenario.

The threshold is sensitive to three values

  • If manual reconciliation takes 10 minutes, its cost drops to 150 ₽. Break-even occurs at approximately 128.6 returns, and automation becomes cheaper for the first time starting with the 129th return.

  • If exceptions account for 50%, the variable cost of automation rises to 75 ₽. Break-even is 90 returns; the first whole-unit gain occurs with the 91st return.

  • If support costs 10 000 ₽ per month, the fixed component rises to 17 500 ₽. Break-even is approximately 97.2 returns; the first whole-unit gain occurs with the 98th return.

Change one parameter at a time and preserve the source of the numbers. Otherwise, after several edits, it becomes impossible to understand why the threshold shifted and what initial data were used.

What must be included in the full estimate

Development is just one line item. Before selecting a variant, verify that state descriptions, payment system integration, receipts, operation logs, permissions, notifications, exception handling, monitoring, test environments, and post-API-change support are included.

  • One-time expenses: assessment, development, migration of incomplete returns, training, and acceptance testing.

  • Ongoing expenses: support, monitoring, updates, and re-verification after platform or payment processor changes.

  • Variable costs: manual review of exceptions, customer inquiries, and reconciliation of discrepancies.

  • Cost of error: repeated financial impact, missed receipt, false completed status, or lost event during external system unavailability.

It is useful to compare the calculation with a review of saved hours and actual costs. There, released time and actual cost reduction are separated — for refunds, this distinction also holds.

How to accept automation

Acceptance must verify not a polished demonstration, but the final state and failure scenarios. Every operation needs a stable identifier: repeating the same event must return the same result and not create a second refund.

  1. Successful scenario: the order, payment, receipt, and external registry contain a single operation with a consistent amount.

  2. Duplicate: a secondary delivery of the same event does not change the financial result and does not create a second receipt.

  3. Rejection before external effect: the operation remains incomplete, the reason is visible, and retry is safe.

  4. Unknown result: after a timeout, the system first checks the state of the external operation rather than unconditionally starting a new one.

  5. Exception: there is a reason, an owner, an action history, and a review deadline.

  6. Permissions: employees see and perform only authorized operations, while critical actions remain logged.

  7. Control month: the count and total amount of completed returns match at all four control points; discrepancies are analyzed by cause.

The general acceptance checklist for an online store helps prevent loss of access, roles, backups, and critical scenarios around this integration. It supplements but does not replace domain-specific return checks.

When automation is not needed

Manual processing may be more rational with low and irregular volumes, unstable processes, or a high proportion of unique exceptions. Automating a chaotic procedure is an expensive way to get unpredictable results faster.

Practical next step: take one full month with an identical set of operations. Measure the total time for manual reconciliation and the number of completed returns R; to calculate costs, divide the total time by R to obtain the average time per return. For exceptions, measure their count and total review time separately, then find the average time per exception. The median is useful as a characteristic of a typical case, but it does not replace the average when calculating total labor costs. For example, in a sample dataset, nine returns taking 1 minute each and one taking 101 minutes total 110 minutes: the average is 11 minutes, while the median is 1 minute. Multiplying this median by 10 returns yields only 10 minutes instead of 110. Substitute the measured values and prices from actual commercial offers for development and support into the formulas, calculate the expected and adverse scenarios, and after launch, recalculate the model based on the results of the first full month. The number 76 applies only if your assumptions match those of the sample example; it is not a universal threshold.

Discussion 0

Share your experience and ask questions. Comments without links appear after editorial review.

No comments yet. Start the discussion.