Data quality is the real AI bottleneck for UK firms — not the model

Author: StJohn Krog — LinkedIn

UK businesses got wider with AI faster than they got deeper. The Office for National Statistics reports that about 35% of firms with 10+ employees used at least one AI technology by mid-2026, up from roughly 12% in late 2023 — yet the average number of AI technologies per adopting firm only moved from about 1.4 to 1.6, and only around 10% say they use AI extensively across operations.

That gap is often blamed on skills or budget. Increasingly the constraint underneath both is simpler: the data and connections AI needs to act on are still messy, while people quietly fix the mess as they work.

What the research is saying in 2026

Several independent signals line up:

Klarus surveyed mid-market decision makers in the UK and Ireland (published July 2026). Among organisations that had piloted or deployed AI, 83% experience poor data quality and 69% say it is preventing or delaying AI activity. Where projects met expectations, strong data quality was the top success factor (59%). Only 10% had successfully scaled all initiatives beyond pilot.
Dayshape research among UK professional services leaders put poor data quality as the top barrier to effective AI adoption (34%), ahead of system integration (32%), cost (28%) and skills (22%).
UKISUG / Syniti work with SAP user organisations found just 4% “very confident” in data quality, 94% saying silos block real-time decisions, and 89% saying data challenges will slow AI adoption.
Commentary synthesising ONS trends (UK Tech News, Aug 2026) makes the mechanism clear: humans have been correcting customer details, reconciling duplicates and interpreting ambiguous emails without writing those fixes back into systems of record. Copilots hide the gap; agents expose it.

DSIT’s AI Adoption Research adds the commercial punchline: 75% of AI-using firms report workforce productivity gains, but 77% report no revenue change. Shallow tools on shallow data feel productive and leave the P&L untouched. Among barriers, 21% of firms cited data complexity (lack of data or unstructured data); among those who named it, 70% rated it as a significant barrier.

Why “clean everything” is the wrong plan

Waiting for a perfect data lake before you touch AI is how preparation becomes a multi-year programme. Few SMEs will ever reach “fully clean and connected.” You do not need that.

You need use-case-grade data: the fields, documents and identities that matter for the one workflow you want AI to help with — reliable enough that an agent (or a human using a copilot) is not guessing.

Ask three questions:

Which process are we automating or assisting?
Which information does that process depend on?
Where do people currently compensate — re-keying, spreadsheet side-channels, “ask Sarah”, duplicate customer records?

Those compensation points are your backlog. Fix them in priority order for that workflow. Leave the rest alone until the next workflow.

What failure looks like when agents arrive

A person chasing a supplier query can notice that the address in the ERP disagrees with the CRM, stop, and ask. An agent without an explicit rule may carry the wrong address into the next step — and write it somewhere else as if it were fact.

The same pattern shows up in ecommerce (stock vs channel listings), professional services (client matter IDs across time and billing), and field service (job notes vs scheduled assets). The model is rarely the bottleneck. The handoffs between systems and the unstructured residue in email and PDFs are.

A practical 30-day data-for-AI sprint

Week 1 — Map one workflow end to end

Lead-to-cash, order-to-fulfilment, or ticket-to-resolution. List every system, spreadsheet and human “glue” step. Mark where AI would act.

Week 2 — Define the minimum viable truth

For that workflow only: required fields, allowed sources, duplicate rules, “stop and ask a human” conditions. Delete or quarantine fields nobody trusts.

Week 3 — Connect two systems properly

Usually CRM ↔ finance, or commerce ↔ stock. One durable integration beats five CSV exports. Capture corrections back into the system of record.

Week 4 — Pilot AI on the cleaned slice

Shadow mode first. Measure handoff rate, error/rework, cycle time. Every agent failure that traces to data becomes a ticket on the data backlog — not a reason to buy a bigger model.

Connected systems are the other half of the same problem

AI layered on disconnected SaaS multiplies copy-paste. The firms seeing returns treat integration and data quality as AI readiness, not as a separate “IT project someday.” Visibility before clever automation remains the rule: automate a mess and you get a faster mess.

You do not need an enterprise MDM programme. You need:

a named owner for customer / product / supplier master data for the chosen workflow
one ID that travels across the two systems you connected
a weekly 30-minute data clinic on exceptions the AI (or staff) could not resolve

Klarus respondents already recognise the priority: improving data quality was a joint top plan for the next 12 months (43%), alongside stronger guardrails. That is the right sequence — foundations before scale.

Practical takeaway

UK AI adoption has broadened. Depth still waits on data people used to fix by hand. Do not boil the ocean. Pick the workflow, identify the information it depends on, remove the human glue that never gets written back, connect the two systems that matter, then let AI operate on that slice. The model will keep improving. Your bottleneck is whether the business truth underneath it is fit for purpose.

Clarity Growth specialises in connected digital estates for UK growing businesses — websites, ecommerce, CRM, automation and the integrations that make AI usable. We help clients prioritise use-case-grade data and system connections before scaling copilots or agents, so pilots survive contact with real operations instead of stalling on silos and spreadsheet glue.

Hero — “Systems of record”

Ops lead reviewing a single connected dashboard on a laptop, or a simple visual of CRM / commerce / finance tiles with clean arrows. Mood: practical, not sci-fi.

Search: Unsplash/Pexels — “business dashboard analytics laptop”, “operations manager warehouse laptop”.

Credit: Unsplash / Pexels + alt: “Operations lead reviewing a connected business dashboard on a laptop.”

Why this matters for Clarity Growth

Sources

Start with the friction. Build the capability.

Clarity Growth helps organisations identify, design and implement practical automation opportunities across existing systems and workflows.

Discuss an automation opportunity