If your software (or your AI dashboard) produces a weekly list of hundreds or thousands of things that "don't add up," the normal outcome is that nobody looks at it anymore. Not because people are lazy: because out of a thousand lines almost none matter, and checking them one by one takes longer than the list saves you. The problem isn't that the list flags too little. It's that it flags everything, and flagging everything is the same as flagging nothing.
The myth: the more your system flags, the better watched your business is
It sounds like common sense: if a program can detect that something doesn't add up (an impossible margin, a wrong price, stock that doesn't match), more detection should always be better. The finer the filter, the safer you sleep.
It doesn't work that way. A rule that says "show me everything that doesn't add up" can be technically perfect and still be useless if it doesn't tell apart what's actually alive in your business today from what's dead history in your catalog. If ninety five out of a hundred results are about a product that hasn't sold in a year, nobody is going to bother hunting for the five that do matter. And the day something genuinely serious happens, it will be hiding in that same list, the one nobody opens anymore.
Why does a list of thousands of lines end up with nobody looking at it?
It happened, with concrete numbers, to a pharmacy group we work with. They wanted a list of products with a margin that couldn't be real: a wrong selling price, or a purchase cost read incorrectly when it came into the system. The rule itself was simple to calculate. The problem showed up when they looked at the result: most of those "impossible margins" came from products that hadn't had a single unit in stock or a single sale in months or years. Dead entries in the catalog that, by pure arithmetic coincidence, also happened to meet the "this doesn't add up" condition.
Without discarding those entries first, the full list had over 52,000 products. Nobody was ever going to go through it, and there was no need to: almost all of it was noise. A prior step was added that strips out products with no stock and no recent sales before applying the margin rule. With that filter, the list dropped to around 2,900 products, and those 2,900 are the ones that are genuinely alive in the business today, the ones the owner can check, or ask directly to the AI assistant they use to review their catalog.
The change wasn't making the rule smarter. It was deciding, before applying it, which part of the catalog was even worth asking the question about.
The rule: before asking if something doesn't add up, ask if it's still alive
This isn't just about pharmacies or margins. The same pattern shows up in any list that compares two figures over a large catalog, a customer base, or an order list: stock running low, reviews dropping, customers who've gone quiet for a while, prices that don't match a supplier's rate. Any business with hundreds or thousands of products, customers or orders carries the same risk: part of that list no longer represents anything happening today, and if the rule doesn't know that, it counts that part exactly the same as the part that matters.
The decision rule is this: if a result can come up for something that's no longer alive (no movement, no stock, no recent activity), that list isn't ready until it filters that out first. It isn't a technical detail. It's the difference between a list someone opens every week and a list nobody touches again after the second month.
How do you know if your own lists have this problem?
You don't need any new software to check. With what you already have:
- Take the last list your system produced (stock, margin, whatever it is) and count how many of those items have had any sale or movement in the last three months. If most haven't had any, that's the filter you're missing.
- Ask, plainly, when was the last time someone on the team went through that whole list. If the answer is "months ago" or "we get it but don't open it," the problem isn't whoever receives it: it's that the list was never designed to be opened by anyone.
- Ask to see, out of the last results, how many ended in a real action (changing a price, placing an order, calling a customer) versus how many were closed without touching anything because they no longer applied. If the second number is much bigger than the first, that's the noise you need to strip out before fine-tuning the rule any further.
It's the exact same work we do when we bring the data from several of a business's programs into one place: it isn't enough for the AI to calculate correctly; someone has to decide, catalog by catalog, which part of the data no longer counts as the present. You can see what that looks like in the case of a data platform for a pharmacy group.
If you want to check how much of the lists your business already gets is signal and how much is noise, in our services we explain how we approach it with the data you already have. And if you'd rather talk it through directly, reach out and we'll look at your own lists together.
