How to tell whether a job made money
Most engineering SMEs cannot answer this question, which means they cannot answer any of the ones that follow from it. You do not need a system to start. You need twenty jobs and a fortnight.
The question underneath all the others
Which customers should we keep? What should we quote? Do we need another machine? Should we take this contract? Every one of those depends on knowing what work actually costs, and in a great many engineering businesses turning over between one and twenty million pounds, nobody does.
That is not incompetence. It is what happens when a business grows out of a workshop where the owner knew every job personally. The gross margin in the annual accounts is real, but it is an average across work of wildly varying profitability, and an average cannot tell you which end to stop doing.
Do not start with software
The usual response is to look at MRP or job-costing systems. Sometimes that is right, eventually. It is almost always wrong as a first move, for three reasons: implementation takes a year and the answer is needed now; a system fed with the same unreliable hours produces confident nonsense; and you cannot specify what you need until you understand the shape of the problem.
Start by measuring a sample by hand. Twenty jobs, properly, beats four hundred badly.
The method
1. Choose twenty jobs that represent the business
Recently completed and fully invoiced. Deliberately mixed: a couple of large ones, several typical ones, some small ones, at least three from your biggest customer, and — this matters — two or three that everybody already suspects were bad. Do not pick the interesting ones. Pick the ordinary ones.
2. Reconstruct the actual hours
Not from a timesheet system you do not trust. From whatever genuinely exists: job cards, machine logs, the setter's notebook, the works order the job travelled with. Where the record is missing, sit with the person who did the work and reconstruct it. They will remember, and their estimate is far better than the zero you currently have.
Capture setting separately from running. In a job shop with short runs, setting is frequently the whole story, and it is invisible on a quote that thinks in unit times.
3. Add the material actually issued
What was drawn from stores, not what the estimate assumed. Include the offcut you could not use and the bar you had to buy in because stock was wrong.
4. Count the rework honestly
The second setup because the first one was scrapped. The weld that had to be ground out. The parts that came back from the customer. This is the number most often left out and it is frequently the difference between a job that made money and one that did not.
5. Apply one overhead rate, and be transparent about it
Total annual overhead divided by productive hours available. Not a rate from 2016, and not a different rate per machine at this stage — that is a refinement, and refinements come after you know whether the problem is a rounding error or a third of the margin.
6. Put it in one table
Job, customer, quoted price, quoted hours, actual hours, material, rework, recovered margin. Twenty rows. That table has settled more arguments in engineering businesses than any system I have seen installed.
Two or three patterns, nearly every time: one customer is significantly less profitable than everyone believed; short runs are subsidised by long ones because setting is under-recovered; and a specific category of work — often the awkward, urgent, favour-to-a-good-customer sort — loses money consistently and has been treated as goodwill for years.
Then, and only then, decide what to fix
The findings usually point at one of four things. If the estimate is systematically wrong, fix the estimating method — often just setting time and a realistic recovery rate. If the estimate is right and the job overruns, the problem is on the shop floor: sequencing, setting, or rework. If both are right and the margin is still thin, it is a pricing problem and it is a commercial conversation, not an operational one. If the numbers vary wildly with no pattern, you have a data problem and that is the thing to fix first.
Only after that does it make sense to ask what a system needs to do — and by then you can specify it in a page, because you know exactly which numbers you could not get.
Keeping it going
The sample tells you the shape of the problem. Making it permanent needs one habit: quoted hours against actual hours, on every job, reviewed monthly with the people who did the work. Not to apportion blame — the moment it becomes that, the numbers become fiction — but because an estimating process that never learns what happened cannot improve.
Businesses that do this consistently tend to find that within a year they are quoting differently, turning down work they used to chase, and making more money on less turnover. That last part is uncomfortable and it is usually the correct outcome.