ChatGPT, Claude and Copilot already help many people write text. Once they can also see your company data, they can do a lot more: from a quick number to small everyday tasks.
Why the data comes first
An assistant only knows what you show it. If your data is spread across an accounting program, an online store and several Excel files, it cannot tell you how much you sold last month. Sometimes it answers anyway, and the answer sounds convincing but does not match your numbers.
So I start with the data. It goes into one place, tables and columns get clear names, and every measure gets a description, because "profit" means sales minus the cost of goods in one company and also minus advertising in another. I connect the assistant to a single reporting table, read-only. It cannot change or delete anything, and the AI account stays yours.
What you can ask
These examples come from an invented home-goods shop with made-up data. I ask in Lithuanian and check every answer against the report.
- Totals. "How much did we sell over the year?" Answer: €412,437, of which the online store €209,024 and the shop in Kaunas €203,413.
- Comparison. "How did September compare with August?" Answer: €33,110 against €25,701, which is 29% more, with each channel shown separately.
- Profitability. "Which category loses money online?" Answer: candles, minus €10,448 after all online costs, while in the physical shop they earn €12,483.
- Advertising. "Does the candle advertising pay off?" Answer: €27,599 was spent, 56% of the whole ad budget, and the return is 1.48 times against 5.67 to 9.58 times for the other categories.
- Returns. "Which products are returned most often online?" Answer: linen textiles, 13.6% of order lines, against 1.9 to 4.8% for the other categories.
- Weekly rhythm. "Which weekday sells the most in the shop?" Answer: Saturday, €59,438 over the year.
One question gives one number. A series of questions shows where you are losing money: here, candles online and their advertising.
Who can ask
Not only the person who knows how to use the report. A sales manager, an owner or an accountant can ask in their own words without setting any filters. You do not have to ask someone to prepare a separate summary each time, and nobody has to learn Power BI.
When you want more than answers
Once the assistant is in daily use, some clients ask whether it can also do something. Then I add a few agreed actions, for example adding a note to an order, changing an order status or creating a task. You approve every change, and every change is logged (who, when, what the old value was), so it can be undone. The assistant cannot delete anything.
I do not do this straight away. First you need to see how the assistant works with read-only access.
Where the assistant can still be wrong
AI sometimes makes mistakes, even when the data is tidy. That is why the report stays the official source of the numbers. If a number matters for a decision, check it there.
Who this is for
Small and mid-sized companies without an in-house analyst. I start with a free 30-minute call on Google Meet. After it you get a short written summary with a price range, and you decide whether to go ahead. You can book a call here or write to rytis@numeva.lt.