Upload your M-PESA statement and start asking it questions

At stage 1, the AI wrote things for you. At stage 2, it starts answering questions about your information - not the internet's. The clearest first place to feel this is a document nearly every Kenyan business already has: an M-PESA statement.

Your statement is a complete record of money in and out. The problem is that it is hundreds of lines of raw transactions, and no one has time to read it. An AI can read all of it in seconds and answer the questions you actually care about. Here is how - and, just as importantly, how to catch it when it is wrong.

First, get the statement as a file#

Request your official M-PESA statement (you can get a PDF for a chosen period). That file - the real, complete record - is what you hand to the AI. This is the heart of stage 2: you are pointing the assistant at your own data instead of asking it to guess.

The questions an owner actually asks#

Forget clever data-science questions. Ask the ones that keep you up at night.

Where is the money going?#

Here is my M-PESA statement for last month. Group my payments OUT into
categories - stock, rent, salaries, transport, airtime, personal - and tell
me the total for each, largest first. List anything you were unsure how to
categorise separately instead of guessing.

Who owes me, and who is paying late?#

From this statement, list the customers who paid me, with amounts and dates.
Flag any regular customer whose payment this month was smaller or later than
their usual pattern.

What changed this month?#

Compare this month's statement to the previous one I am pasting below.
Tell me the three biggest changes in money in and money out, and by how much.

That last question is the one owners find most useful. It turns two walls of numbers into three plain sentences about what actually shifted.

The answer it got wrong - and how we caught it#

Now the honest part, because this is stage 2's whole lesson.

Asked to total one month's stock payments, the AI returned a figure that looked right - until we checked. It had missed three transactions near the end of a long statement and quietly left them out of the sum. The total it gave was confident, clean, and wrong by several thousand shillings.

We caught it for one reason: we asked it to show its working.

Now show me every individual transaction that went into that stock total,
with dates and amounts, so I can check the sum myself.

With the transactions listed, the gap was obvious. This is the single most important habit at stage 2.

The rule for trusting a number#

Use the AI to find and organise, but verify before you decide. In practice:

• For a total you will act on, ask it to list the lines behind it and spot-check.
• Treat a single big surprising number as a lead to check, not a fact to bank.
• The more a number matters, the more you verify it. This never changes.

Used this way, the statement becomes something you could never read by hand: a searchable, question-answerable record of your cash flow.

Why this is a real step up#

Notice what changed from stage 1. The AI is no longer writing from general knowledge - it is reasoning over your actual transactions. That is the foundation of everything above: an assistant that answers from your price lists, your policies, your records. The M-PESA statement is just the easiest place to feel it working, because the data is already complete and already yours.

Try it this week#

Pull one month's M-PESA statement and ask it the "where is the money going?" question above. Then ask it to show its working and check one total by hand. That single exercise teaches you both the power and the discipline of stage 2. For the full climb, see the five stages of AI adoption.