An assistant that answers questions about your products is only as good as the product information it can read. A tool that drafts replies to customers is only as careful as the records it is given. When an AI trial disappoints, the cause is often found in a spreadsheet, not in the software.

Find out where the information actually is

Most businesses keep what they know in many places: an accounting package, a shared drive, WhatsApp groups, email threads, one person's memory. Make an honest inventory before planning anything.

Note which source is the trusted one for each kind of fact. If the price list in the drive and the price list in the billing system disagree, an AI system will repeat whichever it reads first.

Clean the part you will use

There is no need to tidy everything. Pick the data the first workflow depends on and fix it: remove duplicates, retire outdated documents, give files clear names and dates.

This work is slow and unglamorous, and it pays off whether or not the AI project goes ahead. Staff find things faster, reports agree with each other, and the next project starts from a better place.

Separate personal data early

Customer names, phone numbers, addresses, payment details and employee records are personal data. In India, the Digital Personal Data Protection Act, 2023 sets out how such data may be processed: broadly, for a clear purpose people have been told about and agreed to, with reasonable safeguards and a way for them to ask for correction or erasure.

So check whether sending that data to an outside AI service fits the purpose it was collected for, where it will be processed and whether the provider keeps or learns from it. Where the task does not need a person's identity, remove or mask it before it leaves your systems.

Write down who may use what

Decide which teams and which tools may read each source, and record it. Permissions that seem obvious today are forgotten when a new tool is connected in a hurry. A short written record also makes it easier to answer a customer who asks what happened to their information. Where the rules are unclear, ask a lawyer to review the specifics before data moves.

Let the data shape the ambition

Once the inventory is done, the realistic options become clearer. Well-kept, non-personal data such as product specifications or internal manuals is a good first candidate. Scattered or sensitive data may point to a smaller first step, or to fixing the records before anything else.

Back to the journal