Rule debt: What Happens to a Pricing Engine After 18 Months
The accumulation of rules is a cumulative process, not an isolated incident: it grows with every legitimate decision made in isolation, even though no single decision is open to criticism. Five forms predominate: the exception that has become permanent; the seasonal rule that is never suspended; rules that contradict one another; the orphaned rule whose author has left; and the redundant rule.
The cost isn't what you might think: before it eats into profit margins, debt erodes trust. A team that no longer understands why the system suggests a certain price will stop following it. The solution is a consistent routine rather than a major cleanup effort every three years: a quarterly portfolio review, four questions per rule, and an expiration date set at the time of creation.
A pricing engine that has just been deployed is a well-defined system. Three hundred rules written by people who understand why they exist—documented and consistent with one another. Eighteen months later, there are eight hundred of them. No one knows where two-thirds of them came from; some contradict each other; and a handful produce recommendations that no one dares to implement anymore. The engine hasn’t broken down—it’s become clogged. This problem has a name, a mechanism, and a remedy that isn’t a project but a ritual.
What Is a "Rule Debt," and Why It's Invisible
A pricing engine operates based on a set of rules: minimum margin requirements, price differentials to maintain relative to competitors, product line consistency, rounding rules, and exceptions by supplier, region, and store format. This set of rules is not a static configuration. It evolves, because retail is constantly changing.
The "debt of rules" is the gap between the current set of rules as they stand and what they would be like if we were to rewrite them today, with the benefit of hindsight. It does not consist of mistakes. It consists of decisions that were correct at the time they were made, but are no longer so.
Where do the rules come from that we no longer dare to challenge?
It’s always the same story. A buyer obtains an exemption for a specific transaction. A regional office requests special treatment during construction work. A supplier negotiates a minimum price for its product line for a semester. A team circumvents an unintended consequence by adding a constraint rather than correcting the one that caused it.
Each of these decisions is justifiable. None of them is documented as temporary. And when reason disappears, no one is responsible for repealing the rule, because no one was ever tasked with repealing it.
Why No One Sees Her Passing By
Because a rule violation doesn’t cause an incident. It doesn’t trigger any alerts, doesn’t block any processing, and doesn’t appear on any dashboard. It manifests itself through subtle signs that are initially attributed to something else: a slowly rising rate of rejected recommendations, increasingly lengthy committee deliberations, or a phrase that keeps coming up in meetings: “Anyway, we’ll just have to fall behind on this category.”
This last symptom is the most serious. It means that the team has lost confidence in the system and has set up a parallel management system—often in a spreadsheet—exactly the kind of system the tool was supposed to replace.
Maintenance currently accounts for 61% of IT budgets, and the executives surveyed plan to reduce this share to 27% by 2030. Maintaining existing infrastructure—not building new systems—is the top expense category.
(Cognizant, survey of 1,000 executives and technology leaders from the world’s 2,000 largest companies, November 2025)
This figure does not pertain to pricing; it pertains to information systems in general, and that is precisely what makes it so telling: the mechanics are the same everywhere. What has already been built requires more attention than what remains to be built, and this proportion continues to grow as long as no one actively works to counter it.
The Five Forms of Liabilities
1. The Exception That Became the Norm
An exemption granted for a three-week operation that is still in effect three years later. This is the most common and easiest type to address, because all it takes is making it visible: in most parks, a search for exemptions with no end date turns up several dozen.
2. The seasonal rule has never been disabled
The constraint imposed for the start of the school year, the holidays, or a weather spike—which continues to apply even out of season. It’s more insidious than the previous one because it only takes effect during part of the year: we don’t realize it until it’s causing problems—in other words, too late.
3. Rules That Contradict Each Other
Two legitimate rules, when considered separately, are incompatible for a subset of SKUs. A maximum price difference constraint relative to a competitor and a minimum margin requirement may well be mutually exclusive for an SKU whose purchase price has risen. The system then makes a decision based on an order of priority that no one consciously established, and the result is surprising.
This is the most expensive approach because it produces recommendations that are difficult to understand. And a recommendation that is difficult to understand is a recommendation that gets rejected, as explained in our article on the explainability of price recommendations.
4. The Orphan Rule
The one whose author has left the company or changed positions. The author’s intent is not documented anywhere. No one dares to delete it because no one can guarantee that it doesn’t protect something important. So, as a precaution, it remains in place indefinitely.
5. The redundant rule
Two rules that produce the same effect in two different ways, often written a few months apart by two people who didn't know each other. It doesn't cause any direct harm, but it doubles the work required for any future changes: you fix one, the other continues to take effect, and you conclude that the engine is unpredictable.
The other 5 parts of the series “Managing an AI Pricing System”
The True Cost of Debt
The Cost of Trust: It Comes First
Before it eats into profit margins, an unorganized inventory costs the store customer loyalty. The mechanism is simple: a team follows a recommendation as long as it can justify it to a buyer or store manager. As soon as it can no longer do so, it stops. And it doesn’t just stop following the specific rule in question—it stops applying it to the entire category, and then to the entire store, as a precaution.
That is why rule debt is a governance issue rather than a technical one. It undermines the system's usability.
The cost of margin, which comes next
It lies in the contradictions. When two rules conflict, the system applies the more restrictive one—which is almost always the more conservative option, and therefore the least profitable. A price floor that’s been overlooked on a product line where purchasing terms have improved means profit left on the table with every sale—silently—for months on end.
The Cost of Inaction: The Heaviest Burden
A system that no one understands anymore becomes a system that no one touches. Every change requested by the business is met with the same response: we don’t know what it might break. The organization then stops adjusting its pricing policy—not because it doesn’t want to, but because its system has become too opaque for it to dare to do so.
79% of executives believe their organization will not be able to reduce its technology debt by more than half by 2030. A liability that is allowed to grow cannot be resolved with a single decision—it must be managed gradually.
(Cognizant, survey of 1,000 executives and technology leaders from the world’s 2,000 largest companies, November 2025)
The fleet review: a ritual, not a project
The natural temptation is to do a major cleanup: you identify the problem, launch a complete overhaul, and spend three months on it. It works once, but then the system gets cluttered again, because nothing has changed in the way rules are created and discarded. The only approach that works in the long run is a regular, short-term cycle.
The Rhythm
- Every quarter: review the rules created during that period, any exceptions that have expired, and any conflicts that have been detected. A half-day is sufficient if this process is already in place.
- Every year: a comprehensive review of the fleet, category by category, with the relevant buyers. This is the only time we revisit the old rules.
- Whenever there is an organizational change—whether it’s the departure of a category manager, the merger of two business units, or a change in sales policy—these are the moments when the rules are left hanging.
The Four Questions to Ask About Each Rule
- Why does it exist? If no one can answer that in a single sentence, the rule should be considered for removal, not retained as a precaution.
- Is this reason still valid? Do the supplier, region, season, or competitor in question still exist in the same form?
- Does it actually work? A rule that hasn't been triggered in twelve months doesn't protect anything. The trigger counter is the most useful tool in this review.
- Does it conflict with another one? That's the question that requires a tool, because it can't be detected by eye when there are several hundred rules.
Who decides to remove
The sticking point is rarely technical; it’s political: no one wants to bear the responsibility for having removed the rule that protected something. The answer boils down to one principle:the decision to remove a rule is made collectively, within the same body that created it. If a rule was created by a committee, it must be repealed by a committee. If it was created through an individual ruling, it means the process for creating it was too lax, and it is that process that needs to be corrected. This point is directly related to the governance of exceptions and exemptions, which deals with the right to grant exemptions, whereas this article addresses the consequences that such exemptions leave in their wake.
Prevent Debt from Building Up Again
The expiration date, set at the time of creation
This is the most effective and simplest measure: every new rule is established with an expiration date. If it is to be permanent, this is stated explicitly, along with the reason why. The effect is automatic: instead of having to justify repealing a rule, one must justify extending it. The burden of proof shifts, and that is all that matters.
The author and the intent, not just the parameter
A rule without a written statement becomes orphaned as soon as its author changes jobs. Two lines are all it takes: who, when, why. That’s what will make it possible, eighteen months later, to make a decision in thirty seconds instead of renewing it out of an abundance of caution.
A park you can read
A useful pricing engine should be able to answer three questions without requiring any specific development: how many rules are active within this scope, which ones have never been triggered, and which ones conflict with one another. If your solution can’t answer these questions, the portfolio review will remain theoretical, because it will require manual work that no one will be able to sustain over the long term. This is a selection criterion in its own right, just like the ones listed in our guide to essential pricing software features.
A backlog of rules is therefore not a sign that a system is flawed. It is a sign that it is being used. A backlog that doesn’t change is either a dead backlog or one that no one is looking at. That is why the backlog review is planned from the scoping phase onward, just like the other rituals in the “run” phase of a pricing tool deployment project. The issue isn’t about avoiding liabilities; it’s about deciding how often to settle them.
FAQ
It is the gap between the current set of rules and the set that would be written today with the benefit of hindsight. It does not consist of errors, but rather of decisions that were correct at the time they were made: exceptions granted for a one-time operation, seasonal restrictions that were never lifted, and trade-offs made by someone who has since changed positions.
It is invisible because it does not trigger any incidents. It can be identified by subtle signs: an increase in the rate of rejected recommendations, increasingly lengthy decision-making processes, and a gradual return to parallel management in a spreadsheet.
Three rhythms complement each other. A brief quarterly review, limited to rules created during that period, expired exceptions, and detected conflicts: half a day is enough once the process is established. A comprehensive annual review, category by category, with the relevant buyers—the only time we revisit old rules.
And a review triggered by organizational changes: the departure of a category manager, the merging of business units, or a change in sales policy. These are the moments when the rules are left without a home.
Manual detection is not feasible once the number of rules exceeds a few dozen: a conflict arises only for a subset of SKUs, under specific conditions. The solution must therefore be able to simulate how the product portfolio applies to the entire catalog and flag instances where multiple rules are mutually exclusive.
Two complementary metrics make this work possible: the rule-trigger count, which identifies rules that no longer provide any protection, and the log of the rule applied to each recommendation, which allows you to trace an unexpected price back to its cause.
Yes, provided it is done within a specific framework. A rule whose purpose is not documented anywhere and that has not been triggered in the past twelve months does not protect anything identifiable: maintaining it as a precaution is like paying indefinitely for insurance whose purpose is unknown.
The framework is based on two principles. The decision to remove an item is made by the same body that approved its creation, so that no single person bears sole responsibility. And the removal is simulated before being implemented, in order to assess its impact on the catalog before it is actually removed from the shelves.
The most effective measure is to set a mandatory end date at the time of creation. Every new rule is established with an expiration date; if it is to be permanent, this must be explicitly stated along with the justification. The effect is automatic: from now on, one must justify an extension rather than a repeal, and the burden of proof shifts.
Added to this is the traceability of intent: who created the rule, when, and why—all in two lines. This is what makes it possible to make a decision in thirty seconds eighteen months later, rather than extending it out of an abundance of caution.
The accumulation of rules is a cumulative process, not an isolated incident: it grows with every legitimate decision made in isolation, even though no single decision is open to criticism. Five forms predominate: the exception that has become permanent; the seasonal rule that is never suspended; rules that contradict one another; the orphaned rule whose author has left; and the redundant rule.
The cost isn't what you might think: before it eats into profit margins, debt erodes trust. A team that no longer understands why the system suggests a certain price will stop following it. The solution is a consistent routine rather than a major cleanup effort every three years: a quarterly portfolio review, four questions per rule, and an expiration date set at the time of creation.
Pricing decisions account for only a small portion of the workweek in most organizations. The rest is divided among preparation, coordination, formatting, and justification. This time is not wasted: justifying a price to a buyer is real work. The problem is that it takes the place of what creates value.
The measurement is based on a reconstructed typical week, not on self-reported estimates: the difference between the two is often a factor of two. The correction then follows a specific order: eliminate, then automate, then delegate. Automating a useless task is tantamount to making it permanent.
Reversibility is negotiated before the contract is signed, never after. Once the contract is signed and the system is live, the balance of power has completely shifted. Four assets are at stake: your raw data, your rule set, your trading history, and your competitive matching. Only the first is generally covered by contracts.
The gap between perception and reality is well documented: 89% of executives believe they can switch suppliers in less than a month, but among those who have tried, 58% report that the effort failed or took much longer than expected (Zapier, 2026). The goal isn’t to leave, but to have the option to do so: that’s what maintains a healthy supplier relationship throughout the contract term.