Most small operations run on an unspoken rule: nothing is allowed to fail. It is a rule nobody can follow, so it quietly gets broken every week while everyone pretends otherwise. An error budget audit replaces the pretending with arithmetic. You work through each area of the business, decide how much failure it can actually tolerate, write that tolerance down, and check real performance against it. The method comes from the engineers who keep large web platforms alive, but the audit below needs nothing more than a notebook, two honest hours, and the willingness to admit that “zero mistakes” was never a plan. Work through the ten points in order. Score each one pass or fail. Fix the fails before you add anything new to your plate.
A note on how to run it before you start. Do the audit with whoever actually fields the failures — the person answering complaint emails knows the real miss rate better than any dashboard or gut feeling. Use last quarter’s actual numbers where you have them, and honest estimates where you do not; an estimate you revisit beats a blank you ignore. And agree upfront that the audit is about the operation, not about anyone’s competence. The whole point of budgeting failure is to stop treating it as a personal shame and start treating it as a resource to be allocated. Done in that spirit, the ten points below take an afternoon and change how the next quarter gets planned.
The 10-point error budget audit
1. Every promise to customers has a written tolerance
Pull up the commitments you make publicly — delivery in three days, replies within 24 hours, site always up. For each, ask: what miss rate can we genuinely absorb before customers walk? If “delivery in three days” really means “97 times out of 100”, write exactly that. Pass only if each promise carries a stated tolerance someone chose on purpose. An implied 100% is an automatic fail.
2. The tolerance is a number, not an adjective
“Mostly reliable”, “rarely late” and “almost never down” are moods, not targets. Convert every tolerance from point 1 into a figure you can count: missed orders per hundred, hours of downtime per quarter, complaints per month. If you cannot count it, you cannot budget it, and you will argue about it forever. Pass when every commitment has a numeric ceiling on failure.
3. Someone tracks spend against each budget
A budget nobody monitors is decoration. Assign each tolerance an owner and a simple tally — a shared sheet updated weekly is plenty. The owner’s only job is to know how much of the quarter’s allowance is already gone. Three late deliveries against an allowance of ten? Fine, note it and move on. Eight against ten with six weeks left? That is a flashing light, and you want it flashing early.
4. There is an agreed response when a budget runs out
This is where most operations fail the audit. Decide now, in writing: when the failure allowance for an area is spent, what stops? Sensible answers include freezing changes to that area, pausing new feature work, or pulling a person off growth tasks to stabilise. The response must be automatic enough that nobody has to win an argument to trigger it. If exhausting a budget changes nothing, you never had a budget — you had a diary of disappointments.
5. Risky changes are charged against the budget before they happen
New supplier, new checkout flow, new pricing page — every change carries failure risk. Before approving one, check the relevant budget’s balance. Plenty of allowance remaining means you can move fast and absorb a stumble. A nearly spent allowance means the change waits or ships in a smaller, safer slice. Teams upgrading their stack — for instance while bringing web platforms, cloud systems and AI tools together — should time the riskiest steps for when budgets are freshest, right after a clean month, not during a shaky one.
6. Cash gets the same treatment as operations
Money needs an error budget too. Set an explicit monthly overrun ceiling — say 4% above plan — and a hard rule for what freezes when you hit it. This turns budget reviews from blame sessions into balance checks. It pairs naturally with the discipline of tracing where cash leaks out and shielding your capital: the leak-hunting tells you where the losses come from, the budget tells you when the total is no longer acceptable. Pass only if the overrun ceiling and the freeze rule both exist in writing.
7. The calendar carries deliberate slack
Audit your scheduling. If every project timeline assumes zero sick days, zero vendor delays and zero rework, your plan is fiction with a start date. Build an explicit slack allowance into each project — one buffer week per two months of work is a reasonable opening position — and defend it from people who see “empty” time and want to fill it. The slack is not spare. It is the budget, pre-purchased, for problems you cannot yet name.
8. Small failures are logged, not just survived
You cannot manage a budget whose withdrawals go unrecorded. Keep a plain running log: date, what failed, which budget it hit, rough cost. Five minutes a week. After one quarter the log will tell you something the audit cannot — which tolerances were set wildly wrong. A budget that never gets touched was too generous and is hiding room for faster moves. One that empties by week three was dishonest, and the operation underneath it needs real repair.
9. Growth work and stability work are traded consciously
Every hour spent chasing new business is an hour not spent making current delivery sturdier, and pretending there is no trade-off is how operations burn out. The budget makes the trade explicit: while failure allowances are healthy, tilt hours toward growth; once an allowance is nearly spent, tilt back to stability until it recovers. This single rule resolves the oldest argument in any small firm without a single raised voice. If you are scoping new engagements — perhaps reviewing how a digital services plan gets sequenced for incoming clients — check the stability budgets first and let their balance set the pace of what you take on.
10. Somebody in the operation has actually studied this discipline
Error budgets, service-level targets and the craft of spending reliability wisely form a genuine body of knowledge, not folklore. It grew up inside site reliability engineering, and it is teachable. SRESchool runs instructor-led reliability courses and consulting covering exactly this territory — defining service-level objectives, weighing dependability against delivery speed, and cutting the repetitive toil that eats teams alive. If your operation leans on technology at all, having one person trained in the formal version of this audit pays for itself the first time a budget decision gets made from knowledge instead of instinct. Pass if someone owns the skill. Fail if the whole framework lives in one founder’s head.
Scoring your error budget audit
Count your passes. Eight to ten means you run a genuinely managed operation — your remaining task is reviewing the numbers quarterly so tolerances track reality. Five to seven means the skeleton exists but the muscles are missing; points 3, 4 and 8 (tracking, consequences, logging) are where paper budgets become working ones, so fix those first. Four or fewer means your operation currently manages failure by mood, and the kindest thing you can do is start with point 1 this week: one promise, one written tolerance, one tally.
Resist the urge to fix all ten at once — that is perfectionism sneaking back in through the side door, and perfectionism is precisely the disease this audit treats. Set tolerances you can defend, watch the balances, and let an exhausted budget — not a bad morning — be the thing that changes your plans. Run the audit again in ninety days. The operations that repeat it are the ones where “reliable” stops being a hope and becomes a line item.
