Here’s the reassuring bit first: nearly every small team and not-for-profit I’ve looked after has customer data that’s a genuine mess — duplicates, half-filled records, three spreadsheets and a shoebox of business cards — and it is completely normal. You don’t fix it with a giant six-month migration that nobody has time for. You fix it humanely: in small passes, in the order that helps your people most, and without pretending it’ll ever be perfect.
Why the mess happens (and why it’s not your fault)
Data gets messy because real work is messy. Someone’s away, so a record gets half-entered. A new system gets bolted on, so now the same customer exists in two places. A volunteer leaves and takes the only copy of how the list was organised. None of that is incompetence — it’s what happens when busy people keep the lights on with the tools they’ve got.
The mistake I see is treating it as a moral failing and launching a heroic clean-up that burns everyone out and gets abandoned at 40%. There’s a calmer way.
The humane fix, in order
- Start with the data you actually touch. Don’t clean the whole database. Clean the 200 records you use every week. The dead weight at the bottom can wait — or stay buried forever, and that’s fine.
- Fix the entry, not just the history. If new records keep arriving messy, tidying the old ones is bailing a leaky boat. A simple form with the right required fields stops the mess at the source.
- De-duplicate before you do anything clever. This is the one job where AI genuinely shines: spotting that “J. Smith”, “Jane Smith” and “jane.smith@” are the same person, and flagging it for a human to confirm. Flagging — not auto-merging. You decide.
- Decide what you’re allowed to keep. Old personal data you no longer need isn’t an asset, it’s a liability. Part of cleaning up is honestly deleting what shouldn’t still be there.
Clean the data you use weekly. Stop the leak at the form. Let AI flag the duplicates and let a human make the call.
Where AI helps — and where it doesn’t
AI is great at the boring detection work: matching near-duplicates, spotting a phone number in the email field, standardising “St” and “Street”. It’s a tireless second pair of eyes across thousands of rows.
What it must not do is decide on your behalf. Auto-merging two records that look similar but aren’t — say, a parent and child at the same address — is exactly how you end up sending the wrong person the wrong letter. The pattern that works is the same one we use everywhere: AI proposes, a human disposes. The machine does the looking; the person does the deciding.
And the hard rule from us: sensitive personal information doesn’t go through a general AI tool. Cleaning a community organisation’s client list is not a “paste it into a chatbot” job. It needs an approach that keeps the data where it should be — ideally on Australian infrastructure you actually control.
A realistic first step
If your data is daunting, don’t plan the perfect system. Pick one list — the one your team curses at most — and do a single, scoped pass: de-dupe it, fix the live records, and patch the form that keeps refilling it with rubbish. That one pass usually buys back more time than a year of good intentions, and it tells you whether a bigger tidy-up is even worth it.
If you want a second opinion on what’s safe to automate and what should stay hands-on for your particular data, that’s a good thing to map out together — and we’ll happily tell you if the honest answer is “leave it alone for now.”