When the Numbers Tell the Wrong Story

On August 15, 2026 in Wrigley Field, Joshua Báez cranked a 449-foot home run on the first MLB pitch he ever saw. In his second at bat, he did it again, becoming only the third player in MLB history to hit two home runs in his first two at bats. In his next at bat, he did what no player had ever done before by hitting his third home run in his first three at bats. His debut immediately became one of the premier stories in all of sports and everyone wondered what was next for him. Turns out it was a whole lot of nothing. Báez went 0 for 18.¹

Business makes the same mistake, and it usually starts with the sample rather than the math. Three at bats is too few to tell you how good a hitter will be, and a Top Ten list is just ten customers chosen by size. Neither tells a full story about where the next hit will go or where a loss is going to come from. While neither number was wrong, both were judgment errors due to magnitude reported without context.

Simple vs. Advanced Metrics

When I joined a building products company as part of the Accounts Receivable department, I, like many others in analysis roles, inherited a set of reports and methods from the analyst before me. Chief among those was a series of Top Ten lists, including open balances, amounts past due and open credits. The issue was that every list had the same customers in basically the same order every month and told no story, while AFDA, the allowance for doubtful accounts, was absorbing losses from customers that management never saw. This was a symptom of the actual problem: the reports in place were just that, and no analysis was done on the data.

Stat	Definition	Proposed	Adopted	Weakness / Blind Spot
Batting average (BA)	H / AB	1871	1876	Excludes reaching base by walk and eliminates power
Slugging (SLG)	TB / AB	1867	1923 NL, 1946 AL	Excludes walks and overpays for power; a home run is not four singles
On-base (OBP)	(H+BB+HBP) / (AB+BB+HBP+SF)	1954	1984	Counts every way on base as one, so power disappears
OPS	OBP + SLG	1978	Never official	Adds unlike denominators, weights them equally and ignores park and era
wOBA	Events at their run values, per PA	2006	Never official	No park or era adjustment, and covers the plate appearance only

Distressed customers present one of two ways: one slams their credit limit hard all at once and stops paying; the other slow pays and short pays until the limit fills up behind them. Utilization catches the first before the aging has moved at all, and the aging catches the second while the limit still looks fine, which is why either signal had to be able to raise a customer on its own.

This resulted in the creation of the Risky Index. It weighted past dues and credit utilization measures against themselves before weighting the two against each other. A dollar ninety days past due is not a dollar thirty days past due, just as a customer sitting at 95% of their credit line is not one at 60%; scaling each factor by its own severity is what made them comparable in the first place. The metric was designed so no single warning signal could overpower the other, but either one on its own would draw attention to the distressed customer.

Since losses to major customers were written off to sales allowance and the mid- to low-level losses hit AFDA, the Risky Index identified and focused on the customers who would cause the unexpected risk.

As a mitigation tool, Risky was measurably successful. It reduced AFDA losses by over 90% in year one of implementation and avoided another $900k loss in year two. No customer touched the AFDA that Risky hadn't already identified months ahead, allowing our collectors to take action proactively and clear distressed customers long before they became losses. Customers that were previously overlooked were now on a Top Ten list that actually quantified their risk to the business.

Risky did not need new data to run: the past dues and credit utilization were already on every report I inherited. What it did need was the weighting and tuning, using past AFDA hits as learning cues. The weighting of them against each other and the incorporation of prior knowledge is what made them predict something.

The one issue was that, on its own, the number that Risky produced didn't actually mean anything. The score only meant something next to something else: next to its own history it tracked whether a customer's risk was changing, and next to the rest of the book it told us which fire to fight that week.

The Eye Test

Joshua Báez clearly has a bright future ahead of him. His slump ended going into September and midway through the month he is by some measures a top 15% hitter, which is the kind of swing both players and businesses go through. Nobody needed a number to see either end of it because the eye test said it all.

Risky automated the eye test, with the trend telling a collector if their efforts were working and the ranking telling them who to pursue first. Their expertise could be applied properly: identifying whether this was a cash flow problem that would clear in sixty days or a business quietly winding down, and whether the answer was a payment plan, a credit hold or a settlement. The Risky Index sized the problem and put it in order, empowering the team to solve it.