One customer, four records: the duplicate problem nobody budgets for
A duplicate customer record is what happens when one buyer reaches you more than one way and your system does not realise it is the same person. In home & garden sales this is not an edge case - it is the normal condition, because a considered purchase naturally involves several contacts over several weeks through several channels. The cost is not untidiness. It is a pipeline that overstates itself, a team that looks disorganised in front of customers, and hot leads hidden inside three lukewarm ones.
How one person becomes four records
Follow a realistic buyer for a garden room.
Saturday, she uses the configurator on your site and downloads the PDF, leaving a personal Gmail address and the name Kate. On Tuesday her partner phones with a question about groundworks, leaving a mobile number and a different surname. On Thursday she messages your Instagram from an account with a handle and no real name. Two weeks later she submits a quote request from work, as Katherine, with a company email and the same mobile.
Four touchpoints. One buyer, one project, one budget. A classic CRM shows four leads - or, worse, three leads and one orphaned Instagram thread nobody ever connected to anything.
What it actually costs
Your pipeline is wrong in a specific direction. Duplicates always inflate. You count four opportunities where there is one, so forecast is overstated and conversion rate is understated - the same buyer appears as three losses and one win. Anyone making decisions on those numbers is making them on fiction.
Your team looks careless at the worst moment. The most common way this surfaces is two people from your company replying to the same buyer with different answers, or a follow-up email that asks something she already answered in the chat. For a €12,000 considered purchase, “they don’t seem to know what’s going on” is an expensive impression to leave.
Your best lead is invisible. This is the one that costs real money. Kate is your hottest prospect precisely because she has engaged four times across four channels. But split across four records, she looks like four unremarkable enquiries. The signal that should have moved her to the top of the queue is exactly the signal that got destroyed.
Deduplication becomes somebody’s Friday. Manual merging is a chore that gets deferred, so the data degrades until the CRM stops being trusted, at which point people start keeping their own spreadsheets and you have a worse problem.
Why “just check before you create” does not work
The usual advice is process: search before you create a record. It fails for three reasons.
Most enquiries do not arrive by hand any more - a webhook creates them, and a webhook does not search first. The information you would search on is often absent or different (a different email, a nickname, no name at all on social). And the person best placed to notice is a salesperson in the middle of doing something else.
This is a data problem, not a discipline problem, and discipline is the wrong tool for it.
Resolving identity properly
The fix is to match every incoming signal against existing records automatically, with an explicit confidence level, and to be honest about the middle ground.
Strong signals stand on their own. A matching email address is near-certain. A matching phone number is very strong - shared handsets exist but are rare in this context.
Medium signals are suggestive, not conclusive. The same configurator session ID, or a link back to the same saved design, is good evidence. A shared company domain says these two people work together, which matters for a trade enquiry and means nothing for @gmail.com.
Weak signals are worth surfacing but never worth auto-merging. Same surname. Same postcode. Similar name spelled differently.
Then the important part: do not force a binary decision. Above a high confidence threshold, merge automatically. Below it, do not silently create a duplicate or silently merge - attach the candidate matches to the record, ranked, and let a person confirm in one click. Ambiguity is real, and a system that pretends otherwise will either fragment your data or corrupt it.
The other half is making merges reversible. Every automatic merge should be undoable, because the day you merge two genuinely different customers is the day people stop trusting the whole system.
What good looks like
You know it is working when opening a customer shows you one timeline: the configurator session, the phone call with its transcript, the Instagram exchange, the quote request - in order, with one engagement score across all of it, and no one on your team wondering whether there is another record somewhere with the rest of the story.
That is not a reporting nicety. It is the difference between Kate arriving in your morning queue at the top, flagged as a buyer who has engaged four times in three weeks, and Kate sitting in four places looking like nobody in particular.
The short version
- Duplicates are the default in considered sales, because buyers reach you repeatedly and inconsistently.
- They inflate your pipeline, embarrass your team, and hide your hottest leads.
- Process discipline cannot fix it - most records are created by integrations, not people.
- Match on confidence tiers, auto-merge only when confident, surface ranked candidates otherwise, and make every merge reversible.
Identity resolution is part of what Anders does - see also what an AI CRM is.