How to segment a store's customers?

Not all of a store's customers behave the same way. Some come every week, others a few times a year. Some make large baskets, others buy little but regularly. Some have just made their first purchase while former regulars are starting to stop coming back.

Customer segmentation means grouping customers who share characteristics or behaviors in order to better understand your customer base and tailor the actions aimed at it.

The goal is therefore not to create as many segments as possible. It is to build groups relevant enough to lead to a different decision or action.

Why segment your customers?

A customer base taken as a whole gives little actionable information. Imagine 5,000 identified customers. Knowing that their overall average basket is €42 doesn't tell you:

• which ones buy regularly;
• which ones are visiting for the first time;
• which ones generate the largest amounts;
• which ones no longer come back;
• which product categories interest certain groups;
• which customers seem to be starting to drift away.

Segmentation lets you move from a mass of customers to groups whose behavior can be analyzed separately. It can then be used to tailor a message, an offer, a follow-up or simply the analysis of your customer base.

Which criteria should you use to segment your customers?

There is no universal criterion. The choice depends on what the business is trying to understand or do.

Recency

How long has it been since the customer last bought? Recency helps distinguish recently active customers from those whose last purchase is getting old. It must, however, be interpreted according to the purchase cycle. A customer absent for 30 days may be unusual in a store visited every week and perfectly normal in a sector where purchases are spaced several months apart.

Purchase frequency

How many times does the customer buy over a given period? Frequency helps distinguish regular customers from occasional buyers. Here again, context matters: six purchases a year can be a high frequency for one business and a very low one for another.

Amount spent

How much revenue has the customer generated over the period studied? The period must always be specified. A ranking over the last three months won't necessarily give the same results as a ranking over the last two years.

Average basket

How much does the customer spend on average per transaction? Two customers can generate €1,000 in annual revenue with totally different behaviors:

• 20 purchases of €50;
• 2 purchases of €500.

The annual amount is the same, but the commercial relationship is not. See also increasing the average basket.

Products or categories purchased

Purchases also let you create segments based on observed interests. For example:

• regular buyers of a category;
• customers who have already bought a brand;
• customers who have never bought a given range;
• buyers of a product that needs periodic renewal.

This kind of segmentation can be particularly useful to avoid sending an offer to customers it doesn't suit.

Channel or location

When the data allows, it can also be worthwhile to distinguish behaviors by:

• physical store;
• online store;
• click & collect;
• location visited.

In a network, this dimension makes it possible, among other things, to identify customers who visit several points of sale.

Combining several criteria

The most useful segmentations rarely rely on a single criterion. Take:

Customers with no purchase in 90 days

This segment can include a former regular as well as a customer who made a single purchase three months ago. They probably don't have the same importance or need the same action. So you can refine it:

Customers who made at least 6 purchases over the last 12 months and haven't bought anything for 90 days.

Then possibly:

Customers who made at least 6 purchases over the last 12 months, with no purchase in 90 days and who spent more than €500 over the period.

Each additional criterion shrinks the population but increases the precision of the segment. You should, however, avoid the opposite excess: a segment so precise that it contains almost no one can lose its operational value.

The RFM method

A classic segmentation method combines three dimensions:

• Recency: date of the last purchase;
• Frequency: number of purchases;
• Monetary value: amount spent.

This is the RFM method. Customers are generally given a score on each of these dimensions, which then makes it possible to group profiles with similar behaviors. For example, a customer can be:

• very recent;
• very frequent;
• high in amount.

Another may show a high historical amount but low recency. RFM therefore helps distinguish behaviors that revenue alone wouldn't reveal.

It is not, however, required for every segmentation. For a campaign about a specific product category, the products purchased can be far more relevant than a full RFM score. The method must always answer the need, not the other way around. See also identifying your best customers.

Which segments can you create?

There is no universal list of segments to replicate in every business. A few groups can nevertheless serve as a starting point:

Segmenting doesn't mean freezing customers

A segment isn't necessarily permanent. A new customer can become a regular. A regular customer can start to drift away. An inactive customer can come back. Behavior evolves with each purchase.

A segmentation used to drive recurring actions must therefore be recalculated regularly from the most recent data. This is particularly important for segments based on recency or frequency.

How do you use segments?

Segmentation becomes useful when it genuinely changes the way you act.

It then also becomes a tool for managing the customer relationship.

Beware of segments that are too broad

“Inactive customers,” “good customers” or “loyal customers” may seem precise enough. They aren't always. A segment must be defined by measurable criteria. For example:

Loyal customers

is an ambiguous notion. Whereas:

Customers who made at least 10 purchases over the last 12 months

describes a verifiable rule. This doesn't mean ten purchases is universally the right definition of loyalty. The threshold must be chosen based on the business. A good segmentation relies on explicit, understandable criteria.

Beware of segments that are too small as well

Adding criteria improves precision, but also shrinks the size of the group. If you ask for:

customers who bought at least 12 times, spent more than €1,000, bought a specific category, visited a given store and haven't bought anything for exactly 60 days,

the resulting segment may become too narrow to be truly useful. You need to find a balance between targeting relevance and a workable group size.

How do you measure whether a segmentation is useful?

Creating segments isn't an end in itself. After an action, it is useful to measure what happened. Depending on the goal, you can look at:

• the return rate of the customers you followed up with;
• the revenue generated by the segment;
• the number of customers who made a new purchase;
• the use of an offer;
• the change in purchase frequency.

This also lets you check whether the criteria chosen were really relevant. A highly sophisticated segment isn't necessarily better than a simple one if it doesn't lead to better results.

Segmenting with AI

Creating a segment in a traditional tool generally requires defining several filters. With an AI able to query the business's data, the logic can be expressed directly in natural language. For example:

“Identify the customers who haven't bought anything for 90 days.”

Then:

“Keep only those who made at least 6 purchases over the previous 12 months.”

Then:

“Among them, keep those who spent more than €100 a month on average.”

The value is being able to build and refine the segment step by step without having to rephrase the whole query at each stage. The quality of the result nonetheless depends on the available data and on the precise definition of the criteria requested.

How can Gillia help?

Gillia lets you query the available customer data in natural language to build groups that match a specific need. For example:

“Which customers regularly buy this product family?”

The merchant can then refine:

“Keep those who bought at least three times over the last six months.”

Then use this group as the basis for a loyalty action or a targeted message. The value is being able to move step by step from intent to segment and then to action, while keeping a validation step before any external action. When the same segment needs to be checked regularly, the query can also become a recurring routine.

This is exactly where Gillia's role as an AI agent comes into its own.

Segmenting in a multi-location business

In a network, the scale of analysis must be chosen carefully. A customer can buy in several stores. If customer data is consolidated at the brand level, their behavior can be analyzed globally:

• total frequency;
• total amount;
• stores visited;
• categories purchased across the network.

It can also be relevant to segment by location to answer a local question. The two approaches aren't mutually exclusive. The scope of the segment simply needs to match the decision you are trying to make (see multi-store management). To go further, explore the Gillia use cases related to customer knowledge: create, edit or target a customer group or prepare a targeted action on a selection of customers.

A segmentation must stay useful

A good segmentation is neither the most complex nor the one with the most criteria. It must let you clearly answer three questions:

• Who belongs to the group?
• Why is this group different from the others?
• Which decision or action will you adapt thanks to this distinction?

If the third question has no answer, the segment is probably useless. Customer segmentation becomes truly valuable when it turns the available data into groups that are understandable, measurable and actionable.