How AI investigates the rules behind a price and tests what should change
A pricing engine can follow every rule and still produce a price that surprises the team that set them. The engine applies the rules consistently. The combination is not obvious. Priorities, exceptions, and relationships across products interact, and a request as simple as making six-packs more competitive can require someone to reconstruct all of them before a change is safe.
That reconstruction is the expensive part of the work. It shows up at implementation and again whenever the strategy moves. We put agents on it. They read the live configuration and the data, test explanations for what they find, and hand a specialist a finding or a proposed change. The specialist decides what is commercially true.
Three facts have to line up before anyone should change a line of logic:
- What the retailer meant.
- What the rules actually do.
- What prices appear.
Treat those as one fact and a sensible request becomes a surprise on the shelf.
The beverage aisle
Illustrative example
This aisle is built on three relationships. Larger packs should offer better value per can. Equivalent packs from competing national brands should sit at parity. Own-brand should hold a gap to the national brands. Some diet variants ride along with those relationships.
None of it holds unconditionally. Parity may apply only to matched packs. A promotion may override the pack ladder. A margin floor may override the promotion. Some rules cover the category; others cover selected products, or apply only when particular data is available.
Now the team is asked to make six-packs more competitive, because the six-pack is the pack shoppers use to judge whether the store is expensive. The request sounds like one change. It leaves several decisions unstated:
- Which rules will apply to those six-packs?
- Which rule wins when two disagree?
- If the six-pack moves, what else has to move to preserve a relationship, and what should be allowed to break?
A lower target can be overridden by a rule that runs later. Preserving parity can require a linked pack to move too. Scope is a separate decision from priority: a promotional exception intended for selected six-packs should not silently become the rule for every pack in the category.
An agent can trace those conditions across customer-specific logic and surface the decisions a proposed change requires.
Some of what you need is not in the configuration. A retailer that regularly transacts below its displayed price needs a different reading of the data from one that never discounts. That has to be established with the business. Agents need it for the same reason a new specialist does.
The report that did not hold
One of our internal diagnostic reports said that very few recommended prices were reaching a retailer’s shelf. If that had been true it was a serious failure, so we asked a challenge agent to test what could make the claim false.
The measurement was counting repeated sends of the same recommendation as separate failures. The agent counted distinct price requests instead and gave the shelf a longer window to catch up. Counted that way, the original claim did not hold.
The agent did not stop on the better number. It asked whether the shop was also reaching nearby prices we had never recommended, which would make the matches less convincing, and whether apparent matches were simply returns to prices the shop had held before. It chose those two checks after seeing its results. Nobody specified them as the next step.
It withdrew the disconnection claim and rewrote the report. The corrected report went to specialist review. Nothing in the retailer’s configuration changed.
Two seats
The same investigative pattern shows up in two seats, serving delivery specialists and the retailer’s pricing team.
- Delivery. Internal agents work for the Quicklizard specialists who implement and expand a customer’s use of the platform. QL Investigator and the diagnostic agents sit here. Investigator stays internal.
- The retailer’s team. Product agents work for the customer’s own pricing team. Strategy Challenger and pricing explanations sit here.
Neither seat is there to change a price on its own. Both are there to make the three facts above visible to the person who will.
Investigator
Solutions architects and customer success managers point Investigator at an account. It reads the configuration, the data, the model outputs, and our notes on how the platform behaves. They need to know whether the customer’s existing logic will support a new capability, and what has to change before it is worth trying.
Investigator helps assess readiness and recommend pilots. It also generates and tests draft pricing configurations. A specialist checks the proposal, takes commercial questions back to the customer, and implements the approved configuration. An exception buried in the existing rules may matter more to that decision than a promising model output.
For existing Quicklizard customers adopting new AI capabilities, agent-assisted delivery and process redesign have shortened implementation cycles from months to weeks.
A pilot in preparation
While preparing the next phase of an AI pricing pilot for a consumer-durables business, our team used agents to investigate which products were suitable, then modeled how those candidates would be priced under the existing configuration, using the model outputs available to it.
Some products passed the initial checks and would have decreased in price where the pilot intended increases. Others exposed questions about which rules took priority, or how price limits should apply. Eligibility had not established whether the live logic would do what the pilot needed.
The findings changed the proposed product selection. They also split the remaining work in two: adjustments available through configuration, and a requirement that needs development. The next phase can be scoped around that difference.
Pilot measurement has the same shape of problem. Three different numbers get called the price:
- The recommendation the engine generated.
- The price that was applied.
- The price a shopper paid.
Report the first as if it were the third and the pilot will look decisive and prove nothing. Those distinctions have to be named before anyone treats the result as adoption or as a commercial outcome.
Strategy Challenger
A product assigned to generate profit is following a competitor’s price. The pricing manager needs to know why. If a margin floor keeps determining the final price, changing the strategy may accomplish little until someone looks at that constraint.
Strategy Challenger is the product seat for that investigation. It reads how rules, guardrails, and commercial objectives interact, and it proposes improvements for review. The manager decides whether the current behavior still serves the product’s role.
| Part of the review | What the agent examines or proposes | What the pricing team decides |
|---|---|---|
| Compare behavior with intent | Where pricing logic appears to conflict with commercial objectives | Whether the stated product roles and priorities are still correct |
| Examine constraints | Where guardrails repeatedly determine or constrain recommendations | Whether a limit is protecting the strategy or preventing it from taking effect |
| Review a proposed change | Improvements to the interactions between rules, guardrails, and objectives | Whether to accept, revise, or reject the proposal before implementation |
The configuration does not hold the retailer’s whole context. A competitor-matching strategy on profit generators may be a turf war the retailer chose to fight on house brands. A strategy that looks dormant may be waiting for a seasonal calendar. The agent surfaces the mismatch and explains its reasoning; the pricing team knows which mismatches are on purpose.
Pricing explanations answer the everyday question of why one product received its price. They connect the relevant history, strategy, and evidence in plain language. Back in the beverage aisle, that means seeing how a pack relationship, a promotional exception, or a margin floor influenced a particular six-pack, and deciding whether it warrants a closer look.
What to ask
The confirmed intent, the account of what the logic was doing, and the reason a proposal was accepted or turned down are worth keeping. That record gives the next category, market, or capability a starting point. Agents can read it back. So can people.
Four questions are useful when someone offers agents for this work:
- How will you establish what our pricing logic is meant to achieve, beyond describing what it already does?
- When we ask why a product got its price, what evidence will we see about the rules and how they interacted?
- Who on my team approves a proposal, how long does one take them, and what happens to one nobody gets to for two weeks?
- When we add a category, a market, or a capability, what does the second investigation get to reuse?
The six-pack still needs a commercial decision. See which other packs would move before you make it.