CRM implementation12 minAugust 2026

    CRM Data Migration: Clean Data In Your New CRM

    Dirty data is the quiet killer of a CRM rollout. How to scope the migration, deduplicate before you import, and decide which history is actually worth moving.

    Migration is the part of an implementation that people plan least and regret most. It looks like a technical step, so it gets treated like one, and it gets scheduled near the end when the calendar is already tight. Then the export turns out to hold three versions of your biggest customer, and suddenly the whole project is about data. We have done this often enough to say plainly: the migration is where trust in the new system is won or lost, and it deserves to sit at the front of the plan rather than the back.

    01

    Why dirty data breaks more than reports

    The obvious cost of bad data is a wrong number in a dashboard. That is not the expensive part.

    The expensive part is what happens next. Someone spots the wrong number, mentions it, and the room quietly decides the system cannot be trusted. From then on people keep their own list on the side, because their own list is right. Now you have two sources of truth and you are paying for one of them.

    That loop is very hard to reverse. Cleaning the data afterwards fixes the numbers but not the habit. This is why we treat migration as an adoption problem wearing a technical costume.

    You get one first impression of a new CRM. If the first report someone opens is visibly wrong, you spend the next year arguing with a spreadsheet you thought you had replaced.

    02

    Scope the migration before you touch anything

    The first decision is not how to move the data. It is how much of it deserves to move.

    Most teams default to everything, because everything feels safe. It is not safe. Every record you carry across is a record someone might one day rely on, and old records are exactly the ones nobody can verify anymore.

    A useful rule: migrate what you will act on, archive what you might need to look up, and delete what you would be embarrassed to have kept.

    • Open deals and their full history: migrate. This is the operational core.
    • Active customers and their contacts: migrate, after deduplication.
    • Closed won deals: migrate the outcome and the value, usually not every note.
    • Closed lost deals older than two years: archive rather than migrate, unless you actively re-approach them.
    • Contacts with no activity in three years and no email: leave them in the archive.
    • Custom fields nobody has filled in for a year: do not recreate them in the new system.
    03

    Count before you clean

    Before deciding how to fix the data, find out how bad it is. This takes an afternoon and it changes every other decision in the project.

    Export everything to a spreadsheet and count. Not impressions, actual counts. The numbers are usually worse than the room expects, and that is exactly why having them early is so useful.

    • How many organisations share a normalised name, ignoring legal suffixes and case?
    • How many contacts share an email address, and how many have no email at all?
    • How many organisations share a domain but have different names?
    • How many records have no owner, or an owner who has left the company?
    • How many deals sit in a stage nobody has used in the last six months?

    In practice, the duplicate rate in a CRM that has been running for a few years without governance sits high enough to surprise everyone in the room. Measure yours. Do not assume you are the exception.

    04

    Deduplicate before the import, never after

    Merging duplicates in a spreadsheet is tedious. Merging them inside a live CRM, after your team has started attaching activity to both versions, is genuinely painful and sometimes irreversible.

    So the order matters more than the method. Clean in the export, import once, and keep a deduplication check running afterwards so the problem does not grow back.

    Matching on company name alone is the classic mistake. Names differ by legal form, by spelling, by whether someone typed the trade name. Domain is a much stronger signal, with name as a secondary check.

    • Match organisations on email domain first, then on a normalised name as a second pass.
    • Match people on email address, and treat name matches as suggestions that a human confirms.
    • Decide in advance which record wins a merge: usually the one with the most recent real activity.
    • Never auto merge on a fuzzy match without review. Two similar names can be two real companies.
    • Keep a log of what was merged. You will be asked about one specific customer within a month.

    We built our own deduplication tooling because doing this by hand at scale is where migrations quietly go wrong. Whatever tool you use, the principle is the same: decide the matching rule before you run it, not while you are looking at the results.

    05

    Import in a way you can undo

    Treat the first import as a rehearsal, always. Even when the data looks clean, especially when the data looks clean.

    The point of the rehearsal is not to check whether the import runs. It is to look at real records with your own eyes and notice the things no validation rule catches.

    • Import into a test environment or a separate sandbox first.
    • Pick ten records you personally know and check them field by field.
    • Check the boring things: currency, date format, decimal separator, country codes.
    • Verify relationships, not just records. Is the contact attached to the right organisation?
    • Import in a fixed order: organisations, then people, then deals, then activities.
    • Keep the source export untouched and findable for at least six months.
    06

    Keeping it clean after go live

    A migration gives you one clean moment. Without a rule, the data starts degrading again from week one.

    The degradation is rarely dramatic. It is a rep creating a new organisation because search did not find the existing one, repeated a few hundred times over a year.

    • Make the search behaviour good enough that creating a duplicate is harder than finding the original.
    • Run a duplicate check monthly rather than annually. Small batches stay reviewable.
    • Give one person the authority to say no to new fields. Field sprawl is data decay in slow motion.
    • Fix the intake, not just the database. Most duplicates enter through forms and imports, not by hand.

    Frequently asked questions