How AI agents inherit your data — and why that matters

An AI agent works from your data, so its results are only as good as that data is. If your information is scattered, out of date or contradictory, the agent doesn't fix it — it acts on it, confidently and at scale. Which is why sorting your data and process comes before, not after, adding AI. The most common reason an AI project disappoints isn't the AI. It's the state of the data it inherited.

"Garbage in, confident garbage out"

The old data adage — garbage in, garbage out — is worse with AI agents, because they don't just report the bad data, they take action on it. An agent working from a stale customer list emails the wrong people. One working from contradictory stock figures makes the wrong call on an order. And it does so with exactly the same confidence it would if the data were perfect.

That confidence is the dangerous part. A person handed dodgy data often senses something's off and checks. An agent has no such instinct. It treats every input as true and proceeds, which means a small data problem can turn into a large operational one before anyone notices. The agent isn't malfunctioning — it's faithfully acting on what you gave it.

Why is scattered data the real problem?

Most businesses' data isn't wrong so much as fragmented. The same information lives in several tools and spreadsheets that quietly disagree — the booking system says one thing, the accounts say another, and a spreadsheet says a third. No single copy is obviously "the wrong one"; they've just drifted apart over time.

An agent asked to act has to pick a source. If the sources conflict, it can't know which is right — it just chooses one and runs. The fragmentation itself is what makes AI unreliable, not any single bad number. This is the same underlying issue that makes it hard to get a straight answer from your own reporting: when data is scattered, nothing downstream can be trusted, whether that's a dashboard or an agent.

Why does this matter for the AI hype?

The pitch is simple: "add AI and get results." The reality is that AI amplifies whatever's underneath. Good process and clean data, and it shines — fast, tireless, accurate. Mess, and it makes the mess faster and more confident. AI is a multiplier, and a multiplier applied to a negative number gives you a bigger negative.

That's why the businesses getting real value from AI are the ones that fixed the foundations first. They're not smarter about AI; they just did the unglamorous work of connecting and cleaning their data before pointing an agent at it. The same principle runs through when to use AI agents in your business: the technology is the last step, not the first.

The right sequence

• •Connect and clean your data into one accurate source of truth, so there's a single version an agent can rely on.
• •Fix the process so the workflow feeding the data is sound — see why automation fails when the process is broken.
• •Then add agents where they genuinely help, with a human checking early until you trust them.

Skip the first two and AI becomes a liability, not an asset. This is also why a connected system tends to make everything downstream easier: when the data is already together and already accurate, both your reporting and any AI you add have solid ground to stand on.

A worked example: the agent that emailed the wrong customers

A growing e-commerce business wants an AI agent to handle first-line customer replies — order status, delivery dates, simple questions. Sensible use of an agent, in principle. But the customer data lives in three places: the storefront, a shipping platform, and an old spreadsheet the founder still uses for wholesale accounts. They quietly disagree about addresses, order status and who's a current customer.

The agent goes live and starts confidently telling people their orders have shipped when the shipping platform says otherwise, because it pulled the status from the storefront. It emails delivery updates to addresses that were changed months ago in one system but not the others. It isn't broken — it's faithfully acting on whichever source it happened to read, and the sources don't agree. Every wrong reply lands with total confidence, which makes it worse, not better.

Roll the agent back, connect the three sources into one accurate record, and switch it on again — and suddenly it's genuinely useful. Same agent, same code; the only thing that changed was the ground it stood on. The AI was never the problem, and it was never going to be the fix. The data was both.

How do you know if your data is ready for AI?

A few plain questions tell you most of what you need. If you asked two different systems the same thing — a customer's current address, whether an invoice is paid, how much stock is on hand — would they agree? Is there one place you'd point to as the source of truth, or several that have drifted apart? When a number looks wrong today, how long does it take to work out which copy is right?

If those questions make you wince, your data isn't ready for an agent yet — and that's useful to know before you spend money, not after. The good news is that the work to get ready isn't AI work at all. It's connecting your systems so information lives in one accurate place and flows without re-keying. Do that, and you don't just unlock AI; you make your everyday reporting trustworthy too. The foundations pay off whether or not you ever add an agent on top.

Related reading

• •Pillar: Business process automation: a practical guide
• •What should stay human: the limits of automation
• •How to automate without creating a new mess

The honest first step

Whether your data is ready for AI comes down to how connected and trustworthy it is today — which starts with understanding where your information lives and where it disagrees. That mapping is the prerequisite for any useful automation, agent-based or not. Sort the foundations, and AI has something solid to work from; skip them, and it just amplifies the problem.

Tell us what you run, and we'll show you what's possible.