You don't need tidy data before you start automating

Nearly every conversation about automating something ends in the same place: "we're interested, but first we need to sort our data out." Then six months go by. The spreadsheet still has three tabs that don't reconcile, the CRM is still half empty, and the quotes folder is still called quotes_2024_FINAL_v3. If you're waiting until everything is clean before you start, bad news: that Monday never arrives. The good news is that for the work you actually want off your plate, you don't need it.

Where the myth comes from, and why it sounds so sensible

"Garbage in, garbage out" belongs to an earlier era of computing, and in that era it was true. It came from dashboards, reports and forecasting models: if you wanted a system to tell you how much you'd sell in November, you needed three years of sales, consistently categorised, with no duplicates. One customer entered twice and the number was wrong. There, yes — data first.

What's changed is that an AI agent isn't doing that job. It isn't training on your history or averaging anything: it reads what's in front of it, in whatever language it's written in, and acts. It has far more in common with a new hire than with a report.

And you don't ask a new hire for a clean database before letting them pick up the phone. You tell them the opening hours, where the price list lives, what they must never promise, and who to ask when they get stuck. That's enough to be useful on day one, however chaotic your filing is. Agents are the same.

What an agent actually needs

Run through the three things required to get one working and notice that "tidy data" isn't among them:

• Somewhere to read the answer. Your price list, the services page, a PDF, a two-page document you write yourself.
• Rules about what it never does. What it won't promise, what it won't discount, what always comes through you.
• Somewhere to put the result. The WhatsApp thread itself, a row in a sheet, an email to your inbox.

The difference matters. Sorting out a company's data is a project: months of it, arguments about categories, someone wrestling with exports. Writing down the ten answers you repeat most is an afternoon. For 80% of what you want to hand over, the afternoon is enough.

Think of a garage answering "is my car ready?" thirty times a day. That doesn't need a pristine CRM or twelve years of migrated history. It needs whoever knows to record the vehicle's status in one fixed place. That's a field, not a migration.

Three things you could set up on Monday

1. Answering what you already answer twenty times a day. Hours, address, parking, whether you take a particular insurer, how long a quote takes. The data source is you and your website. Zero cleanup.

2. Sorting and routing what comes in. An agent that reads emails and messages and labels them: quote, complaint, supplier, invoice, noise. It needs none of your history, because the message it has to classify is right there. This is the most underrated one — it touches no data of yours and gives you back the first half hour of the day.

3. Chasing things that are already written down somewhere. Appointment reminders, overdue invoices, quotes that went quiet. Here you do need a reliable source, but you usually have one: the calendar you actually work from and the software you actually invoice with are reasonably tidy, precisely because you use them daily.

The only awkward bit in the third is when the diary lives in three places at once — the boss's Google Calendar, the notebook at reception, the salesperson's phone. That isn't fixed by tidying data. It's fixed by deciding which one wins: a ten-minute decision you've been postponing for two years.

The tidying happens afterwards, and it's free

Here's the part almost nobody sees coming: putting an agent to work is the best data audit you'll ever run, and it doesn't cost you a separate project.

The moment it starts, the mess surfaces on its own, with names attached. It can't quote a price because the PDF and the website disagree. It sends two reminders to the same customer because they're in the system under two phone numbers. It gets stuck on the same question every time because that answer was never written down anywhere. Within two weeks you have a concrete list of what's genuinely broken, ranked by how often it bites.

That list is worth far more than a general clean-up, because a general clean-up tidies everything equally — including the fields nobody ever opens and the thousand customers who haven't been back since 2019.

Tidying data you aren't using is tidying a cupboard nobody is going to open.

There's a useful side effect too. Because each fix comes from a real incident ("this went wrong today, with this customer"), the team actually does it. An abstract data-quality initiative dies in week three.

When the myth is right

It isn't always a myth, and that's worth saying plainly. Three situations call for tidying first:

When the agent acts on the data rather than just reading it. If it's confirming a dosage, telling someone whether a treatment is covered, or committing stock you don't have, a bad record stops being an annoyance and becomes an error that repeats reliably — faster, and to more people. An agent amplifies whatever you feed it. Where money, health or firm commitments are involved, verify the source first.

When there's no single version of the truth. If the same service is priced in three places at three amounts, that isn't a cleanliness problem. It's that nobody has decided which one is correct. Decide it before you automate, even if only for that one service.

When you're about to point an agent at a folder of old personal data. Before letting it read twelve years of client files, it pays to know what's in there. That isn't tidying — it's knowing what you're exposing, which is a GDPR conversation, not a spreadsheet one.

The short version: cleanliness matters enormously where the agent decides, and rather little where it just reads and passes the message on.

Start with what you already know by heart

The safest first task is the one you could answer with your eyes shut at three in the morning. If you know it by heart, your data is already organised — in your head — and getting it out is an afternoon of writing, not a six-month programme.

Try this in the next week: take every message and email that came in over seven days, count which ones repeat, and write the good answer to the top three. That's your source. Everything else — the spreadsheet, the half-empty CRM, the folder with the unspeakable filenames — can carry on waiting, because those three don't need it.

Our email manager almost always starts this way: reading what's already arriving and sorting it before you open the inbox. It doesn't need your CRM up to date. It needs you to tell it what counts as urgent. You approve, it executes.