TL;DR
Rule-based repricing does what you tell it. AI repricing works out what to do. Neither is better, and in 2026 the question isn’t which one you buy. Most serious tools offer both, so the real decision is which engine you assign to which SKU: rules where you need to explain the price (MAP lines, B2B tiers, private label), AI where the volatility outruns anything you could maintain by hand (the contested long tail). If you’re picking a side, you’re answering the wrong question.
Ask this online and you’ll get a fight. AI evangelists say rules are obsolete. Rules people say AI is a black box burning their margin while nobody’s looking.
They’re both describing real failures. They’re just describing failures of the wrong engine on the wrong SKU.
So here’s the honest comparison: how each actually works, what each is genuinely bad at, who each suits, and then the part most articles skip, which is how to run both without them fighting each other.
New to automated pricing? Start with what repricing is and come back.
What rule-based repricing actually is
You write conditions. The tool executes them. Nothing more mysterious than that.
IF the Buy Box holder is priced above my floor THEN price 1% below them UNLESS that breaches my floor, in which case hold
Every price change traces back to something you wrote. If a price looks wrong, you can find the rule that did it in about thirty seconds. That auditability is the entire value proposition, and it’s worth more than people give it credit for.
Where it genuinely wins
- When you have to explain the price. MAP agreements, B2B tiers, brand commitments. “The algorithm decided” is not an answer you can give a supplier.
- When the logic is genuinely simple. A quiet category with two competitors doesn’t need machine learning. It needs a floor and a rule.
- When you’re learning. Watching your own rules fire teaches you how the Buy Box behaves. That intuition is worth having before you delegate it.
- When you need to debug. Something’s wrong on 40 SKUs. With rules, you read the rule. With AI, you file a support ticket.
Where it genuinely fails
- It’s frozen. A rule written in January knows nothing about March. It doesn’t notice that a competitor changed behaviour, that demand shifted, or that your category got more crowded. It just keeps executing.
- It doesn’t scale past your attention. Twenty rules is a system. Two hundred is a liability, because you won’t maintain them and the stale ones keep firing.
- It races. “Always undercut the lowest” is a rule. It’s also how price wars start, and our price war guide covers where that ends.
- It can’t see round corners. A rule reacts to a competitor’s price. It can’t notice that this competitor always goes quiet at 9pm, or that a SKU holds the box comfortably 40p higher.
What AI repricing actually is
The tool reads competitor behaviour, sales velocity, Buy Box state, stock levels and timing together, then picks a price. It learns from what happened last time.
The important word is learns. If the tool behaves identically in month six to month one, it isn’t AI, it’s rules with a badge on.
Where it genuinely wins
- The long tail. Three thousand contested SKUs, each with its own competitor set and rhythm. You were never going to write rules for those, and if you had, you’d never have updated them.
- Pattern recognition. The stuff you’d spot if you had time to stare at one listing for a month, applied to all of them at once.
- Pricing up. This is the underrated one. Good AI hunts the ceiling, raising your price while you hold the box to find where demand stops. Most rule sets only ever move down, which means you’ve automated your losses and left your gains manual.
- Volatility. When the market moves faster than you can rewrite instructions.
Where it genuinely fails
- You can’t ask it why. For most SKUs, fine. For the one where your supplier just rang about MAP, not fine.
- It needs a fortnight. There’s a learning period, and sellers who judge it on day two yank the settings before any pattern emerges.
- It’s confident on thin data. A SKU that sells twice a month doesn’t give the model much to learn from.
- “AI” is a marketing word. Plenty of tools run a few extra conditionals and call it intelligence. The test is whether it updates from outcomes.
Side by side
| Rule-based | AI-driven | |
| Decides by | Conditions you wrote | Patterns it learned |
| Transparency | Full, every change traceable | Limited, you see outcomes not reasoning |
| Adapts | Only when you rewrite it | Continuously |
| Prices upward | Only if you built that rule | Yes, actively |
| Scales to | About as far as you’ll maintain it | The whole catalogue |
| Time to value | Immediate | One to two weeks of learning |
| Debugging | Read the rule | Read the reports |
| Thin-data SKUs | Fine | Weak |
| Explaining to a supplier | Easy | Awkward |
| Suits | MAP, B2B, private label, small catalogues | Contested long tail, volatile categories |
Read that as a division of labour, not a scoreboard.
Who each suits
Rules suit you if: you’re under roughly 500 SKUs and can genuinely maintain the logic; you’re in a stable category where prices move weekly rather than hourly; you have MAP obligations or B2B tiers where the price needs justifying; or you’re new enough that watching the rules fire is still teaching you something.
AI suits you if: you’re past roughly 1,000 SKUs; your categories move constantly; you’ve stopped updating rules because there are too many; or you want prices to climb rather than only defend.
And if you’re between the two, which is most established sellers, you’re the hybrid case. Which brings us to the useful part.
The honest answer: it’s an assignment decision
Here’s what nobody selling you either engine will say plainly.
In 2026, most serious repricers offer both. So “rule-based vs AI” stopped being a purchase decision and became a configuration decision. The question isn’t which engine you buy. It’s which engine each SKU gets.
| SKU type | Engine | Why |
| MAP-restricted lines | Rules | The price must be defensible |
| B2B and tiered offers | Rules | Amazon Business logic needs to be explicit |
| Private label, uncontested | Rules or ceiling-hunt AI | Little competitor signal to learn from |
| Flagship products | Rules | Brand positioning beats optimisation |
| Contested wholesale | AI | Volatility outruns any rule you’d maintain |
| The long tail | AI | You were never writing rules for 2,000 SKUs |
| Clearance and aged stock | Rules | You want a specific, deliberate outcome |
| New launches | Rules first, AI later | No data to learn from yet |
For a mixed catalogue, most sellers end up with AI across the contested majority and rules across the segments that need explaining. The split follows your catalogue, not a formula.
The repricing strategies page covers the patterns each engine runs, and the rules guide gives you copyable conditions if you’re building the rules half.
What both engines need underneath
Worth being blunt: this whole debate is downstream of something more important.
Neither engine can protect a floor that’s wrong. If your minimum price is a number you typed last year, AI will lose money efficiently and rules will lose money predictably. That’s the entire difference.
Build the floor from landed cost plus every fee plus your target margin, and make sure it recalculates when Amazon’s fees move. Our net margin guide covers the calculation, and minimum price floors work off net position rather than a flat number.
Get that right and either engine works. Get it wrong and neither does.
How to run both without them fighting
- One engine per SKU. Never both on the same listing. They’ll argue, and you’ll spend a week working out which one moved the price.
- Segment first, assign second. Group by margin, competition and how much you need to explain the price. Then assign.
- Give AI a fortnight. It has a learning period. Judging it on day two is the most common reason sellers conclude “AI doesn’t work”.
- Watch the ratio. Private label should barely move. Contested ASINs should move constantly. If that’s inverted, your engines are on the wrong SKUs.
- Read win rate against profit. Always together. Buy Box share on its own is trivially easy to buy at your floor, and the win rate guide covers reading it properly.
Repricer runs both engines side by side and lets you assign per group, which is the point: its AI Buy Box optimizer handles the contested tail while rules hold the segments that need explaining. If configuring that across thousands of SKUs is the job that never gets done, managed setup puts someone on it with you, and analytics and reporting shows whether the assignment was right.
FAQ
Is AI repricing better than rule-based repricing? Neither is better in the abstract, and the framing is the problem. Rules win where the price has to be explainable and the logic is simple enough to maintain. AI wins where volatility outruns any rule set you’d realistically keep current. Most established sellers run both, assigned per segment, which is why “which is better” is the wrong question to be asking.
Can I use rule-based and AI repricing at the same time? Yes, and most serious catalogues should. The one hard rule: never both on the same SKU, because they’ll fight and you’ll have no idea which engine caused a given price. Assign one engine per listing or per group, typically AI on contested lines and rules on MAP, B2B, private label and clearance.
Do I need AI if my catalogue is small? Probably not. Under about 500 SKUs, rule-based logic is cleaner, cheaper to reason about and far easier to debug when something looks wrong. AI starts earning its place past roughly 1,000 SKUs, where the volatility across the catalogue outruns whatever rules you could maintain by hand.
Is AI repricing a black box? Partly, and that’s the honest trade. You see outcomes rather than reasoning, which is fine on 2,000 contested SKUs and uncomfortable on the one where a supplier is asking why you breached MAP. That’s precisely why the segments needing explanation should stay on rules rather than being handed to the algorithm for tidiness.
How do I know if a tool’s “AI” is real? Ask whether it updates from outcomes. Real algorithmic pricing learns what won the Buy Box at what price and adjusts accordingly, so it behaves differently in month six than month one. If the tool does the same thing forever, it’s rules with a marketing badge, and you should evaluate it as rules.
Which engine protects my margin better? Neither, on its own. Margin is protected by the floor underneath both, not by the engine on top. A floor calculated from landed cost plus fees plus target margin, that recalculates when fees change, is what stops either engine selling at a loss. Get that wrong and AI just loses money faster than rules do.
Where to start
Don’t pick a side. Open your catalogue and mark each segment with one question: would I need to explain this price to someone?
Everything you’d have to justify (MAP, B2B, flagship, clearance) goes on rules. Everything else, especially the contested tail you’ve never had time to think about, goes to the algorithm.
Then fix your floors, because neither engine saves you from a bad one.
If you want both engines running side by side against your own catalogue:
