Mapping Your AI Data Inventory Before Your First Vendor Demo
Most AI conversations in Asian businesses begin with a tool: a chatbot, a document summarizer, an automated report generator. The vendor demo is polished, the use case looks obvious, and the team starts imagining savings. What usually gets skipped is the unglamorous question underneath every AI tool: what data would this system actually need, where does that data live, and who is allowed to touch it? This article offers a different entry point. Before you evaluate any product, build a plain inventory of your data and the people responsible for it. The result will tell you which AI ideas are realistic, which need preparation first, and which questions to put to any vendor before you sign anything.
Start With the Questions the Tool Will Ask of Your Data
An AI system is only as useful as the information it can reach and interpret. Before comparing products, write down the three or four decisions or tasks you hope AI will help with. For a customer support team, that might be drafting replies to recurring questions. For an operations manager, it might be flagging late shipments. For a finance lead, it might be categorizing expense records. Each task implies a question about data: what does the system need to see in order to answer well?
Be concrete. A drafting task may need past support tickets, product documentation, and tone guidelines. A flagging task may need timestamps, order status fields, and carrier updates. Writing these requirements down turns a vague ambition into a list you can check against reality. It also exposes gaps early, such as a team that answers most customer questions by phone and never records anything in a shared system.
Build a Simple Data Inventory
Create a table, spreadsheet, or document with one row per data source. For each row, record the name of the source, the type of information it holds, the person or team who owns it, where it is physically or digitally stored, how often it is updated, and whether it contains personal information about customers or employees. Keep the language plain. The goal is to know what exists, not to produce a formal compliance document.
Many businesses discover at this stage that critical knowledge lives in personal inboxes, spreadsheets on one employee's laptop, or chat threads that disappear. Record those too. An honest inventory is more valuable than a tidy one, because the AI tool will be fed by whatever is really there, not by what you wish were there.
Assess Readiness Before Access
For each data source, ask four readiness questions. First, is the information accurate and reasonably current? Second, is it consistently labeled, so that a field called 'status' means the same thing across records? Third, can you export or access it without relying on one person's manual effort? Fourth, do you have a clear answer about whether it may be shared with an outside service?
A simple scoring approach works well. Rate each source from one to three on each question, then look at the totals. Sources scoring high across the board are candidates for early experiments. Sources with low scores on consistency or access are preparation work: cleaning labels, documenting definitions, or moving files out of personal drives. Sources with unclear sharing rights should not feed any external tool until someone responsible has reviewed them.
Questions to Put to Any AI Vendor
Once your inventory is ready, you can evaluate vendors against it rather than against their marketing. Ask them to explain, in writing, what data their product needs, whether your data is used to train shared models, where data is processed and stored, how long it is retained, and how you can delete it. Ask what happens to your data if you end the contract. Ask who at the vendor can access your inputs and under what circumstances.
Do not assume answers. If a vendor cannot describe its data handling clearly, treat that as information. Also ask about integration: which of your inventoried sources can connect directly, which require manual export, and what the vendor expects you to change. A tool that requires reformatting every record each week has a hidden operating cost that belongs in your comparison.
A Short Checklist Before Your Next Demo
Before scheduling any AI vendor demo, confirm that you have named the two or three tasks you want help with, listed the data sources those tasks depend on, assigned an owner to each source, scored each source for readiness, and identified which sources contain personal information. Then prepare the written questions listed above. Bring the inventory to the demo and ask the vendor to walk through how their product would use your specific sources, not a generic example.
Teams that work this way tend to make calmer decisions. They pilot where the data is ready, fix what is not, and avoid committing to tools that cannot see the information they actually need. The inventory is not a one-time project. Revisit it when you change systems or add a market, because your data landscape will shift as your operations grow.