# How should a DTC brand read attribution during a flash sale? | Attribution Playbook

Source: https://attributionplaybook.com/blog/2026-10-01-attribution-flash-sale-promotion.html

Oct 1, 2026 - Attribution Playbook


During a flash sale, stop trusting your normal attribution. Everything converts, consideration windows collapse to hours, and last-click gives all the credit to whatever the customer touched last, usually email or direct. The honest way to read a sale is incrementality: what revenue happened because of the sale that would not have happened anyway.

## Why sales break attribution models

Attribution models are built on normal buying behavior: discovery, consideration, comparison, purchase, spread over days or weeks. A flash sale compresses all of it into hours. Customers who were going to buy next month buy today. Customers who never considered you buy on impulse. The model's assumptions about journey length, touchpoint value, and baseline conversion all fail at once, and the model does not know it is failing. It keeps producing confident, wrong answers.

The most dangerous wrong answer is channel credit. During a sale, email and SMS look like heroes because they deliver the announcement and capture the click. But much of that revenue was already in the pipeline: existing demand, pulled forward by the discount. Crediting the announcement channel for the full order value overstates its contribution and leads straight to over-investing in the next sale's blast.

## Compress the window, do not widen it

The instinct during a sale is to widen attribution windows to "capture everything." Do the opposite. A 48-hour sale should be read on a 48-hour window, with a short pre-period for baseline. Long windows during sales do not add information; they add noise, sweeping in purchases that had nothing to do with the promotion and crediting them to it.

Set the window before the sale starts and write it down. Post-hoc window shopping, trying 7-day, then 14-day, then 30-day until the numbers look good, is how brands lie to themselves with data. The window is a methodological choice, not a dial for optimism.

## Watch the halo, not just the orders

Sales have effects beyond the discounted SKUs. Full-price items get added to carts alongside sale items. Email list growth spikes. New customers acquired during the sale may become repeat buyers. An attribution read that only counts sale-SKU revenue misses the halo, and the halo is often where the real value lives.

Tag new customers acquired during the sale and follow their repeat purchase rate over the next 90 days. A sale that acquires discount-only buyers who never return is a very different event from a sale that acquires future full-price customers, and no same-week attribution model can tell them apart. The cohort view is the only honest one.

## Discount codes as truth serum

During sales, unique discount codes are the most reliable attribution signal you have. A code tied to a channel tells you exactly which touchpoint drove the order, no modeling required. This is one reason to run channel-specific codes during sales even when a sitewide code would be simpler: the measurement value exceeds the operational cost.

Protect the signal. If codes leak to coupon sites, the channel attribution degrades, and you are back to modeling. Monitor code usage by source during the sale and be ready to kill leaked codes. The truth serum only works if it stays in the right hands.

## Post-purchase surveys fail during sales

The usual fallback, asking customers "how did you hear about us," degrades during flash sales. Response rates drop because buyers are in a hurry, and the answers skew: everyone says the channel that delivered the discount, even if they have followed the brand for a year. Survey data during sales measures the announcement, not the journey.

Do not throw the survey out, but do not weight it normally either. Note the sale period in the data and discount its influence on channel-mix decisions. Better yet, compare the sale-period survey mix against the baseline mix; the delta tells you which channels actually moved during the event versus which just got louder.

## The one metric that matters

After all the channel analysis, judge the sale on one number: incremental revenue. Revenue during the sale period minus the baseline revenue you would have earned anyway, minus the margin given up in discounts, minus the cost of fulfilling the spike. Everything else is commentary.

Estimating the baseline is the hard part, and it is worth doing properly: same weekdays from prior weeks, adjusted for trend and seasonality. A sale that "did $500K" against a $400K baseline with $80K in discount cost and $30K in extra support is a $10K decision wearing a $500K costume. Incrementality keeps the costume off. Run the number honestly, write it down, and let it inform the next sale's discount depth before anyone gets excited.

## Frequently asked questions

### Why does normal attribution fail during a flash sale?

Because the models assume ordinary journeys spread over days, and a sale compresses everything into hours. Consideration collapses, future demand is pulled forward, and last click hands the credit to whatever delivered the announcement rather than whatever created the demand.

### Which channels get over credited in a sale?

Email and SMS, typically, because they announce the sale and capture the final click. Much of the revenue they appear to drive was already in the pipeline as existing demand that the discount merely timed.

### What is the honest way to read sale performance?

Incrementality: what revenue happened because of the sale that would not have happened anyway, judged against baseline demand and the weeks after, when pulled forward purchases show up as a dip.

### Should sale results set channel budgets?

No. Sale attribution describes an event, not a channel. Budgets set from sale dashboards fund announcement channels for demand creation they did not do, and starve the prospecting that filled the pipeline the sale emptied.