A striker’s value is easy to see. Goals, assists, shots on target: every one is an event, timestamped and counted.

A centre-back’s best work often leaves no trace at all. The run he tracked. The pass that was never played because he was standing in the way. The striker who drifted out of the game because there was no space to drift into. You can count tackles, interceptions and blocks, but the most valuable thing a good defender produces is a goal that didn’t happen. And a non-event doesn’t show up on a stats sheet.

That gap has cost even the best managers. In The Numbers Game, Chris Anderson and David Sally tell the story of Sir Alex Ferguson looking at Jaap Stam’s numbers and noticing that his tackle count was falling. The obvious reading was that Stam was getting older and losing his edge, and in 2001 Manchester United sold him to Lazio. Stam went on to play several more years at the top level.

The better reading was the opposite. With age and experience, Stam had got better at reading the game. He was in the right place earlier, breaking up attacks before a last-ditch tackle was ever needed. Fewer tackles wasn’t a sign of decline. It was a sign of a defender preventing the situations that required them. The data was accurate. The inference was wrong, because the thing that mattered most was never counted.

Businesses have exactly the same problem, and most of them never notice it.

Two kinds of invisible

The first kind is prevention. The bookkeeper who catches a supplier double-charging before it’s paid. The weekly check that spots a pricing error before a hundred orders go out at the wrong price. The person who knows which customer always needs chasing twice. When they do their job well, nothing happens. Nothing happening is very hard to put a value on, so it usually gets valued at zero.

The second kind is opportunity turned away. There’s a story I’ve always liked about PayPal. In 2008 it bought Fraud Sciences, an Israeli start-up that assessed transaction risk using a buyer’s online footprint. As Dan Senor and Saul Singer tell it in Start-up Nation, when the start-up was tested on PayPal’s own transactions, it didn’t just catch fraud PayPal had missed. It also identified good transactions that PayPal’s own rules had rejected.

That second finding is the uncomfortable one. Every rejected transaction was a customer who wanted to pay and was turned away, and PayPal’s systems had no way of knowing. A rejected customer generates no data. Nobody follows up on them. In credit, this blind spot has a name: reject inference. You only ever learn how the people you approved behaved. The ones you turned down stay invisible.

Non-events can be measured

The good news is that “it didn’t happen” isn’t the end of the analysis. It’s the start of a comparison.

Statisticians do this routinely with mortality. To work out the effect of an event, they estimate how many deaths would normally have been expected, based on previous years, and compare that with what actually happened. In 2020, the year of the lockdowns, road deaths in Great Britain fell by 17%, to 1,460. Nobody can point to the individual people who weren’t killed, but the gap between expected and actual is real, and it can be counted.

Football analytics has been moving the same way. The newer models don’t just count what a defender did. They estimate how likely the opposition was to score before and after each action, and credit the player with the difference. Some go further and model where an average defender would have stood, then compare that with where this one actually was. In both cases the trick is the same: build a credible picture of what would have happened, and measure against it.

The non-event becomes measurable the moment you have a baseline.

What this means for an owner

You don’t need a data science team to apply this. Three habits get most of the way.

Keep a baseline. Before changing a process, or before someone leaves, note what “normal” looks like: credit notes issued per month, invoices queried, write-offs, stock-outs, complaints. Most of these numbers already exist somewhere. If a check is removed and they drift upwards, you’ve measured what the check was worth, unfortunately after the fact. It’s far better to know it beforehand.

Count the catches. A defender’s interceptions are an imperfect measure, but they’re far better than nothing. The same goes for prevention in a business. A simple log of “caught before it went out” (the wrong price, the duplicate invoice, the order that would have shipped to an old address) turns invisible work into something you can see, discuss and value.

Occasionally look at what you turned away. If you have rules that reject things automatically, such as minimum order sizes, credit checks or lead filters, the rejects are worth an occasional look. Lenders do this deliberately: they approve a small, controlled sample of applications their rules would have declined, purely to learn whether the rules are right. You can do a low-risk version of the same thing: review a month of declined leads or refused orders and ask how many you’d now say yes to.

Why it matters now

This is about to matter a great deal more. Automation is very good at doing measurable things faster. It’s much worse at noticing the quiet, preventative work that people do without being asked, often without realising they’re doing it.

Automate a process without first understanding what it was quietly preventing, and you may find out the hard way. That’s the business equivalent of selling your centre-back because his tackle count fell.

So before you decide something isn’t worth the cost, it’s worth asking what has not been happening while it was there.

Who in your business is quietly stopping goals going in?

Sources

Delphi Decide helps owners see what their numbers are really telling them, including the things that didn’t happen. If you’d like to talk it through, get in touch.