AI won’t fix retail pricing without decision intelligence

It’s Monday morning on a grocer’s pricing team...

The team pulls weekend sales from one system, the latest comp shop file from another, and cost letters from a shared inbox. The numbers disagree, so the first hours of the week go to reconciling them and debating which version of the truth to trust. By the time the price batch is ready for the tag run, a competitor’s ad has landed and the market has moved again.

Over nearly two decades in retail technology, I’ve watched some version of that Monday play out at more retailers than I can count. Between legacy tools and myriad spreadsheets, too much of the busywork falls on people who should be setting strategy. That leaves every pricing team weighing the same question: how do we apply AI? The real competitive advantage in retail pricing isn’t simply having better AI models. It’s what your organization learns from every pricing decision: what worked, what didn’t, and why. Over time, that institutional intelligence becomes something competitors can’t easily replicate.

Leadership’s answer is usually to buy AI, and I understand why. AI can accelerate pricing workflows and improve decision-making, but the quality of recommendations still depends on reliable data, validated demand models, and sound optimization science. But if you stack AI on a shaky data foundation, it only accelerates poor decisions.

That Monday-morning team needs a unified version of the truth before it needs AI. AI often gets sold as a magic bullet, but mature price optimization pays off in stages, through incremental tweaks and course corrections. Retailers should assess their maturity across data, rules, forecasting, and optimization, then prioritize the gaps that most constrain business outcomes. Progress is not always linear, and different pricing functions may mature at different rates.

Is your data ready for the AI you’re buying?

The AI money is already committed. In NVIDIA’s latest survey of retail and CPG companies, nine in 10 said they’re increasing their AI budgets. As 2027 plans take shape, that sometimes means a mandate reaches the pricing team before anyone agrees on what AI should do.

Organizations often invest heavily in AI while still struggling with inconsistent data, disconnected processes, and teams that spend more time validating information than acting on it. McKinsey’s State of Grocery North America 2026 makes a similar point: AI pilots generate excitement but often fail to change how decisions get made.

The bottleneck usually sits underneath the technology, in the data and the pricing rules. Meanwhile, the market promises AI agents that will make sense of all your data and do the work for you. Sounds great, right? Those promises are getting more real with time, but they’ll only pay off for teams that built their foundation first.

So before you turn AI loose on your data, take an honest look at where your pricing operation stands. Three questions tell you most of what you need to know:

• Is your competitive intelligence reliable, current, and matched to your own items, including private label and fresh?
• Do your pricing rules work together, so a rule for your Key Value Items (KVIs) never undercuts a price family or a margin floor?
• Do you have pricing science you trust to forecast what a price or promotion will do?

Whichever question you can’t answer with confidence is your starting point.

There’s also the question of whether to build that science yourself. General-purpose AI can help automate workflows and interpret information, but it is not a substitute for retail-specific demand modeling, elasticity estimation, and constrained price optimization. Price optimization specialists have already spent years, in some cases decades, refining the optimization engine. I’ll own my bias, since that’s the business I’m in. Still, your best people belong on strategy, and rebuilding that engine in-house pulls them away from it.

At ClearDemand, we’ve built a simple pricing maturity model around the four stages we see in retailers, each defined by the question it can answer: Foundational (what happened?), Structured (did we follow our rules?), Predictive (what will happen?), and Connected (what is the best action, with the guesswork removed?). I think of the three moves between them as crawl, walk, and run.

What does Structured look like?

The move from Foundational to Structured is about visibility: one current view of your price position against the competitors that matter, built on competitive data clean enough to support rules your team trusts.

With that view in place, rules can do their job:

• KVIs held to a defined competitive position
• Price families that keep the 12-ounce and the 24-ounce in step
• Margin floors, zone strategy, and ending-digit conventions applied the same way every cycle

Here, AI mostly works in the plumbing: matching products, flagging anomalies, and cleaning data so the Monday ambiguity shrinks.

Rules have a limit, though, and they can go out of spec quickly and dramatically. One regional chain we work with anchored a high-priority rule to the priciest competitor in its market. When that competitor became one of the cheapest, the rule inverted the chain’s price position for years before a single rules review caught it.

Rules tell you what’s allowed, but not what’s likely to work best.

When does the system start learning?

The move from Structured to Predictive is about forecasting. You keep your guardrails and add elasticity underneath them, so the team sees what a price move will do to units and margin before it reaches the shelf. Promotions get the same discipline: ads are planned against base price, and each temporary price reduction (TPR) is judged on its incremental lift.

This is where the loop closes, because the system can now learn from its own work. Historical model validation helps assess forecast accuracy, while outcome measurement evaluates the results of pricing decisions. Estimating incremental impact requires a credible baseline or counterfactual to distinguish the effect of the decision from other market factors.

One grocer we work with measured its promotion history and found that roughly one in three promotions lost money, so it retired them all.

At this stage, your pricing team can give merchants and category managers clear answers and prove its value to the wider organization.

The real measure of pricing maturity isn’t the sophistication of the model. It’s whether retailers can consistently translate recommendations into measurable improvements in margin, revenue, price competitiveness, and merchant productivity. Forecast accuracy matters, but it is a means to better business decisions, not the business outcome itself.

What does Connected look like?

The Connected stage brings competitive intelligence, regular pricing, and promotions together through a shared data foundation and coordinated demand intelligence, enabling consistent, explainable recommendations and measurable business outcomes.

That’s where vertical, agentic AI levels up the whole organization, because it finally has a clean, measured history to reason over.

Despite the creative marketing and buzzwords, the whole industry, ClearDemand included, is still building toward this stage. Every team we’re helping get there started with the foundation.

Why does every measured decision improve the next one?

I call this compounding intelligence: a continuous feedback loop where pricing and promotion decisions are measured, evaluated, and used to refine future models and recommendations.

A measured price change teaches the model more about elasticity in that category, zone, and season. Measured promotions show what a TPR did to base volume and the rest of the basket, including the items it quietly cannibalized. Capturing merchant overrides and their rationale adds valuable business context, helping teams understand where model recommendations align with or diverge from operational realities.

Anyone can license the same AI models. Your decision history belongs to you, and it grows every week you measure. That’s the pricing operation I want to help every retailer build, where each decision leaves the organization a little smarter than before.

Where should your team start?

Roughly 70% of the grocery and c-store operators who come to us begin in Foundational or Structured, so if that’s your team, you’re in good company and in the right place. If you want a read on where you stand heading into 2027, our self-guided pricing maturity assessment takes about four minutes.

Let’s go back to that Monday morning, once the team is fully Connected. Competitive data is continuously matched and validated, pricing rules are monitored for exceptions, and forecast-driven recommendations help merchants evaluate the potential impact of price changes before execution. The price batch reaches the tag run before the competitor’s ad lands, and the team spends the morning making decisions. That’s when AI finally has something worth accelerating.

So, what did your pricing team learn from last week’s price changes? And which stage is your team in today?

Get in touch with us and let’s brainstorm.

Frequently asked questions

What is AI price optimization?

AI price optimization uses machine learning to forecast how units, revenue, and margin respond to a price change, then recommends the strongest price that still respects a retailer’s rules. In grocery, those rules include KVI positioning, price families, margin floors, and zone strategy.

Do grocers need perfect data before using AI in pricing?

The bar is consistent, trusted data, and the first move in pricing maturity, from Foundational to Structured, builds it: competitive data matched to your items, rules applied the same way every cycle, and one current view of price position. Most grocers have more usable data than they expect.

How long does it take to see value from price optimization?

On ClearDemand’s platform, first competitive intelligence value arrives in 30 days, and competitive intelligence plus rules-based pricing in 45. Forecasting and measured results build from there as the team moves toward Predictive.