Audiences AIThe Discount You Cannot Aim
The standard advice on discounting is sound. Work out which customers actually needed the incentive and stop paying the ones who were coming anyway. Hold out a control group, measure the incremental sale, and keep tightening eligibility until the promotion budget stops leaking. For a coupon or a win-back offer, that discipline is exactly right, and we would give it to any client who asked.
It rests on one assumption: that you choose who sees the offer.
A markdown breaks that assumption. A storewide sale has no audience settings. Neither does the marketplace event you are enrolled in, or the quick-commerce listing that price-matches whatever is live. Every visitor sees the struck-through price: the loyal customer three weeks from her next full-price order and the bargain hunter who was never coming otherwise. Exposure is total by design, so cohort gating, the move that rescues a targeted offer, has nothing to hold onto.
What remains is depth. And depth does something the P&L never books.
Research on wait-or-buy behaviour finds that consumers "are less patient for small markdowns and more patient for large ones" (Baucells, Osadchiy and Ovchinnikov, 2016, Operations Research). Read that as an operating statement. The depth you become known for sets how long your customers are willing to wait for you. A 60 percent event teaches patience, and the invoice for the lesson arrives a season later, when full-price weeks run soft and everyone in the room calls it demand.
There is a second cost inside the same move. A survey of discounting in Indian e-commerce found 48.5% of respondents saying discounts of 50 percent or more left them concerned about product quality (Gaikwad et al., April 2026, IJIRT; a small open-access study, so hold the precision lightly). The direction is the finding: the markdown that moved the unit also argued against the product.
Why does nobody catch this? Because each channel runs its own calendar. The site team plans its sale days. The marketplace team enrolls in the events its account managers pitch. The quick-commerce listing follows whichever price is live. Each calendar looks disciplined on its own. The customer experiences the union of all of them. Suppose the site is discounted 40 days a year, marketplace events add another 60, and price-matching adds 30 with some overlap. A customer with three apps on her phone can now find a lower price from you for roughly a third of the year. Those figures are illustrative. The structure that produces them is real: nobody in the building owns the union, so nobody has ever counted it.
So the working question shifts from who should get the discount to how much of the price surface is open at all. Four moves, in order.
Count the exposure days
Add up the days in a year that a discounted price of yours is visible anywhere: own site, app, marketplace listings, quick commerce. This is a counting exercise, built from promotion calendars and a weekly crawl of your own listings, and it needs no model. The number is the honest denominator for every markdown conversation that follows, because a 25 percent event means something different at 30 exposure days than at 150. Across the pricing work we have done, we have yet to meet a brand that had already produced it.
Read the gap around the sale window
Inter-purchase gaps carry the second signal. When gaps stretch as a known sale window approaches and snap shut once it opens, customers are timing you. That is trained patience, and it produces the same topline as fading demand: soft weeks and a sell-through chart that drifts. The two conditions call for opposite responses. Fading demand argues for a deeper cut. Trained patience argues that the deeper cut is the cause. A team reading only the weekly revenue line will pick wrong, because the distinguishing evidence sits in the spacing of individual customers' orders, and that spacing is rarely on anyone's dashboard.
Gate the surfaces you own
Site, app, email and logged-in pricing are the surfaces where you still pick the audience. The eligibility playbook from the top of this piece applies there, and only there. So the sequence matters. Pull depth off the open surfaces first and let the owned ones carry it: a discount shown to a logged-in lapsed customer is a targeted offer, while the same discount on an open listing is an exposure day for everyone. On owned surfaces, depth becomes a targeting decision again.
Move depth behind membership
A closed structure takes this a step further: the lower price becomes a benefit of a paid membership the customer chose to join. The same IJIRT survey puts repurchase intent at the top of its scale for 46.2% of Lenskart's Gold MAX members against 17.1% of non-members, and reports that members held their preference even when competitors discounted deeper. One honest limit from our own engagements: we have not seen an Indian brand grow full-price volume by removing discounts alone. Every case that held demand paired the discipline with a closed structure of some kind. Treat marketplace sale events the same way, as an exposure commitment you renew annually, and price the renewal before you sign it.
The markdown gets approved in a margin meeting, with margin arithmetic on the screen. Its behaviour is temporal. It moves demand across weeks and books its cost in a quarter where nobody connects the effect back to the decision. That is also why the diagnostic can start this week, from calendars you already have, before any modelling begins.
How many days a year can a customer see a discounted price from you, anywhere?
Interpret reads exposure days and inter-purchase gaps out of the transaction data you already hold. See Interpret.