Poor data isn't a showstopper for AI

Many SME executives have reached the same conclusion after an honest audit of their systems: the data is fragmented, inconsistent, and spread across too many places. The logical next move feels obvious. Fix the data first, then think about AI.

It is a reasonable position. The research seems to support it. But the conclusion it leads to is costing businesses months of lost progress.

The evidence tells a more nuanced story. Many of the AI applications that deliver the fastest, most practical gains for SMEs do not depend on spotless, structured data at all. The real skill is knowing which use cases demand data quality and which do not.

Why so many leaders believe this myth

The concern is not unfounded. Recent UK polling found that 34% of professional services leaders name poor data quality as the single biggest barrier to effective AI adoption. It ranks ahead of integration costs and skills gaps.

Separate research linked to ServiceNow found that 73% of UK executives cite inadequate data accuracy, access and management as a major obstacle to AI rollout.

Those are striking numbers. They reflect genuine operational reality for many businesses. Data issues are real. Fragmented systems are real. The problem is the conclusion many leaders draw from them.

The myth is not that data quality matters. It does. The myth is that imperfect data blocks AI across the board, making it sensible to wait for a full data overhaul before starting.

What the evidence actually shows

AI applications do not all carry the same data requirements. Some use cases genuinely demand high-quality, structured data. Others thrive on unstructured text and existing process flows, where modest governance is enough to get started.

Data-heavy use cases include financial forecasting, demand planning, predictive analytics, and any application where the AI is drawing statistical conclusions from historical records. These do need clean, well-labelled, consistent data. A poorly structured dataset will produce unreliable outputs, and investing in data quality here is time well spent.

Data-light use cases are a different matter entirely. Email triage, document drafting, meeting summarisation, invoice routing, and workflow automation between existing tools all operate primarily on unstructured text. They do not require a unified data warehouse or a multi-year data migration.

An email is already structured enough for an AI to read, categorise, and draft a response. A Word document is already structured enough for an AI to summarise or redraft. An incoming invoice already contains the fields an automation needs to route it correctly. None of these use cases is waiting on your data estate to be rebuilt.

The cost of waiting for IT managers and SME executives

For IT managers and SME executives, the practical cost of the myth is not theoretical. It shows up in the decisions that get deferred.

Email triage that still consumes an hour of a senior manager's morning. Document drafting that still takes a junior team member most of an afternoon. Approval workflows that still move by forwarding emails and chasing replies. These are not data problems. They are process problems that AI can address today.

If SMEs treat imperfect data as a blanket veto on AI, they postpone gains in precisely the areas where AI delivers fastest. The businesses that do move forward during this period will embed working habits, institutional knowledge, and process efficiency that take time to replicate.

There is also a subtler cost. Early AI pilots surface the specific data issues that actually matter. Attempting a full data overhaul before any AI deployment means spending budget on problems that may never affect the use cases you end up prioritising.

Splitting use cases: a practical starting point

The intervention that consistently works is segmentation. Not all-or-nothing, but a deliberate split between what can start now and what requires data investment first.

A simple approach for SME executives and IT managers to apply:

1. List the AI applications your team is considering or has been asked to evaluate.

2. For each one, ask a single question: does this application draw statistical conclusions from structured historical records, or does it process existing text and trigger actions in connected tools?

3. If the latter, it is likely data-light. Pilot it now. Use what you learn to identify the specific data gaps that would improve it further.

4. If the former, scope the data work required before committing to the application.

This approach replaces a binary decision (do AI or fix data first) with a sequenced one. Near-term wins fund the case for longer-term data investment, and early pilots generate the evidence needed to prioritise that investment accurately.

Considerations and realistic limitations

Honesty matters here. Not every data-light use case will run without any governance work. Where automation involves personal data, including customer emails or employee records, UK GDPR obligations apply regardless of how light the technical data requirements are.

The ICO's guidance for SMEs on AI preparation recommends a basic data protection impact assessment before launching any AI pilot that touches personal data. The purpose limitation and data accuracy principles under UK GDPR do not disappear because the use case is operationally simple. This is not a reason to delay indefinitely. It is a reason to build a two-page DPIA into your pilot plan rather than skipping it.

It is also worth being clear that data-light does not mean data-free. Automations that connect your email platform to your document management system still need someone to check that the right data is flowing to the right place. The governance bar is lower. It is not zero.

If you want practical support with this, gecco's Automations service connects your existing tools using no-code workflows, helping SME teams start with email triage, document routing, and approval workflows without waiting for full data modernisation.

Your next step

If your leadership team is still treating data quality as a blanket reason to hold back on AI, that is exactly the kind of readiness gap the AI Readiness survey is built to surface.

Take the AI Readiness survey. You will get access to 65+ free resources and a custom AI Readiness report. We then offer a free 45-minute AI Readiness call to walk through your results.

Most SME AI wins do not require pristine data. The real blocker is misunderstanding which use cases demand structured data and which do not. If you want help identifying where your quickest wins sit, gecco's Automations service is built for exactly that.

Website · LinkedIn · Case Studies · Newsletter