Promotion management with AI: how to measure the real ROI of your campaigns

The commercial lead at a consumer goods distributor was happily showing me the results of a promotion: a 20% discount that had cleared the stock in a few days. The promo "worked", he said. But when we closed the month and looked at margin, the picture was different: he had sold a lot, yes, but earned less money than a normal month with no promo. The discount ate the entire volume upside, and on top of that he had sold cheaply what he would have sold at full price anyway.

That is the most expensive misunderstanding in promotion management: confusing movement with result. A promo selling a lot does not mean it paid off. The right question is not "how much did I sell?", but "how much of that would I not have sold anyway without the promo?". And that question almost never gets answered, because it requires separating two things that look identical in the sales spreadsheet.

The two things are incremental demand (genuinely new sales the promo generated) and forward buying (customers who were going to buy anyway and took the chance to stock up cheap). The first is a return; the second is giving away margin. When you do not distinguish them, you end up celebrating campaigns that actually lost you money.

In this article you will see why half of all promotions do not pay off, how a layer of AI on top of your ERP separates real demand from forward buying, a case of a distributor that stopped giving away margin, and the steps for measuring the true ROI of your campaigns.

---

1. The problem: selling a lot is not earning

1.1 The volume mirage

An aggressive promo always sells. Drop the price enough and the stock moves. The problem is that volume is a deceptive metric: it shows activity, not profitability. A campaign can break records in units sold and, at the same time, leave less margin than doing nothing would have.

1.2 Incremental demand versus forward buying

Here is the heart of the matter:

• Incremental demand. Sales that would not exist without the promo: new customers, consumption pulled from a competitor toward you, products that otherwise would not have sold. This is real gain.
• Forward buying. Customers who were going to buy from you anyway and simply took advantage of the discount to stock up. You did not sell more, you sold the same amount cheaper, and on top of that you lose those sales in the following weeks.

A good promo maximizes the first and minimizes the second. A bad promo is almost entirely forward buying.

1.3 The hidden cost of a badly measured promo

The cost of not measuring properly is not only the margin given away in one campaign. It is that, without data, you repeat the promos that "sold a lot" (even if they lost money) and discard the ones that sold less but paid off. Without measuring incrementality, you optimize toward the wrong place, campaign after campaign.

2. What AI does with promotions

2.1 Estimating the baseline

To know whether a promo generated new demand, you first have to know how much you would have sold without it. That baseline is hard to calculate by eye, because it depends on seasonality, trend and each customer's behaviour. The AI layer, working on your ERP data, estimates it, and against that base it measures how much of the promotional sales was genuinely incremental.

2.2 Detecting forward buying

AI also detects the forward-buying pattern: customers who bought heavily during the promo and then disappeared from the following weeks. That post-promo "valley" is the footprint of the pull-forward, and it is key to understanding the campaign's real result, not just the result in the moment.

2.3 Connecting promotion to inventory and profitability

A promo does not live alone. Moving stock at a discount connects with reducing excess inventory when the goal is clearance, and with per-customer profitability analysis, so you do not give away margin to somebody who was already buying at full price. AI crosses all of that so the promo has a clear, measurable objective.

3. A real case: the distributor that stopped giving away margin

3.1 Before

A beverage and food distributor ran promotions constantly, out of habit and commercial pressure. It evaluated them by volume sold, so almost all of them "worked". Nobody measured what happened to margin or to the following weeks' sales. The result was a calendar full of promos and a profitability that never quite grew despite the effort.

3.2 A phased implementation

1. Retrospective measurement (month 1). The previous months' promos were analysed against the estimated baseline, separating incremental demand from forward buying.
2. Design rules (month 2). With those learnings, it was defined which products and customers responded with new demand and which only stocked up cheaply.
3. Targeted campaigns (month 3). Mass promos were replaced with more focused campaigns aimed at generating real incrementality, supported by RFM segmentation.

3.3 After

The uncomfortable discovery was that a significant share of the previous promos — close to half — had been mostly forward buying: margin given away to customers who were going to buy anyway. By cutting those campaigns and focusing the promotional budget on the ones generating real demand, the distributor ran fewer promos and earned more. Margin improved steadily without losing meaningful volume, because they stopped subsidizing sales they already had locked in.

4. Step-by-step implementation

4.1 Define the objective of each promo

Before designing a campaign, answer what it is for: generating new demand, winning customers, clearing inventory, defending an account from a competitor? Each objective is measured differently. A promo with no clear objective is impossible to evaluate.

4.2 Establish the baseline

With AI, estimate how much you would sell without the promo. That number is the yardstick: everything above the baseline is the only thing that counts as the campaign's result.

4.3 Measure the aftermath, not just the campaign window

The classic mistake is closing the measurement when the promo ends. You have to look at the following weeks to detect the forward-buying valley. A promo only pays off if the sum of during and after comes out positive.

4.4 Target, do not broadcast

Mass promos give away margin to whoever does not need it. Aiming the campaign at the right customer and product, using your ERP data, is what separates a profitable promo from one that just moves stock cheaply.

5. ROI and measurable benefits

5.1 What to measure

The key indicators:

• Incrementality (sales above the baseline).
• Net campaign margin (including the effect of the discount).
• Post-promo effect (the drop in the following weeks).
• Campaign ROI (incremental margin over the promotional investment).

5.2 The typical return

The return rarely comes from running more promos: it comes from running fewer and better ones. Cutting forward buying and focusing the budget on real incrementality usually improves margin without sacrificing meaningful volume. It is one of the few levers that improves profitability without touching cost.

5.3 The cultural benefit

There is a change that is hard to measure but important: the team stops celebrating volume and starts discussing margin. When a promo is evaluated by its real ROI rather than by how much it moved, every promotional decision improves.

6. Common mistakes in managing promotions

6.1 Measuring success by volume sold

The root mistake is celebrating the promo that "cleared the stock". Volume shows movement, not profitability. A campaign can break unit records and leave less margin than a normal month. The right question is never how much you sold, but how much of that you would not have sold anyway without the promo. That difference — incrementality — is the only thing that measures whether the campaign paid off.

6.2 Closing the measurement when the promo ends

Looking only at what happened during the campaign is deceptive, because it ignores the valley afterwards. If many customers stocked up cheap, the following weeks fall, and that fall is part of the promo's result. A campaign only pays off if the sum of during and after comes out positive. Measuring only the peak is like celebrating a loan without counting that it has to be repaid.

6.3 Broadcasting mass promos instead of targeting them

A mass promo gives a discount to everybody, including the customer who was going to buy at list price anyway. That is margin thrown away. Aiming the campaign at the customer and product where it genuinely generates new demand — leaning on segmentation — is what separates a profitable promo from one that just moves stock cheaply. Less reach, better aim, more return.

6.4 Repeating the promos "that worked"

If you evaluate by volume, you will repeat the campaigns that sold most, which are often the ones that lost the most margin. That is how the error becomes chronic: you optimize toward the wrong place, season after season. Only by measuring real incrementality can you know which campaigns deserve repeating and which are better retired, however much they "sold a ton".

Ready to know which of your promos actually pay off?

A promotion selling a lot does not mean it makes money. Separating incremental demand from forward buying, with a layer of AI on top of your ERP, is the difference between celebrating volume and building margin. Half your campaigns may be giving away profitability without you noticing.

Want to see how it works in practice? Book a demo and we will show you how to measure the real ROI of your promotions from your own data.