Sep 12, 2026 · 7 min read
Running the numbers
What an unhandled exception costs after it leaves your log
Most teams can name their uptime and nobody can name this. Here is the arithmetic we walk through in week one, with a checkout timeout as the worked example.
Sep 12, 2026 · 7 min read
Dana Reyes
Engineer at Artisan
- Published
- Sep 12, 2026
- Length
- 7 min read
Everybody agrees an unhandled exception is bad. Almost nobody has a number for it, which is why the work to prevent one keeps losing to the work that has a ticket.
So here is the arithmetic, using a checkout timeout on a store doing $2.4M a year. Put your own figures in; the shape holds for a SaaS trial flow, a signup, or a job that quietly drops work.
The four lines
- The abandoned session. 0.9% of checkouts hit the timeout for six hours. At 1,100 checkouts a day and an average order of $88, that is $43 an hour walking out.
- The support tail. Eleven tickets, eighteen minutes each with the reply and the refund, at $31 an hour loaded: $102.
- The engineering hour. Two people, ninety minutes, mostly spent deciding whether it is the gateway or us: $155.
- The trust line. The one nobody wants to price. We use the store's own repeat rate — customers whose first order errored come back 22% less often — which on eleven customers is about $190 of expected margin.
Call it $700 for six hours of one bug nobody was paged about. Twice a quarter is $5,600, which is more than most teams spend on monitoring, hosting and error tracking put together.
Why the log was not the answer
The exception was in the log the whole time. It was in the log eleven thousand times, next to nine hundred thousand other lines, and the pattern that mattered — always the same route, always after a 8,000 ms wait on one outbound call — is not a thing a person finds by scrolling.
Logs answer a question you already know to ask. An issue arrives having asked it.
What actually moves the number
Two things, in this order.
- Grouping. Eleven thousand events become one issue with a count and a first-seen time. That is the difference between noticing in six hours and noticing in four minutes.
- A threshold, not an alert on every error. Page on a rate — more than fifty events in five minutes on this project — and the first 404 of the morning stops waking anyone up.
If you want one number to put in a planning doc, use the abandoned-session line alone. It is the most conservative of the four and the only one nobody argues with.
Keep reading
Catch the next one in four minutes.
Install the SDK, deploy, and watch the first exception arrive grouped, traced and assigned. Free for 5,000 events a month.