AI in RGM: Where it adds value (and where it doesn’t)
How CPG teams can apply AI in RGM: use-case fit, architecture, and data readiness.
Across the first half of the year, a series of events and executive conversations in the CPG ecosystem kept circling back to the same theme: AI is increasingly part of the RGM agenda, but the real question is still unresolved—where does it genuinely add value, and where doesn’t it? This article captures the reflections that surfaced most consistently in discussions with CPG leaders, technology partners, and revenue management experts.
The takeaway was practical rather than provocative: AI delivers in RGM only when the use case is right—and only when the foundation is ready. What follows is a grounded view of best-fit use cases, real-world limits, the architecture emerging for RGM-grade AI, and why data quality still matters most.
This article builds on the series’ starting point. For the broader context on foundations, adoption, and the five-year RGM transformation journey, read the first contribution here.
Key Takeaways:
AI value depends on the use case: it shines in scenario-heavy, pattern-rich decisions.
AI extrapolates well, but struggles with novelty: it can’t predict what data it hasn’t seen.
Data quality is the bottleneck: start by using AI to detect and reconcile issues.
What can AI truly do in RGM for CPG companies?
A provocative question emerged during the discussions: Do we really need AI in RGM?
The answer was nuanced: it depends entirely on the use case.
The first example involved a seasonal category where the solution was psychological rather than algorithmic, and an optimization tool predicted the opposite outcome.
Why? Because the tool optimized for rational behavior, while consumers experience effort costs, loss aversion, choice paralysis, and framing effects.
Here’s the crucial insight: AI is excellent at extrapolation. It’s terrible at predicting genuinely novel behavior that hasn’t appeared in historical data.
The second example was portfolio optimization—well suited for AI to run thousands of scenarios and identify non-linear relationships.
Even then, qualitative insights were incorporated. The tool was used as a benchmark, not as an oracle.
This distinction matters: AI creates value in RGM by augmenting human judgment.
So, where AI actually adds value?
According to research cited during the discussions, companies optimizing trade investments and commercial decisions holistically—not in silos—see measurable P&L uplift compared to competitors.
> A line that stood out
Companies optimizing trade investments and commercial decisions beyond isolated analytics like promo or pricing separately are rewarded already today with a financial uplift in their P&L. There’s something between 2 to 4 percentage points of uplift versus competitors not being able to do that.
The architecture that’s emerging
Rather than bolting AI onto existing systems, leading approaches are building integrated platforms that address real bottlenecks.
At the foundation is the data layer—connecting across TPM systems, ERPs, and data sources, even when they differ across regions.
Above that sits augmented data management: cleaning data, aligning sell-in with sell-out, detecting promotions missing from planning systems, allocating promo costs.
Next come foundational functions like trade promotion management and retail execution.
Above that sits the analytics layer, moving beyond traditional regression models to neural networks.
Purpose-built RGM technology adds an intelligence hub with encoded business knowledge.
On top sits an agentic framework—AI assistants that analyze questions, generate recommendations, and eventually take actions within parameters.
The final layer is visualization and synchronization across functions.
What makes this different from a generic AI chatbot is RGM-specific knowledge and business context.
> A line that stood out
The revenue management assistant is giving an answer not just as text but visualized. And it’s visualized with the same visuals you would get if you went in the tool. You click, you drag, you drop, you make your selection.
The data quality imperative
None of this works without strong data quality at the core.
Data quality problems are often invisible until scaling across markets, retailers, and categories.
A practical conclusion is to use AI to solve the data quality problem: automated anomaly detection, intelligent reconciliation, and pattern recognition for definition drift.
> A line that stood out
We’re using hundreds of models at the regional level. Every month, they degrade. We have to constantly retrain. The cost is enormous. And I’m not even sure the accuracy justifies it.
Conclusion: from experimentation to impact
AI in RGM isn’t a magic upgrade. It’s a capability that works only in the right context.
The use case comes first. AI can extrapolate from patterns and accelerate scenario-heavy decisions—but it struggles when behavior is genuinely new, or when the context has shifted.
The foundation still decides the outcome. Without reliable data, AI doesn’t create clarity—it amplifies noise.
What works in practice is a disciplined approach: build the data layer, strengthen data management, add business context, and use AI to augment judgment—not replace it. Start where the bottleneck is most visible: data quality. That’s where AI can help first—and where scalable value begins.
Closing thoughts
Start practical. Scale only when the foundation holds.
AI in RGM is most effective when the use case fits the nature of the decision and the foundation is strong.
Start where AI can create immediate value—often in data quality and reconciliation—then expand toward more advanced decision support.