The Single Point of Failure Audit for Shop Skills

Why this matters

Every shop has at least one task that stops dead when a specific person is out. Most owners know roughly who that person is. Very few know how many of those tasks exist, how often they come up, or what a week without that person actually costs in rescheduled work and subbed-out jobs. The audit below turns a vague worry into a ranked list you can work through in slow weeks. It is not a training plan and it is not a strategy debate about whether to cross-train or specialize. It is an instrument: run it, get a number, decide what to close and what to consciously accept.

A skills single point of failure is a task, not a person

The distinction is the whole method. If you start by listing people, you write down "Tech A is our best guy" and learn nothing you can act on. If you start by listing the recurring work your shop actually sells, each line item is either covered by more than one person or it is not, and the ones that are not become your worklist.

A task is a single point of failure when exactly one person in the shop can complete it, unsupervised, to standard. That definition has a bar in it, and the bar is where most audits go soft. "Could probably figure it out" is not coverage. "Watched it done twice" is not coverage. Set the test at: has done it solo, within the last twelve months, and the work passed inspection or did not generate a callback. Anything short of that scores zero, not half. Rounding a half up to a one is exactly how a shop discovers on the day it matters that the backup cannot.

Access counts as a skill for this purpose. If one person holds the supplier portal login, the relationship with the inspector who returns calls, or the only key to the parts cage, the task behind it is a single point of failure even though nobody would list it as a skill. These are usually the cheapest gaps on the board to close and the easiest to miss entirely.

Build the task inventory from what you already dispatch

Do not brainstorm the list. Pull the last twelve months of dispatched work and group it into recurring task types until the groups stop feeling useful. Most residential-service shops land somewhere between eighteen and thirty types. Too few and the audit hides real gaps inside a broad bucket; too many and you spend an afternoon scoring variations of the same skill.

Add three categories the dispatch record will not show you:

• Back-office tasks that gate field work. Filing a warranty claim, pulling a permit, closing out an inspection, ordering from a supplier who only deals with one of your people.
• Qualification-gated tasks. Anything a person legally cannot do without a license, certification, or documented training. These behave differently in the audit and you need them flagged.
• Judgment calls. Quoting a replacement, deciding whether a system is safe to leave running, telling a customer no. These are the hardest to transfer and they almost never appear on a dispatch report.

Score exposure, do not just count heads

Once you know which tasks have coverage of one, the raw count tells you very little. Six single points of failure sounds bad, but if five of them come up twice a year and can wait a week, and one of them happens weekly and stops a job cold, you have one problem, not six.

Score each uncovered task on two axes, one to three each.

Exposure is frequency multiplied by recoverability, so it ranges from 1 to 9. Multiply rather than add: a weekly task that costs you nothing to defer is genuinely lower risk than a rare task that shuts a job down, and addition flattens that difference.

Score frequency on peak-season frequency, not the annual average. A commissioning task that happens forty times a year sounds monthly on an annual average and is actually four a week for ten weeks, which is when the holder is also most likely to burn out or take the vacation they deferred.

A worked audit: a five-tech shop

A shop with five field techs plus an owner who still runs calls part-time. The dispatch pull produces twenty-four recurring task types.

Coverage comes back: eleven tasks have four or more capable people, seven have two or three, and six have exactly one. Six out of twenty-four means 25% of the shop's recurring work runs on a single body.

The six, with who holds them and the scoring:

Total exposure across the six is 31. Tech A holds four of the six, which is 67% by count. But Tech A's four score 9 plus 6 plus 6 plus 3, which is 24 of 31, or about 77% of the shop's total exposure.

That gap between 67% and 77% is the reason to weight at all. Counting heads says "Tech A is most of our risk." Weighting says Tech A is more of your risk than the headcount suggests, and it tells you which of his four to attack first.

Now re-sort the bottom of the list by cost to close. Rank by exposure to know what matters, then check what each fix actually takes, because the two orders are not the same.

• Warranty filing scores only 3, the second-lowest on the board. Closing it is roughly 2 hours of screen-sharing with the office coordinator plus adding a second login. That is 1.5 exposure points closed per hour spent, and the hours are non-billable office time you can spend on a rain day.
• Startup and commissioning scores 9, the worst on the board, and looks like the biggest lift. It is not. Commissioning already happens weekly on billable work. Bringing a second tech along on eight of them over two months adds close to zero incremental non-billable hours, because the second body is riding on jobs the shop is already selling. The best deal on the whole board is the item that ranks first.
• Confined-space entry scores 3 and needs formal outside qualification, call it a full day of class plus a shop drill before anyone signs off. Roughly 0.375 exposure points per hour, the worst ratio on the list.

So the working order is: commissioning first (highest exposure, near-zero incremental cost), warranty filing second (cheapest close on the board, bank it on a rain day), then control programming and the older equipment line, then the inspection walkthrough.

Confined-space entry gets a different decision entirely.

Accepting a gap is a legitimate answer, if you write down the fallback

Confined-space work comes up a handful of times a year. Qualifying a second person costs a day of class and ongoing refreshers for a task that generates a rounding error of revenue. The right call for this shop is probably to accept the gap and route those jobs to a qualified sub when Tech D is unavailable.

That is a real answer, and an audit that produces zero accept decisions is usually dishonest. But an accepted gap only counts as accepted if the fallback is written down and reachable: which sub, their number, what they charge relative to your rate, and how long their lead time runs. An accepted gap with a documented fallback is a plan. An accepted gap with nothing behind it is just a gap you have stopped looking at.

Any task where the safety of the crew depends on the holder being present is a poor candidate for acceptance regardless of frequency. Rare and hazardous is exactly the combination where the backup will be improvising under pressure. For those, either qualify a second person or write a hard rule that the work does not proceed without the qualified person on site, and enforce it when a customer pushes.

What changes the answer

• Shop size. In a two-tech shop nearly everything scores as a single point of failure and the audit stops discriminating. Run it on the owner's tasks only, and shift the fix from "train a second body" to "write the fallback down." A two-person shop closes gaps with documentation and standing sub relationships, not with cross-training it cannot afford.
• License and certification gates. A qualification-gated task cannot be closed by cross-training. The fix is hiring, sponsoring someone through the qualification on a multi-month timeline, or a standing contract relationship. Score these normally, then flag them so they do not sit on a training list where they will never move.
• A holder who is leaving. If the person is two years from retirement or has told you they are looking, score recoverability on "when they are gone permanently," not "when they are out sick." A 2 becomes a 3, exposure jumps, and the item deserves to move up the list ahead of things that scored higher under the sick-day assumption.
• A new service line. Adding a service adds tasks that start at coverage of one by definition, usually held by whoever pushed to add the line. Re-run the audit after any launch rather than waiting for the annual pass.

The failure modes

Auditing people instead of tasks. Produces a ranking of your techs, which you already had, and no worklist.

Generous coverage scoring. The most common defect by far. A shop scores a tech as covered because they helped once or because nobody wants to say out loud that they cannot. Check the coverage claim against dispatch history: if a task went to the same person on every single ticket over twelve months, coverage is one, whatever the matrix says. Dispatch behavior is a better witness than self-assessment.

Treating the audit as a one-time exercise. Coverage decays. People leave, equipment lines change, a certification lapses. Run it annually, plus after any departure and after any new service launch.

Closing gaps on paper. A tech marked as trained who has never done the task solo is a coverage claim, not coverage. The close is not the training session, it is the first unsupervised job that passes.

How to verify you got this right

Pick the top-ranked gap you believe you closed and run a no-notice drill. Schedule that task on a day the original holder is genuinely unavailable, or have them ride along under a rule of total silence unless something unsafe is about to happen. If the backup completes it to standard without asking for a rescue, the coverage is real and you can score it as a two.

Second check: hand your top three gaps to the person who dispatches every day and ask them to name who they would actually send. If dispatch names the original holder for all three, nothing has changed on the ground no matter what the matrix says.

Third check: read your accept list back. Every accepted gap should have a named fallback and a phone number next to it. Any accepted gap that reads only as "we would figure it out" has not been decided, it has been deferred.

References

• OSHA general industry standards on qualified-person and training requirements for hazardous tasks
• U.S. Small Business Administration (SBA), business continuity planning guidance for small firms
• See related: Cross-Training vs Specialization Decision Matrix
• See related: The Annual Skills Review SOP
• See related: How to Document Tribal Knowledge So It Outlives the Person