Is your data ready for AI? A check for small businesses
By Keith Hamilton · · 4 min read
AI is only as useful as the information it works from, so before you buy any tool, ask: is your data ready for AI? Give it an out-of-date price list and it will quote the wrong price, confidently. The good news is that getting data ready for AI does not mean a large data project. For a small or medium-sized business, it means checking that the information one task depends on is findable, accurate and safe to use, and this article shows you how.
What does "data" mean in a small business?
In a small business, data means the everyday information you already hold:
- Spreadsheets, often several versions of the same one.
- Email inboxes, including shared ones.
- Documents in a shared drive: policies, procedures, price lists, templates.
- Your CRM, accounts package, booking system or order system.
- Your website.
- Knowledge in people's heads that has never been written down.
AI works with most of these. The question is whether the information in them is in good enough shape for the task you have in mind.
Where should you start preparing data for AI?
Start from the task, not the data. Do not try to tidy everything: pick the first task you want AI to help with, then list the information that task needs.
For example, an AI tool that answers customer questions about delivery needs your delivery policy, your current charges and access to order status. It does not need your payroll records or last year's marketing plans.
If you have not chosen a task yet, our guide on where to start with AI will help.
How do you check whether your data is ready for AI?
Ask five questions about the information your chosen task depends on.
1. Where does it live?
There should be one place where the right version is kept. If the answer to "where is it?" is "it depends who you ask", that is the first thing to fix.
2. Is it accurate and current?
Check when it was last reviewed and who is responsible for keeping it up to date. Out-of-date information produces wrong answers, because the AI has no way of knowing it has been superseded.
3. Is it consistent?
Things should be recorded the same way each time. A customer list where the same company appears three times with slightly different names, or dates typed in several formats, causes problems for any system, not only AI.
4. Can a system reach it?
Software reaches some information easily and some only with extra work. A well-organised spreadsheet or a system with an API (a standard way for software to connect) is easy. Scanned documents, handwritten notes and information held only in someone's memory are harder: not impossible, but they add effort and cost.
5. Who is allowed to see it?
Only the people entitled to see information should be able to reach it through AI. If an AI tool can read a folder, it can repeat what is in it to anyone who asks, so check access before you connect anything.
What are the common data problems and fixes?
These are the five problems to look for when improving data quality for AI in a small business, and each has a simple fix:
- Several versions of the truth. Agree which version is current, archive the rest and name one owner.
- Duplicate records. A one-off clean-up of the records the task uses, then a simple rule for how new ones are added.
- Free-text fields used for everything. Add a few clear categories for the information the task depends on.
- Knowledge in people's heads. Ask the people who know to write down the answers to the questions they are asked most often. A page or two is enough to start.
- Scanned or image-only documents. Check whether the originals exist in a form that software can read before deciding to work from scans.
What does this look like in practice?
Problems surface as soon as you gather the documents a task needs. A firm setting up an assistant to answer staff questions about internal policies starts by collecting its policy documents. That is the point at which it finds, for example, three versions of the holiday policy in different folders, or an expenses procedure last updated before the approval process changed.
Fed straight into an assistant, those documents give staff different answers depending on which version it finds. A few hours spent agreeing the current versions, removing the old ones and naming an owner for each fixes the problem before it starts. Our article on internal knowledge assistants covers how to keep that content current once it is in use.
Do you need perfect data?
No. Waiting for perfect data is a good way never to start. Aim for information that is good enough for the first task, and improve it as you go, because running a tool on real work quickly shows you where the gaps are.
How should you handle personal data?
If the task involves information about customers or staff, UK GDPR applies. Use only the personal data the task needs, know where it will be processed, and check your supplier's terms.
Our UK GDPR and AI checklist sets out what to check before you switch anything on.
How do you check your AI data readiness quickly?
Answer these five questions:
- Have we chosen one task and listed the information it needs?
- Is there a single, current version of each source?
- Does each source have a named owner?
- Can the tool reach the information it needs, and only that information?
- Have we checked any personal data against UK GDPR?
Where Forwardcycle fits
Our AI Opportunity Diagnostic includes a process inventory that shows which tasks are ready for AI now and which need some groundwork first. For a quick view of where you stand, take the free AI Readiness Assessment, or arrange a free call.