Most of what goes wrong in a small business does not arrive as an event. It arrives as a drift. A number that used to sit in one place now sits somewhere slightly worse, has done for a while, and nobody decided it.
This playbook sets up a watch for that. The number worth telling you about is almost always a change against your own normal, rather than a fixed figure.
Why a fixed alert is the wrong tool
Say you set an alert at ₹50,000 of packaging spend in a month. For a workshop that packs forty orders a week, ₹50,000 is a month gone badly wrong. For a distributor shipping pallets, it is a quiet Tuesday. The same figure is an emergency for one business and beneath notice for another, and it is often both for the same business, six months apart.
So a fixed alert lands in one of two states. Either it never fires, and you slowly stop believing it is switched on at all. Or it fires every single month, you mute it, and it is now decoration. Both of these look like a working alert on the screen.
Instead, a drift watch asks is this number different from what this business normally does? That question has a useful answer for a jeweller, a clinic and a courier without anyone tuning a figure by hand.
The four shapes drift usually takes
Each of these is invisible in a single month and obvious across three.
- Supplier prices creeping. A component moves from ₹120 to ₹126, then to ₹131, then to ₹138. Every one of those is a rounding error you would never argue about on a phone call. Together they are a 15% increase you never agreed to, sitting inside every invoice you send.
- A margin that slipped and stayed slipped. You used to make 22 points on a line and now you make 20. Sales are fine, the bank balance is fine, nothing looks broken. On ₹12,00,000 of monthly sales, those two points are ₹24,000 a month (roughly a salary) leaving without a decision anywhere behind it.
- A marketplace commission that changed. The rate card moved by a notice that arrived in a mailbox nobody reads, and your settlements have been a little thinner ever since. The orders still come, the payouts still land, and the difference is buried in a deduction line inside a settlement file.
- A customer who quietly stopped ordering. Nobody complains and nobody cancels. A shop that ordered ₹40,000 every month has ordered ₹6,000 for two months running. You find out at year-end, by which time they have a new supplier and a routine that does not include you.
None of these is a failure of attention. You looked at every one of those numbers the month it happened, and every one of them looked fine, because in a single month it was fine.
Set up the watch
Pick one thing that would actually hurt
Choose one drift: the one that would cost you most if it ran for a year unnoticed. That is usually purchase prices or margin for a trader, top-customer volume for a distributor, marketplace deductions for an online seller.Create an agent named for that job
On Automations, create an agent and call it what it watches, such as Margin watch or Supplier price watch. Tick only the parts of your books it needs. A price watch has no business reading payroll, and if you never tick it, it cannot.Add the automation and set the comparison, not a limit
Choose how far back it compares (the last three months is the usual starting point) and how big a move is worth a message. A sensible first setting is wide: tell me when something moves more than a tenth against its own recent normal. You can tighten it once you have seen what it says.Read the monthly cost, then switch it on
The figure is shown before you commit. A watch that runs weekly costs very little, and on a quiet week (most weeks) the check finds nothing and costs nothing at all. See What automations cost.Give it a month before you judge it
The first fortnight of any drift watch is you learning what it calls a change. If it is noisy, widen the threshold. If it has said nothing at all and you know something moved, narrow it. Both are ordinary.
Be honest about what it needs
A drift watch compares you to yourself, which means it is useless until there is a yourself to compare against.
- It needs history. A workspace three weeks old has nothing to measure a change against. Everything is the first of its kind, so everything is either silence or a false alarm. Give it two or three months of real entries first; a watch switched on in week one cannot work yet.
- Seasonal businesses need like-for-like. If you sell sweets, October is not a drift from September; it is Diwali. A watch comparing this month to last month will scream every festival and then apologise every January. Compare a season to the same season last year, and if a full year of history does not exist yet, treat the alerts as questions rather than findings.
- It can only see what you have recorded. If half your purchases arrive as a photograph of a bill in three weeks’ time, the drift arrives three weeks late too. The watch is only ever as current as your books.
What to do when it does say something
The value of a drift watch is that it hands you a specific question with a date on it. This supplier’s rate has moved three times since April. This customer has ordered a fifth of their usual for two months. That is a conversation you can have without accusing anyone of anything, because you are asking about a number rather than about a person.
The finding stops with you. An agent will not renegotiate a rate, will not message the customer who went quiet, and will not adjust a price. See What an agent will never do. Every run it makes, including the quiet ones, is on the record.
Related: What an agent is, Budgets & ratio analysis, Report insights & anomalies.