You have a trustworthy daily feed of matched competitor prices landing where people can see it. The last question is what anyone should do with it, and this is where a good data project can still destroy value.
The naive rule — be one cent below the cheapest competitor — is the one everybody writes first. It is worth understanding precisely why it is bad before replacing it.
Why matching the cheapest fails
It hands your pricing to your least informed competitor. Somewhere in any set of rivals is a seller clearing stock, miscalculating shipping, or simply making a mistake. A rule that follows the minimum follows that seller, and the rest of the market follows you.
It ignores everything a shopper is actually comparing. Delivery speed and cost, returns, stock availability, whether the buyer trusts the site. Being three percent dearer with next-day delivery and real stock is a winning position in most categories; a rule that only reads a number cannot see that.
It races. If two sellers in a market both run undercut rules, the price falls to the floor within days, and neither of them chose that outcome. This is visible and documented on marketplaces, and it is the main reason serious repricing systems include a floor and a cooldown.
And it reacts to noise. A competitor who is out of stock still displays a price, often a stale one. Following it means repricing against a product nobody can buy.
What a usable rule contains
A competitor price is one input. These are the others.
- 1
A margin floor, per product
The price below which you would rather lose the sale. Non-negotiable, checked last, and it should override every other part of the rule. Most repricing disasters are a missing or a global-rather-than-per-product floor.
- 2
A reference set, not a minimum
Decide which competitors the rule may react to — usually the two or three from lesson two that genuinely take your customers — and use a position within that set, such as "the second cheapest" or "the median", rather than the absolute minimum. It is dramatically more stable.
- 3
An availability condition
Ignore the price of any competitor whose page says out of stock. This single condition removes a large share of spurious triggers, and it is only possible because lesson four told you to keep the availability text.
- 4
A movement threshold and a cooldown
Do not act on a change smaller than, say, one percent, and do not change the same product's price more than once in a defined window. Thresholds stop you chasing rounding; cooldowns stop two automated systems from spiralling.
- 5
A total-cost comparison where you can get it
Where you captured shipping, compare delivered price rather than shelf price. A rival who is two percent cheaper and charges for delivery is not cheaper, and a rule reading only the shelf price will tell you they are.
Three rules that tend to hold up
Illustrative shapes rather than recommendations. The right rule depends on your category and your position in it.
| Situation | Rule shape | Why it works |
|---|---|---|
| You are the service leader | Stay within +4% of the reference median, never below the floor | Captures the price-sensitive shopper without giving away the premium you have earned |
| Clearing end-of-life stock | Match the second cheapest in-stock reference, floor at cost | Moves units without leading the market down on live lines |
| Enforcing a supplier's minimum price | Never price below MAP; alert when a reference breaks it | The output is an evidence trail for your supplier, not a price change |
Measure the rule, not the feed
The last trap is reporting on the wrong thing. It is tempting to report pipeline health — rows collected, match rate, uptime — because those numbers are easy and they go up. They are not the point.
The questions worth answering quarterly are: how many of our in-scope products are currently priced outside the band we intended, how quickly do we react when a reference competitor moves, and what happened to margin and volume on the lines the rule touched compared with the lines it did not. That last comparison is the only honest measure of whether any of this worked, and it requires that you deliberately leave a set of products out of the rule so there is something to compare against.
Hold one back. A control group of a few hundred SKUs costs you almost nothing and is the difference between knowing the project paid for itself and believing it.
Worked example: one SKU through three rules
A product you sell at €99.00 that costs you €72.00, so €27.00 of margin, with a floor set at 12% — €80.64. Three competitors are visible this morning: €104.00 in stock, €96.50 in stock, and €84.00 marked "2–3 weeks". The first rule cuts your margin by more than half to beat a price the shopper cannot buy today. The third rule raises the price.
| Rule | Reference price | New price | Margin kept |
|---|---|---|---|
| Match the cheapest, any availability | €84.00 | €83.90 | €11.90 |
| Match the cheapest that is in stock | €96.50 | €96.40 | €24.40 |
| Sit 1% under the median of in-stock rivals | median of €104.00 and €96.50 = €100.25 | €99.25 | €27.25 |
What usually goes wrong
Each of these is a rule that works on the day it is written and costs money a month later.
- No availability condition. The cheapest figure on the page is very often the price of something nobody can ship, and it drags your whole range down behind it.
- No floor. A rule without a per-product margin floor will follow a competitor clearing stock all the way into a loss, at machine speed, overnight.
- No movement threshold. Reacting to a four-cent change generates hundreds of price updates a day, each one a row in a channel feed and a fresh chance to be rejected.
- No cooldown. Two automated repricers pointed at each other will walk a price downwards in a loop, and the loop finishes long before anyone reads the alert.
- Automating in week one. Run it advisory and count the overrides: disagreement on one row in twenty means it is ready, and one in four means the rule is wrong, not the human.
- No control group. Without a held-out set of SKUs the rule never has to prove it beat doing nothing, and nobody can answer whether the pipeline paid for itself.
That is the course. If you want the collection stage handled, the retailer write-ups show what a run on a specific site returns and what it costs.
Browse the retailer write-ups