"Our data is a mess" is one of the most common things we hear from businesses thinking about AI. It's usually said as a reason to wait. The reality is more encouraging: you rarely need perfect data to get value from AI. You need data that's accessible, accurate enough for the specific job, and allowed to be used for that purpose. Those are much more achievable goals.
Readiness is per workflow
Data readiness isn't a company-wide score. Your customer records might be in excellent shape while your product specifications live in scattered spreadsheets. What matters is whether the data needed for the specific workflow you want to improve is ready. That's good news, because it means you can fix a small, focused set of data rather than everything at once.
The five questions
For any workflow you're considering, ask these five questions about the data it depends on.
1. Where does it live?
List the systems involved: CRM, ERP, accounting, shared drives, email, spreadsheets. Be honest about the unofficial ones. If critical information lives in one person's inbox or a personal spreadsheet, that's the first thing to address.
2. Can software get to it?
AI and automation need a way to read and often write data. Check whether each system has an API, a built-in integration or at least a reliable export. Older or heavily customized systems may need more work. This question often decides what's practical.
3. Is it accurate enough?
Accurate enough depends on the job. An assistant that drafts replies for a person to review can tolerate some errors. An automation that updates invoices directly needs to be more reliable. Look at a sample of real records: are key fields filled in? Are there duplicates? Are statuses up to date? You'll quickly learn whether the data can be trusted for this purpose.
4. Is it consistent?
Consistency matters more than perfection. If customer names, product codes, dates and categories are entered the same way, software can work with them. If each person has their own way of entering things, you'll need some cleanup or rules to standardize them.
5. Are we allowed to use it this way?
This is the question most often skipped. Data may be restricted by privacy laws, industry regulations, customer contracts or your own policies. Customer personal information, health data, financial account information and client confidential material need particular care. Before sending data to an AI service, confirm that the use is permitted and that the service's data terms are acceptable. Our guide to AI risk and compliance basics covers this in more detail.
Documents are data too
Much of a mid-sized business's knowledge lives in documents: policies, procedures, contracts, proposals, product sheets and email. Modern AI can work with these directly, which is one of its most useful capabilities. But document collections have their own readiness issues:
- Outdated versions. If three versions of a procedure exist, an assistant may quote the wrong one. Identify the current, official version.
- Scattered storage. Documents spread across personal drives and email are hard to use. Consolidate the ones that matter.
- Access permissions. An internal assistant should respect who is allowed to see what. Don't give everyone access to HR or finance documents through a chatbot.
- Scanned images. Scanned documents need text recognition before they can be searched, and quality varies.
The fixes that matter most
You don't need a big data project before starting with AI. These focused fixes deliver most of the benefit:
- Pick a system of record for each type of information in your workflow, and stop keeping parallel copies in spreadsheets.
- Standardize a few key fields, such as status, category and owner, with drop-down choices instead of free text.
- Deduplicate the records your workflow depends on.
- Retire outdated documents or clearly mark the current version.
- Label sensitive data so everyone, including your automations, knows what needs extra care.
- Assign an owner who is responsible for keeping each data set in good shape.
AI can help clean up, too
AI is useful for data preparation itself. It can standardize inconsistent entries, flag likely duplicates, extract structured fields from free-text notes and summarize document collections. As with any AI output, have a person review changes before they're applied to important records, at least until you've seen how reliable it is on your data.
Signs you're ready to start
- You can name the systems that hold the data for your target workflow.
- At least one of those systems can be connected to or exported from reliably.
- A sample of records looks accurate enough for the job, with human review where needed.
- You've confirmed that using the data for this purpose is allowed.
- Someone owns the data and can make decisions about it.
If you can check most of these boxes for a single workflow, you're ready to begin, even if the rest of your data is still a work in progress. Try our readiness checklist for a quick self-assessment, or see how our data and systems integration service can help connect things up.