The system that spots a departing customer before the sales rep does

The sales rep was surprised. Yet all the signals had been sitting in the system since March: in the order history, in the invoices, in the payments. What was missing was someone to connect them and say out loud that this customer was leaving.

Why the sales rep is the last to know

In B2B, customers rarely announce their departure directly. Before the decision falls, the buyer spends many months testing the competition: moving part of the volume there, stretching the gaps between orders with the current supplier, caring less about paying invoices on time. Leaving looks like a slow fading - and a fading customer generates no tasks: he does not call, does not complain, does not ask about delivery dates.

A sales rep handling a few dozen customers inevitably deals with the loud ones: current orders, complaints, negotiations. Sales reports do not help either, because they show totals - and in a total, one buyer's decline can vanish if another happens to grow. A manual customer review once a quarter catches the problem when the decision on the other side has often already been made.

The heart of it: the company holds the data about the coming departure long before the customer's phone call. What is missing is a mechanism that reads that data every day.

A mechanism in three parts

Such a mechanism can be built in the system the company already has - where orders, invoices and payments are recorded. It works anywhere regular customers place repeat orders. It consists of three parts.

1. The customer's rhythm, calculated from their own history. Every regular buyer has a natural rhythm: a typical gap between orders, a typical volume, typical payment behaviour. The system calculates that rhythm from the last two years and treats it as the reference point; the length of that period can be set freely in the configuration. Comparing a customer with their own history is fairer than comparing them with the company average, because a buyer who orders quarterly is not an at-risk customer for that reason alone.

2. Points for deviations instead of single alarms. One weaker month is noise that happens to everyone. So the system does not react to a single signal; it adds up deviations: a lengthening gap between orders, shrinking volumes, growing payment arrears, orders narrowing to an ever smaller set of items. Only several signals at once raise the risk score. This is also the natural place for AI: telling a seasonal dip from quiet leaving, catching a pattern that is hard to write down as an explicit rule. Simple rules catch the obvious cases, AI adds the less obvious ones.

3. A “red light” with a specific customer and the data. When the risk score crosses the threshold, the system marks the customer as at risk and sends an alert wherever it is needed - with the full picture: what the rhythm looked like, what changed and since when. Such an alert ends in a task, for example a conversation with the customer this week, rather than a vague sense that “something is going on”. The difference is fundamental: a conversation held in April saves the relationship, the same conversation in August is just listening to a decision.

What this means for a business owner

The sums are worth doing on your own numbers. Take one regular customer and their average monthly margin. Multiply it by the number of months it realistically takes to win a comparable customer - in manufacturing that is usually a long process: enquiries, samples, audits, negotiations, first trial orders. All that time, the hole left by the lost buyer stays in the production plan, and fixed costs do not go away: machines depreciate whether they run or not, the shop floor and the people cost the same every month.

Then there is the other side of the sums: keeping an existing customer is cheaper than winning a new one, and the conversation triggered by an early alert costs an hour of the sales rep's or the owner's time. Setting those two amounts side by side - the value of the relationship at risk and the cost of an early reaction - every business owner can do for themselves. That comparison is the entire business case for this mechanism.

Where to start

The good news: the data is already there. Order history, invoices and payments sit in the sales or accounting system. A sensible order of work looks like this: first, a list of regular customers with their share of revenue - to know whose departure really hurts. Then each customer's rhythm calculated from the last two years: gaps between orders, volumes, payment punctuality. Finally, simple rules comparing current behaviour with that rhythm, and an alert sent wherever it is needed.

If you would like to see what such a mechanism could look like on your data, write to me or book 30 minutes in the calendar below.

More articles about what can be built in your company are in the Knowledge section.