Webinar: Generative AI for Revenue Growth Management

Generative AI for Revenue Growth Management

GenAI won’t fix RGM until your prediction and optimisation stack is strong enough to trust.

The GenAI halo effect in RGM: impressive demos, weak decision outcomes

A lot of RGM teams are treating generative AI as the missing capability—assuming that if a model can write, summarise, and strategise, it should also be able to “tell us the right price.” Ingo argues that this is the wrong starting point. The risk isn’t under-adoption. It’s building confidence on top of methods that oversimplify how shoppers actually choose.

This session reframes GenAI for RGM as a sequence of layers: clean and integrated data, precise purchase prediction, optimisation to select offers, and only then a chat-style interface that makes the system usable at scale. The point isn’t to add a new tool. It’s to make better commercial decisions more repeatable.

What You'll Learn

Why LLM-style GenAI is the wrong mental model for pricing decisions.
Language models predict the next word from what came before. RGM needs to predict the next purchase from market context—price, promo, assortment, competitors—not just time-series patterns. [12:07]
Why elasticity-based thinking collapses under real category complexity.
Ingo uses simple demand curves to show how quickly “one elasticity number” becomes unstable and unhelpful—especially once you move beyond a single product into substitution across a category. [16:11]
What a practical GenAI stack for RGM actually looks like.
A clear four-layer progression: data preparation and integration, purchase prediction, optimisation to select offers, and a prompt interface as the final usability layer—not the core engine. [23:04]
How “virtual shoppers” change the quality of prediction, not just the speed of analysis.
Instead of assuming a functional demand curve, the approach simulates many shopper types with different preferences and behaviours, then predicts how demand shifts across the whole category when you change prices, promos, or packs. [26:52]
How to use GenAI to make decision-making easier without pretending it can invent the decision.
The most credible use of a chat interface is to translate business questions into constrained optimisation runs—offering options and trade-offs aligned to company rules, terms, and strategy. [32:33]

Meet the Speaker

Ingo Reinhardt

Co-Founder and Managing Director at Buynomics

Before Buynomics, Ingo was a Senior Director with Simon-Kucher & Partners, a global leader in pricing. He holds a Ph.D. in Management from the University of Cologne and Master's degrees in Management and Mathematics. Ingo was a PostDoc at the University of Oxford and published in the Strategic Management Journal.

Session Highlights

GenAI interest is exploding—but most organisations still rate their maturity as low.
The audience poll shows roughly two thirds of participants reporting low or very low GenAI maturity, with almost no one claiming full implementation. [08:21]

RGM’s “next word” is the next purchase—and that changes everything about what the model must learn.
Ingo explains why purchase outcomes depend less on sequence and more on context: price, competitor set, assortment, promotions, and market configuration. [14:48]

Elasticities simplify the math by simplifying reality—and that’s the problem.
A single elasticity number shifts depending on price point and assumptions about demand shape, making it a fragile foundation for GenAI-driven decisions in real categories. [16:20]

GenAI for RGM is a stack: data → prediction → optimisation → interface.
The prompt layer is the last mile. Without the first three layers, a chat tool can sound confident while producing ungrounded recommendations. [23:04]

Virtual shoppers provide a mechanism for substitution, not just a curve fit.
By simulating many shoppers choosing among products, the model can predict where volume goes when you change one item—capturing cross-effects that basic methods miss. [27:16]

The “halo effect” is the commercial risk: capability in language creates false confidence in pricing.
Because GPT can do surprising tasks, people assume it can solve pricing automatically. Ingo argues the responsible approach is tool-using AI: language for interaction, specialised models for prediction and optimisation. [36:26]

Q&A

How mature are most organisations in GenAI for RGM today—really?
The audience poll indicates most teams are still early: around two thirds rate their maturity as low or very low, with almost no one claiming fully implemented, effective capability. Ingo’s view is that this is expected because the field is moving quickly and “fully implemented” is still undefined for many teams. [08:21]

How do you forecast price moves outside historical ranges?
Ingo notes extrapolation is inherently harder than interpolation, but within a reasonable “bubble” around the status quo, models can still perform well. The risk grows when changes become extreme (doubling or tripling), where the historical signal no longer anchors behaviour credibly. [42:39]

Is building a data lake a precondition for GenAI in RGM?
Ingo’s position is that the predictive system must have access to relevant data, but that can be achieved through different architectures. The key requirement is not the label (“data lake”) but whether the data needed for prediction and optimisation is available, integrated, and usable. [43:12]

What’s the main difference between “GenAI with virtual shoppers” and ChatGPT-style GenAI?
LLMs predict the next word from text context, trained on massive language corpora. RGM prediction depends more on market configuration—price, promo, assortment, competitors—and the available commercial data is far smaller than internet-scale text, so you can’t simply apply an LLM to pricing decisions directly. [45:05]

How do you get virtual shoppers to behave like real shoppers across markets and channels?
Ingo explains they are built from the data available in the specific context—country, channel, category—and typically at the level where decisions are actually made. Granularity depends on both data availability and whether teams truly act differently at that level (e.g., regional decisions vs. national ones). [48:02]

Where does the chat interface actually add value, if it can’t “invent” the right price?
The interface becomes valuable when it translates business questions into model-backed options and constraints—e.g., “How do we reach 5% net revenue in the UK?”—and returns trade-offs based on what the company is willing to change (prices only vs. including promo/PPA). It also helps encode company-specific rules, terminology, and institutional knowledge so recommendations stay realistic. [32:33]