How many customers do you need before the groups mean anything?

There is no fixed number, because the thresholds are the tertiles of your own shop's distribution: the screen calculates the groups only when there are enough customers to describe a distribution and when spend is not identical for everyone. If it tells you segmentation was not calculated, that is not an error: it means any banding would be arbitrary with that data. Meanwhile the individual cards and the reorder forecast remain perfectly usable.

Can the platform message drifting customers by itself?

No. The screen calculates groups, spend, frequency and the reorder forecast, but it sends no email and no messages and there is no scheduled sending. Contact is made by a person, working from the list you built. This is a real limit and it also has an upside: it forces you to choose who to contact instead of writing to everybody, and a short list contacted well outperforms a long list contacted automatically.

Why is the run out date missing for some customers?

For two possible reasons, and the screen tells you which. The first is that the pack weight is not declared in the catalogue: without it, daily consumption cannot be estimated. The second is that the customer has only one recorded purchase, so there is no interval to measure yet. The first case is solved by completing the catalogue, the second by waiting for the second purchase or simply asking the customer how long a pack lasts.

How often should the contact round be run?

Every two weeks. The useful window for an effective contact closes shortly after the forecast run out date, and for food that date typically falls within six weeks of the previous purchase. A quarterly round consistently arrives once the customer has already bought elsewhere and already formed the new habit, which is the hard part to reverse. Twenty names every two weeks beats eighty every three months.

Running a pet shop45 minLevel: IntermediateUpdated on August 4, 2026

Following up the customers who stop coming back

How to read the customer groups, build the list of people who stopped buying and test the four verifiable reasons a customer stops coming back.

Audience: Pet shops
Species: All species
Scope: Italy, European Union, Valid everywhere

What you need before you start

Un accesso all'area negozio con il permesso di consultare la clientela
Almeno sei mesi di ordini consegnati e prestazioni completate registrati nel sistema
Il catalogo compilato con il peso della confezione, se vuoi la previsione di riacquisto
Una persona incaricata dei contatti e un'ora protetta a settimana da dedicarci

A customer does not leave with a farewell letter. They stretch the gap between one purchase and the next, skip a booking, buy the large bag elsewhere and only the rest from you, and at some point stop altogether. The transition has no recognisable moment, which is why almost no shop notices until the year end figures tell a different story from the expected one. The difference between noticing and not noticing lies not in instinct but in looking at two numbers per customer, the date of the last purchase and the usual interval between purchases, and doing it on a fixed rhythm. This guide explains how the customer screen builds its groups, how to read the reorder forecast, how to build the list of people drifting away and which four causes of drift you can actually verify. It also states what the system does not do.

A customer does not leave, they stop coming back

The way a shop loses a customer is almost always the same. First they buy everything they need from you. Then they find somewhere else something you did not have that day. Then they get used to buying that item elsewhere. Then they realise they might as well pick up the rest in the same place. None of these steps produces a complaint, so none of them ever reaches your ears.

The only available signal is time. A customer who buys a bag of food every five weeks and has not appeared by week nine has already bought elsewhere, almost certainly. At that point contact still makes sense; three months later it becomes a cold call nobody welcomes. The useful window is narrow and it can be calculated, and calculating it is exactly what the reorder forecast does.

Where the numbers on the screen come from

Before interpreting any group you need to know what enters the calculation, otherwise you take decisions from a partial photograph. The rule is simple and has an important consequence for anyone who sells mainly across the counter.

Only delivered orders and completed services count: a cancelled order or an unfinished booking never enters the calculation.
A counter sale not recorded as an order does not exist for this screen: if a large share of your sales goes through the till unrecorded, the groups describe only a slice of your customer base.
The thresholds separating the groups are the tertiles of your own shop's distribution, not absolute values: they move as the data moves and cannot be compared with another shop's.
Segmentation is not calculated when there are too few customers for tertiles to describe a distribution, or when everyone spends the same.
Frequency stays unmeasurable until enough customers have at least two purchases: with a single purchase there is no interval to measure.
A customer with animals of different species is counted under each species, so the sum of species served can exceed the number of customers.

Reading the groups and the thresholds

Groups come from crossing two bands, spend and frequency, each split into low, medium and high according to your shop's tertiles. The point to hold onto is that a customer in the low band is not a bad customer: they are a customer who spends less than most of your customers, which also depends on how many animals they have and how big those animals are.

The customer groups and what they suggest
Group	What it describes	What is worth doing
Best customers	High spend and high frequency	Protect them: an absence should be noticed within two weeks
Frequent	High frequency with spend that is not high	Check what they buy elsewhere, usually the large pack
High spend	High spend with frequency that is not high	Watch the reorder forecast, long intervals are natural here
Regular	Middle bands on both axes	The group where recovery pays best, because it is the largest
Occasional	Low bands on both axes	Contact only with a concrete reason, otherwise you spend goodwill
New	Too short a history to be placed	The second purchase is the decisive moment and needs support
Unclassified	Not enough data for a band	No automatic action, more recorded purchases are needed

The individual customer card adds what the groups cannot say: total spend, average basket, number of orders and services, date of the last purchase and the yearly purchase rate. It is those five figures together, not the group label, that tell you whether someone is drifting out.

Building the drift list in nine steps

Fix one day every two weeks
The rhythm matters more than the depth of the analysis. Every two weeks keeps the contact window open; once a quarter always arrives late, because the average reorder interval for pet food is shorter than a quarter.
Open the customer screen and read the four tiles
Customers, total spend, orders and services. They show whether anything moved since the last round: a drop in order count with the same number of customers means intervals are stretching, and that is the first sign of collective drift.
Check whether segmentation was calculated
If the screen says it cannot be calculated, do not push the groups: it means there are too few customers or that spend has no spread. In that case work directly from the individual cards, which remain usable.
Start with best customers and frequent buyers
These are the groups where an absence is most unusual and most expensive. Scroll the cards and note anyone whose last purchase is older than their own usual interval: it is that comparison that counts, not a single threshold for everyone.
Open the reorder forecast for the names you noted
The forecast shows the cadence in days, the estimated run out date with its uncertainty range and the date from which reordering makes sense. Look at the confidence too: low, medium or high. A low confidence forecast is a hint, not a fact.
Keep only those marked due or late
Urgency has four values: on track, soon, due and late. The first two should not be contacted, because contacting someone who needs nothing yet burns the credit you will need later. The last two are your list.
Drop anyone with a single purchase
With one purchase there is no interval, and the screen says so explicitly. Those people are not drifting customers: they are customers who have not made a second purchase yet, which is a different problem handled a different way.
Write a suspected cause next to each name
Before calling, write down which of the four causes in the next section looks most likely, based on what they used to buy and what they stopped buying. Calling without a hypothesis produces a generic conversation that teaches you nothing.
Make contact, listen and record the answer
You make the contact: the platform sends no messages and no email. The purpose of the call is not to sell but to find out why: the sale, when there is a reason for one, comes on its own within the same conversation.

The four verifiable causes of drift

There are many reasons a customer stops coming, but only four can be checked against the data you hold and corrected by a decision of your own. The others, such as a house move or the loss of an animal, should simply be recorded so that the row is closed and the contact never repeated.

Causes, the trace in the data and how to check
Cause	Trace in the data	How to verify it
Repeated stockouts	The customer stopped buying one specific product and continues with the rest	Check the dates on which that product was out of stock
Wrong pack size	Short reorder intervals and a low average basket	Compare the effective cost per kilogram across the available sizes
A service never repeated	Completed services that stop after the first or second	Reread the notes on the work record for that service
Diet or species change	The customer still buys but completely different products	Ask directly: there is often a prescription behind it
Causes outside your control	A clean stop with no earlier signals	One respectful contact, then close the row
Incomplete recording	A long standing customer showing as absent for months	Check whether their purchases go through the till unrecorded

The most frequent and most correctable cause is the second one. A customer who always buys the small pack because that is what you display pays more per kilogram and sooner or later finds the large pack elsewhere. The screen calculates which size is cheapest per effective kilogram and how much a year would be saved against the worst size: that is something you can say across the counter in ten seconds, and it changes a buying habit without any discount.

What to offer and what not to offer

Faced with a list of drifting customers the temptation is to offer everyone a discount. It is the most expensive move and the least effective, because it rewards those who were coming back anyway and teaches everybody else to wait for the next discount. A voucher only makes sense when it solves the cause you identified.

If the cause is a stockout, the right offer is not a discount but a guarantee of availability: say when the product is back and put one aside.
If the cause is pack size, propose the better size and explain the effective cost per kilogram: that is information, not promotion, and it stays true afterwards.
If the cause is a service never repeated, ask what did not work before offering to repeat it: a second identical service produces an identical outcome.
If you do use a voucher, give it a short validity window and a minimum spend consistent with that customer's average basket, not with your target.
Never use a voucher on somebody simply ahead of their forecast: you are paying for a purchase that was coming anyway.
Record the redemption as a redemption: counted uses are the only way to know whether the voucher produced purchases or only lost margin.

Measuring recovery without fooling yourself

The correct measure is not how many customers you contacted, nor how many replied. It is how many made a purchase within sixty days of the contact, compared with how many would have done so anyway. The second number cannot be obtained precisely, but it can be approximated well by keeping a small part of the list uncontacted and watching what happens.

Indicators for the fortnightly round
Indicator	How it is calculated	What a decline means
Rows worked	Names contacted out of the whole list	The protected hour is not being protected
Reason captured	Contacts where you obtained a cause	The conversation starts from the sale rather than the question
Return within sixty days	Contacted customers who bought again	You are calling too late relative to the forecast
Correctable causes	Share of the first three causes among those captured	The problem is organisational rather than commercial
Rows closed	Contacts with a cause outside your control	No action needed, it only prevents calling again

After three rounds, roughly six weeks, you will have collected enough reasons to see a concentration. If half the answers point to the same cause, the useful work is no longer individual recovery but correcting that cause, and at that point the list shortens on its own.

Frequently asked questions

How many customers do you need before the groups mean anything?: There is no fixed number, because the thresholds are the tertiles of your own shop's distribution: the screen calculates the groups only when there are enough customers to describe a distribution and when spend is not identical for everyone. If it tells you segmentation was not calculated, that is not an error: it means any banding would be arbitrary with that data. Meanwhile the individual cards and the reorder forecast remain perfectly usable.
Can the platform message drifting customers by itself?: No. The screen calculates groups, spend, frequency and the reorder forecast, but it sends no email and no messages and there is no scheduled sending. Contact is made by a person, working from the list you built. This is a real limit and it also has an upside: it forces you to choose who to contact instead of writing to everybody, and a short list contacted well outperforms a long list contacted automatically.
Why is the run out date missing for some customers?: For two possible reasons, and the screen tells you which. The first is that the pack weight is not declared in the catalogue: without it, daily consumption cannot be estimated. The second is that the customer has only one recorded purchase, so there is no interval to measure yet. The first case is solved by completing the catalogue, the second by waiting for the second purchase or simply asking the customer how long a pack lasts.
How often should the contact round be run?: Every two weeks. The useful window for an effective contact closes shortly after the forecast run out date, and for food that date typically falls within six weeks of the previous purchase. A quarterly round consistently arrives once the customer has already bought elsewhere and already formed the new habit, which is the hard part to reverse. Twenty names every two weeks beats eighty every three months.

What to do next

Every two weeks open the customer screen, start from best customers and frequent buyers and note anyone whose last purchase is older than their own usual interval. Keep only those marked due or late, drop anyone with a single purchase, write a suspected cause next to each name and then call to understand rather than to sell. Record the reason in fixed words: after three rounds you will know whether the problem is individual or whether it is one problem affecting everybody.