How we decide whether a piece of data is trustworthy enough to publish

There is a joke that has circulated in Kenyan banking circles for as long as I can remember — and I have been in banking circles long enough to have heard it at Standard Chartered, at I&M, and at least three corporate governance seminars where the tea was weak and the biscuits were stale.

A man walks into a bank in Nairobi and asks for the loan officer. He says he is travelling to Dubai on business for four weeks and needs to borrow Sh5,000. The loan officer, doing what loan officers do, asks for collateral. The man hands over the keys and logbook to a brand new Mercedes-Benz S-Class 500 parked outside the front door. Everything checks out. The loan is approved. The bank's president and officers have a good laugh — a Sh15 million Mercedes as security for a Sh5,000 loan — and an employee drives the car into the underground garage for safekeeping.

Four weeks later the man returns, repays the Sh5,000 plus Sh150.41 in interest, and collects his keys. The loan officer, still puzzled, says: "Sir, we checked you out while you were away. You are a multimillionaire. Why on earth would you borrow Sh5,000?"

The man smiles. "Where else in Nairobi can I park my car for four weeks for Sh150.41 and expect it to be there when I return?"

I love this joke for many reasons, but mainly for what it says about the loan officer. He had a number in front of him — Sh5,000 — and he thought he understood the transaction. He laughed, in fact, because the number looked absurd. A multimillionaire borrowing bus fare. But the loan officer had not asked the one question that mattered: what is this number actually for? He looked at the figure. He did not diagnose it. And the man with the Mercedes walked away with four weeks of secured parking in downtown Nairobi for the price of chai at a cafe.

Which brings me to what I do for a living at Index Digital Systems.

I am the Finance Editor at IDS. That means I am the person who decides whether a number — a currency rate, a mobile-money tariff, a market figure, a claim about the size of Kenya's digital lending book — is fit to publish on our platforms. I have a BSc in Statistics from the University of Nairobi, close to a decade of banking experience, and current consulting work in digital credit. I mention this not to pad the word count, good people, but because it is part of the verification. IDS's editorial standards require that money content is reviewed by a named editor with relevant experience. I am the named editor. If the number is wrong, my name is on it.

So how does the checking work? Let me begin from the beginning.

The first thing I do with any figure is what the loan officer in the joke failed to do. I trace it back to its origin — not the website that displayed it, not the blog that quoted it, not the WhatsApp forward that carried it through Nairobi like a matatu carries gossip on the Thika Road express. I want the primary source. For currency, that is the Central Bank of Kenya's published indicative rates. For stock prices, the Nairobi Securities Exchange. For M-Pesa charges, Safaricom's own tariff schedule — not a screenshot someone posted on X, not a table some website copied from another website that copied it from a third website that may or may not have checked it in 2022.

Why does this matter? Because data degrades as it travels with a persistence that is fairly alarming once you start paying attention. A CBK indicative rate gets rounded by one outlet, converted by another, timestamped by neither, and by the time it reaches Wanjiku scrolling her phone in a Rongai matatu the number she sees may be twelve hours old and Sh0.50 off in either direction. Half a shilling sounds harmless. Multiply it by a Sh5 million diaspora remittance and you are looking at Sh25,000 that has evaporated or materialised depending on which side of the transaction you sit. That is not a rounding error. That is someone's rent.

The second check is the one the loan officer skipped entirely. What does this number actually measure?

I spent four years studying statistics at the University of Nairobi, and if that degree taught me one thing it is that the same number can mean three different things depending on whether you are looking at a mean, a median, or a mode, and whether the denominator is the population, the sample, or something the analyst cobbled together on the back of what I call a hastily-scribbled-on-a-serviette-at-Java calculation. The Sh5,000 in the joke was not a loan. It was a parking fee. The number was correct. The diagnosis was wrong.

Take the claim — repeated for years by the Financial Times, the Economist, and half the fintech conference circuit — that 43 per cent of Kenya's GDP "passes through" M-Pesa. The figure was arithmetically correct. It was also economically nonsensical. Mobile-money volumes count every deposit, withdrawal, and transfer — gross flows. GDP counts goods and services produced. One mwananchi sending Sh1,000 to his mother, who withdraws it, who gives it to Mama Mboga, who deposits it again, generates three mobile-money transactions and zero GDP. The bank officers in our joke looked at a Sh5,000 facility and saw a laughably small loan. The international press looked at 43 per cent and saw a staggeringly large economy. Both missed the question: what is this number actually measuring?

At IDS, if I cannot state clearly what a figure represents — its definition, its scope, the population it covers — it does not publish. Period.

The third check is time. In credit operations at I&M Bank, where I handled lending decisions and RTGS payments, we had a hard rule: a valuation report older than six months was a decorative document, not a risk input. The property was worth Sh8 million in March. By September the market had shifted, the county government had gazetted a new road that either doubled the value or cut it in half depending on where the tarmac landed, and the borrower was waving a piece of paper that told you what the property used to be worth. Useful for nostalgia. Useless for lending.

The same rule applies to publishing. A dollar-shilling rate from yesterday is history. An M-Pesa tariff from before the last revision is wrong — not roughly right, not close enough, wrong. Every time-sensitive figure on our platforms carries a visible "last updated" timestamp. I insisted on this. It is, in my considered opinion, the single most important thing a financial data publisher can do, and also the thing most Kenyan financial websites do not bother to do.

But let me park timestamps aside and talk about the check that makes me most uncomfortable.

Here is what happens. A number arrives that confirms what I already know. Average digital loan size: Sh5,000. That sits comfortably within the range I have seen in my consulting work with SACCOs and digital lenders. My brain relaxes. My scrutiny drops. And that, good people, is precisely when you end up like the loan officer — laughing at a transaction you have not actually understood.

I have trained myself, imperfectly I confess, to be more suspicious of comfortable numbers than surprising ones. A surprising number triggers investigation by instinct. "The shilling strengthened by 3 per cent in a single session? Let me check that twice." A comfortable number triggers nothing, and nothing is where errors nest like termites inside a ceiling board, quietly eating through the structure until the whole thing comes down on your head. So I ask: average across which lenders? Mean or median? Over what period? Weighted how? If the source cannot answer those questions, the number goes into quarantine. It sits in a spreadsheet, dated and attributed, until someone either confirms it or kills it.

In these dying column minutes let me draw your attention to one last thing. Ownership.

Every piece of finance content on IDS carries a named editor. That editor — usually me, for markets, currency, and mobile money — is personally accountable for the accuracy of every figure in it. Not accountable in the abstract, feel-good, corporate-governance-seminar sense where everyone nods solemnly and then goes for tea. Accountable in the sense that if a reader writes in and says "your exchange rate was wrong on Tuesday," my name is on it, and I am the one who has to trace the error, fix it, and publish a correction in the open.

That changes how you treat data. It is fairly easy to publish a number when nobody knows who approved it. It is rather less easy when your name, your photograph, and your professional biography are attached to it like a stamp on a bank guarantee that someone will eventually come to collect on.

In my banking days we called this skin in the game. The loan officer who signed the credit memo was the loan officer who sat in the recovery meeting when the borrower defaulted. You tended, under those conditions, to read the financials with a certain concentration.

Does our process catch everything? No. We are a small operation. We will get things wrong. What we will not do is get things wrong silently. The correction goes up. The timestamp updates. The named editor answers for it.

Which brings me back to the man with the Mercedes and the loan officer who thought he had seen it all. The bank had every piece of data it needed — the logbook, the credit check, the Sh5,000 application, the multimillionaire's net worth. What it did not have was the right question. The officers looked at the number and laughed. The man looked at the number and parked his car.

At IDS we try, every day and with every figure, to be the person who asks what the number is actually for — before we publish it and before someone else drives off with four weeks of free parking while we are still chuckling at the Sh150.41.