What customer data a small business actually needs

A small business can run retention on five things per customer: how to reach them, whether they allowed it, when they bought, what they bought, and when they should be back. Everything else can wait. The mistake I see most is the opposite one: a long list of fields, half of them empty and none of them used.

What is the minimum data per customer?

Five fields.

1. A name and one way to reach them, a phone number or an email you are allowed to use.
2. Permission. Which channels they agreed to, and whether they asked you to stop.
3. Purchase dates. Every visit or order, not only the last.
4. What they bought and how much they paid.
5. Next due. The date they should be back, from what retention is.

That list looks short on purpose. When I describe what a casino retention system needs, the answer is four tables keyed by player: the players, their deposits, their bets and their sessions. Names, phones and emails are optional and only matter if messages go through the system. The same logic holds for a shop. Who, what they paid and when, and what they did.

Why does permission belong in the data?

Because it decides what you may do, customer by customer, and it changes. A customer may have agreed to email and not to text messages. A customer may have asked you to stop last week. If your list does not say so, the first person to send a campaign will message someone who should not be messaged.

Check permission at the moment of sending, not only when you build the list. The state can change between the two. In casinos this is a hard gate for people who excluded themselves, and a sound habit for everyone else. Rules differ by country and by channel, and I am not a lawyer, so ask whoever advises you which channels are open to you in which market.

What does it cost to skip the data?

Here is the quick sum that convinces most owners. Open your list and count one group: people who bought at least twice and have not been back for twice their usual gap.

Say there are 400 of them and an average visit or order is $90. If one in ten comes back after a personal reminder, that is 40 people and $3,600 from a list you already own, before a single ad. These are example numbers.

The 400 is the easy part. The hard part is how many of them you can reach today, with a phone number or email you are allowed to use. In the businesses I have worked with, that is where most plans stall. If half your list has no usable contact, the first campaign is not a message, it is a clean-up.

How do you clean the data?

Start with the records that do the most damage.

• Test and internal records. Your own address, your staff, the order you placed to check the checkout. They distort every count. Casino data has test, service and bot accounts, and a handful of them can skew a whole analysis. I explain why in the engineering write-up on per-player scoring architecture.
• Duplicates. One person under two names or two emails splits their history, so they look like two new customers instead of one regular.
• Missing dates. An order with no date cannot be used for anything. Fill what you can and flag the rest.
• Wrong contact details. Send a plain message to a sample and see what bounces.

Do not clean everything. Clean the last twelve months and the top tenth of customers first. That is where a mistake costs the most.

How much history do you need?

For a manual process, a year is plenty, and even three months helps you start. For a prediction model, the sources I rely on ask for at least six months of reliable history, because the model needs examples of customers who stayed and customers who left. I come back to that in what machine learning does in retention. If you do not have it yet, start recording now. History is the one asset you cannot buy later.

What data can you skip?

Most of it, for now. Date of birth, favourite colour, how they heard of you, a dozen survey answers. These are fine to collect if the answer will change what you do. If it will not, you are paying in effort and in customer patience for a field nobody reads.

A good test before adding a field: name the message or decision it will change. If you cannot, skip it.

Where should the data live?

A spreadsheet is fine to start, as long as one person owns it and one copy is the real one. Move to a booking system, a CRM or a payment platform's customer records when the sheet becomes hard to keep correct. The tool matters less than having five fields filled in for the customers who matter.

If you later connect your data to a prediction tool, you will need someone to describe which table has the transactions and which field means what. For self-written systems this is often the longest part of the work. How to find the players about to leave your online casino shows what that data looks like in one industry.

What can you do today?

Open your list and count three things: customers in the last twelve months, how many have a phone number or email you may use, and how many have at least two purchases with dates.

Divide the second by the first. That share is the size of what you can reach tomorrow. If it is under half, fixing contact details and permissions is your first campaign. Block one afternoon this week to do it, starting with the top tenth by spend.