Is your data ready for AI? What a UK SME actually needs, and what it does not.
By Zain M · 1 October 2026 · 11 min read
Data readiness for a UK SME does not mean a data warehouse. It means four things: knowing which system is the record of truth for each kind of record; that the records an automation needs are in that system rather than in inboxes and spreadsheets; that the documents an AI will read are available as files with enough consistency to be checked; and that the personal data involved has a lawful basis and a retention period. Gartner names poor data quality first among the causes of abandoned AI projects, and the ONS and DSIT surveys show UK firms training their own staff to fill the gap. A one-day assessment, set out below, tells a firm whether the first automation can start now or needs a month of tidying first.
What the evidence says about data and AI failure
Gartner’s four causes of generative AI project abandonment begin with poor data quality. In the July 2026 survey of recruitment agency leaders we cite in our sector guide, 74 per cent named poor data and disconnected CRMs as a top blocker. The ONS’s 2026 article on AI in UK businesses and DSIT’s adoption research both describe a skills gap that firms are filling by training existing staff, and much of what those staff are being trained to do is get the firm’s own records into a usable state. The pattern is that the model is rarely the problem; the inputs are.
The good news for a small firm is that the bar is lower than the phrase “data readiness” suggests. An SME does not need a data platform. It needs to know where its records are and to stop keeping them in three places.
The four questions
Run these for the one process you intend to automate first, not for the whole firm.
What “good enough” looks like for the common first automations
Document intake, invoices, applications, referrals, needs the documents as files and a system to write into; it tolerates variety because the AI reads it and a person checks the exceptions. Enquiry and booking automation needs a calendar the system can write to and a record of who was contacted; it tolerates a poor CRM because it creates the record. Chasing and reminders need one list of who to chase and the channel to reach them, which usually means fixing duplicate contacts first. A knowledge assistant that answers staff or customer questions from your own documents needs those documents in one place, current, and with the out-of-date ones removed, which is the tidying job firms most often underestimate.
| First automation | Needs | Tolerates | Fix first |
|---|---|---|---|
| Document intake | Files, a system to write to, 50 checked examples | Layout variety | Photos of screens; documents kept only in email |
| Enquiry and booking | A writable calendar, a contact record | A weak CRM | Two calendars for one person |
| Chasing and reminders | One contact list, a channel | Missing fields | Duplicate contacts, dead numbers |
| Knowledge assistant | Documents in one place, current | Imperfect formatting | Out-of-date policies still in the folder |
A month of tidying, if it is needed
When the four questions fail, the fix is usually a month of unglamorous work rather than a project: pick the record of truth and stop updating the other copies; merge duplicates; move documents out of inboxes into the system; delete the folder of superseded policies; and write down the lawful basis and retention for each data category. It is work the firm can do itself, and it pays back whether or not any AI follows, because it removes the retyping that was the cost in the first place. Our readiness audit does the assessment and the plan for firms that want it done for them.
Common questions
What does data readiness mean for a small business?
Knowing which system is the record of truth for each kind of record, having the inputs reachable by software, having a sample of real documents with known answers to check the AI against, and having the personal data’s lawful basis and retention written down. Not a data warehouse.
Do we need clean data before using AI?
You need consistent enough data to check the AI’s output, not perfect data. Fifty real examples with known answers matter more than a clean dataset.
Why do AI projects fail on data?
Gartner lists poor data quality first among causes of abandonment, and UK agencies name disconnected systems as their top blocker. Usually the records live in several places and nobody can say which is authoritative.
How long does it take to get data ready?
A day to assess one process. If it fails, usually a month of tidying: one record of truth, duplicates merged, documents moved out of inboxes, retention written down.
Does our CRM need replacing before AI?
Rarely. Most CRMs hold records adequately; the problem is that they are not used as the record of truth. Fixing that is cheaper than replacing the CRM.
Which automation needs the least data preparation?
Enquiry response and booking, because it creates the record rather than reading one. Document intake is next, because the AI reads variety and a person checks exceptions.
Want the assessment done for you?
The AI adoption audit includes the data readiness assessment for the processes worth automating, with the tidying plan where one is needed, credited against the first build.