One person, three records: one from the phone, one from a booking, one from the loyalty sign-up. Each shows two visits, so nobody sees the six. Your best customer is invisible because they are filed under three different names.

Somebody has been coming for eighteen months. They are, by spend, one of your best customers. Nobody at the venue knows this, and neither does any report you have ever looked at.

The reason is that they exist three times. Once as a mobile number from a phone booking, once as a name and email from a form, once as a first name and a table from a Saturday when somebody typed what they heard. Each record shows two or three visits, which is unremarkable. Nobody has ever seen the eight.

How the same person becomes three people

Four routes, and every venue has all four running at once.

The channel decides the identifier

⇒ ⭕ A phone booking captures a number. A web form captures an email. A walk-in captures a name. These do not match each other, and nothing in the process ever tries.

⇒ The Phone Is Your Booking System, Whether You Built One Or Not

The name is typed by ear

⚠️ ⭕ In a loud room, at speed, by different people. One spelling on Friday and another on Saturday is not carelessness; it is what happens when a name is heard rather than read.

The number is entered differently

⇒ ⭕ With a country code, without, with spaces, with a leading zero. The same phone becomes several strings, and an exact-match lookup finds none of the others.

And nobody is looking at the moment it happens

⚠️ ⭕ The person taking the booking cannot see the other records and would not have time to check. The duplicate is created by somebody doing their job properly.

What it actually breaks

This is the part worth being specific about, because "clean data" is not an argument.

Your best customers are invisible

⭕ Split across three records, a top spender looks like three ordinary ones. Every "who are our best customers" list is wrong in the same direction, and always by leaving out the best.

⇒ How Many Regulars Do You Actually Have

The regular count is inflated and the value is deflated

⚠️ ⭕ More records, each worth less. Both halves of the mistake push you toward the same wrong conclusion: that you have many shallow customers rather than fewer deep ones.

⇒ What a Regular Is Actually Worth

You cannot see somebody stop coming

⇒ ⭕ A guest who lapses across three records lapses invisibly, because no single record was ever frequent enough for the gap to look wrong.

⇒ The Regular You Lost Without Noticing

The notes are in the wrong place

⚠️ ⭕ The allergy, the preference, the thing not to mention — recorded once, attached to the record nobody opened tonight. The information exists and does not arrive.

And they get contacted as a stranger

⛔ ⚠️ Somebody who asked not to be contacted on one record can be contacted through another. That is a data protection matter, not a tidiness one — ask what applies to you.

Finding out how bad it is

Twenty minutes, and the answer is usually larger than anybody expects.

Count records with the same phone digits

⇒ ⭕ Strip everything that is not a digit, take the last nine, and group. This one comparison finds the majority of duplicates in most systems.

Count records with the same email

⭕ Lower case, trimmed. ⚠️ Fewer matches than the phone method but almost no false positives — this is the safest one to act on first.

Look for near-identical names on the same nights

⚠️ ⭕ Two similar names on the same table on the same date is one person. This finds the ear-typed ones that the other two methods miss.

And check the ones with exactly one visit

⇒ ⭕ A large population of single-visit records is the signature of this problem, not of a venue that nobody returns to. Most venues read it as the second thing.

⇒ The Guest List You Have Never Cleaned

Merging without losing anything

The danger of cleaning up is doing damage. Four rules make it safe.

Keep every visit, always

⚠️ ⭕ The whole point is the combined history. A merge that keeps one record's visits and discards the other's has made the data worse while appearing to improve it.

Keep both contact details

⇒ ⭕ The phone from one and the email from the other. They are both real and both were used.

Keep the notes from both, with their dates

⭕ ⚠️ Notes conflict. Two records may say different things about the same person, and which is current depends on when each was written — so keep the dates rather than choosing.

And keep a record of what was merged

⛔ ⚠️ Merges are difficult to undo. Before merging anything, know how you would reverse it — and if the answer is "we could not", merge far more cautiously.

⇒ The Report You Open Every Monday

Stopping it happening again

Cleaning up once and changing nothing means doing it again next year.

Search before you create

⭕ The single highest-value habit, and it takes three seconds. By phone digits, not by name — names are the least reliable field you have.

Standardise the phone number at entry

⇒ ⭕ Strip spaces, handle the leading zero, store one format. This one change prevents more duplicates than everything else combined.

Decide which field is the identity

⚠️ ⭕ Pick one — usually the phone. Where there is no agreed identifier, every channel invents its own and you are back where you started.

Make the search fast enough to actually use

⇒ ⭕ If looking somebody up takes longer than typing a new record, staff will type a new record, correctly, because they are serving somebody.

And review it quarterly rather than never

⇒ The Loyalty Scheme Nobody Uses

The cases worth being careful with

Four situations where merging is the wrong instinct.

Two people who share a number

⚠️ ⭕ A couple, a household, a colleague who booked. The phone matches and they are two people, which is why a match should be reviewed rather than applied automatically.

The company and the individual

⇒ The House Account Nobody Closes

Somebody who deliberately keeps them separate

⛔ ⚠️ Some guests have a reason for using different details in different contexts, and in this industry that reason is often serious. Merging without thought can expose something the guest was managing carefully.

⭕ This alone is a reason to review matches rather than merge in bulk.

And anybody who has asked not to be contacted

⛔ Whatever else you do, that request has to survive the merge. ⚠️ How to handle it is a data protection question — ask before you design the process.

What good looks like

The duplicate test

Take your top-spending guest and search for their phone digits. ⇒ One record means this is under control. Two or three means every report you have is wrong.

The single-visit test

⚠️ Look at what proportion of records have exactly one visit. A very high number is usually duplicates, not churn.

The entry test

Working     — the booking screen searches by phone before creating anything
Not working — a new record is the fastest path, so it is the one that gets used

Three to hold

1  Each channel captures a different identifier and nothing reconciles them
2  Your best customer looks like three ordinary ones — the error always hides the top
3  Search by phone digits before creating; standardise the format at entry

Where the record has to sit

One guest, one record, with every visit, both contact details, the notes from all sources with their dates, and any contact preference intact. Kept there, your best customer is visible, a lapse is noticeable, and the note written in March arrives at the table in September. Kept as three, the venue has an accurate list of people who came twice, no idea who its most valuable guests are, and a churn figure that is measuring its own filing.