Project-based manufacturing: running the plant as a portfolio of jobs

Cost books to the job, not to the part

The controlling structure is a work breakdown: engineering hours, procured items, fabrication operations, assembly, testing and commissioning, each with a budget and a code. Shop-floor time booking has to be accurate at that level or the whole reporting layer is fiction. Committed cost matters as much as spent cost, since an order raised for a long-lead item consumes budget the moment it is placed. Software needs project accounting, progress measurement and a link from engineering release through to requisition — a heavier configuration than a parts-based plant ever requires. Retrospective reallocation between codes is the surest sign the structure was built after the work began.

Contention between jobs is the daily decision

Machines, floor space, cranes, test rigs and above all experienced people are shared, so the schedule is a negotiation rather than a calculation. Suitable work is large, low in repetition and long in duration: pressure vessels, packaging machinery, switchgear assemblies, marine fit-out. Capacity planning has to happen at the level of the constrained skill rather than the plant, since a shop with spare welding bays and no coded welders has no capacity at all. Slippage propagates too — one job holding a bay past its window pushes the next toward a delivery penalty.

Materials bought for one job and stranded by another

Purchasing is job-specific, which removes the pooling that makes stock efficient elsewhere. Long-lead items are ordered against a design not yet fully frozen, so a late change turns a delivered item into surplus. Part-built machines hold enormous value on the floor for months. Closeout usually reveals leftovers nobody will reuse: off-cuts, over-ordered fasteners, a contingency motor. A deliberate surplus review at each job's end, with material either returned, transferred or written off, stops the yard filling with the residue of other people's decisions. Transferring surplus between contracts needs a costing rule as well, or the receiving job takes the material while the sending job keeps the charge.

Milestones, retention and the cash curve

Payment follows agreed milestones — design approval, material delivery, factory acceptance, site commissioning — while spending follows the actual work, so the two curves diverge and the plant funds the gap. Retention held until final acceptance can represent the whole profit on a contract. Capital equipment is shared across jobs and justified on the mix rather than on any single award, which makes investment cases harder to write and easier to postpone. Variations decide the outcome: unpriced scope accepted informally by an engineer becomes an unrecoverable cost at handover. Front-loading a milestone at design approval is worth more to the business than a marginally better headline price.

Where project work comes apart

The recurring failures are scope creep accepted verbally, an optimistic engineering estimate nobody revisits after award, and documentation left until the end. Quality obligations here are dossier-shaped: material certificates, weld maps, non-destructive test reports, calibration records and inspection release notes assembled into a handover pack the customer's inspector will audit line by line. Building that pack progressively costs a fraction of reconstructing it after commissioning. Factory acceptance testing is where errors surface, and a failure there is expensive precisely because the customer is standing in the building. Late claims for extra scope rarely succeed once the customer has accepted the equipment, so the paperwork has to keep pace with the build itself.

Frequently asked questions

How do we price a job when the design is not finished?

Separate what is known from what is not. Fix the price on the defined scope, list the assumptions the estimate rests on, and attach a rate schedule to variations so additional work carries a pre-agreed cost. Contingency should be visible internally as a managed pot rather than smeared across line items where it disappears. Record the design freeze in the contract, because the gap between a change before freeze and one afterwards is the gap between a revision and a claim.

What is the right way to measure progress on a part-built machine?

Use physical completion against defined milestones rather than hours booked, which measure effort instead of achievement. Break each work package into observable steps — plates cut, frame welded, machined, assembled, wired, tested — and assign weightings agreed before the job starts. Reviewing earned value against committed cost each period exposes a contract drifting while there is still scope to recover it. Progress claimed by the person doing the work should be verified by somebody who is not.

How do we stop shared resources becoming the bottleneck on every job?

Plan at the level of the scarce skill or asset and publish that plan where all project managers can see it. Most contention comes from each job holding informal reservations nobody else knows about. A single constrained-resource calendar, a named arbitrator for conflicts, and a habit of releasing bays and rigs as soon as they are genuinely finished will recover more capacity than another machine, which often stands idle waiting for the same handful of people.