What AI actually changes in revenue growth management

There is no shortage of claims about what artificial intelligence will do for revenue growth management. Most of them describe a destination without describing the road, which makes them difficult to act on and easy to dismiss.

The published evidence supports a narrower and more useful claim. McKinsey has reported consistently that AI-enabled RGM programmes deliver in the range of 2 to 5 per cent net revenue uplift and 1 to 3 percentage points of gross margin improvement when scaled across a portfolio. Those are meaningful numbers on a large revenue base. They are not transformational in the way the vocabulary around them suggests, and the range is wide because outcomes depend heavily on data maturity and commercial adoption.

What it genuinely changes

Three things, none of them glamorous, account for most of the value.

Coverage

A team of three analysts can rigorously evaluate perhaps the top twenty promotional events across the largest two customers within a planning window. Everything else gets a lighter treatment or gets repeated from last year. The constraint is not skill, it is hours. Automated analysis removes the hours constraint, so every event in every customer carries a proper evaluation rather than the largest handful.

This matters most where the data is thinnest. Smaller markets and smaller customers rarely justify a dedicated analyst, which is precisely why their pricing and promotional decisions receive the least scrutiny, and why the value at stake per hour of analysis is often highest there.

Frequency

Elasticity refreshed once a year, market mix modelling refreshed once a year, and a pricing review once a year, all feeding decisions made every week. That cadence mismatch is normal and it is expensive. Rebuilding the analysis continuously rather than annually closes the gap between when the market moves and when the model notices.

Joint estimation

The most underrated change is that the levers can be estimated together rather than in separate tools. Base elasticity separated from promotional response, baselines built with competitor activity in them, architecture measured on realised price. These are not new techniques. They were simply too laborious to run at portfolio scale by hand, so most organisations approximated them.

What it does not change

Three things are worth being clear about, because overstating them is how programmes lose credibility internally.

The negotiation. A model can size what is at stake in a terms discussion. It has no view on what a specific buyer will accept in a specific quarter, what the relationship can absorb, or which concession buys goodwill that pays back somewhere else. That judgement is commercial and it is human.

The data problem. Analysis cannot compensate for sources that do not join. If sell-out, panel, EPOS and trade promotion data sit on incompatible product hierarchies, no amount of modelling reconciles them. The unglamorous work of building a common spine remains a prerequisite, and it is usually the longest part of any programme.

Accountability. A recommendation that nobody owns does not get implemented. Someone senior has to interrogate the output, decide whether to believe it, and put their name to it in front of a commercial director. That has not changed and should not.

Why most programmes stall

The barriers reported across the industry are organisational rather than technical. Data fragmentation, alignment between functions, and the gap between analytics teams and the commercial decision makers who are supposed to act on the output.

That last gap deserves attention. A model that produces a technically correct answer in a form that an account manager cannot take into a meeting has not created any value. The same research that finds 53 per cent of retailers open to new promotional tactics also finds that the constraint is sales adoption rather than analytics, and identifies financial fluency and negotiation capability as what separates organisations that convert analysis into outcomes from those that do not.

A reasonable way to start

The most common failure is scope. A programme framed as an enterprise transformation acquires a steering committee, a two year roadmap and a budget line that becomes a target during the next cost review.

A narrower start tends to survive better. One category, one retailer or one market. One recurring analytical task that the team already performs and dislikes. Automate that task end to end, run it through a live planning cycle, and compare what the team did with what they would have done. If it holds, extend it. If it does not, the loss is a few weeks rather than a programme.

With 62 per cent of organisations yet to apply generative AI to trade or revenue growth management at all, the realistic prize for moving early is not exotic. It is slightly better baselines, faster post-event analysis, and consistently better funding decisions, compounding quarter after quarter.