Publish Time: 2026-11-18 Origin: Site
Project risk guide · Published by I.C.T · For process engineers, project managers and equipment buyers
Every PCBA factory project carries the same ten categories of risk, and almost every project discovers them in the same order — late. The freeze that was not really a freeze, the transformer that had a twelve-week queue, the crate that cleared customs two weeks after the crew arrived, the feeder list that did not match the BOM, the operator who was hired in week one and trained in week nine. None of these is exotic. What separates a project that absorbs them from one that does not is whether they were written down early enough to be managed, and whether each line had a date attached instead of a colour.
This guide builds a register for PCBA factory setup project risks that runs in the order the project runs: requirement and layout freeze, building and utilities, equipment interfaces and customs, installation and commissioning, material, programming and people, validation, and ramp-up. Each section gives the risk lines that belong in it, the trigger that tells you the risk is becoming real, and a worked figure so the exposure can be put in currency rather than in adjectives. Every number is a stated assumption for illustration, not a measurement from a named project, and none of it should be used without your own data behind it. The equipment that gets installed inside that schedule is described under the line packages that a setup plan has to fit around, and it is the least variable part of the whole exercise.
A risk register fails for a boring reason: it is organised the way the company is organised, so every line belongs to a department instead of to a date. The moment a register is sorted by function, it stops being read by the people who can act on it, because a process engineer cannot resolve a customs line and a finance controller cannot close a floor loading line. The alternative is to sort by the phase in which the risk can still be cheaply influenced. A risk that can be mitigated in the freeze phase and a risk that can only be mitigated during commissioning are different animals, even when they describe the same technical issue. Ranking PCBA factory setup project risks by phase is what turns a document into a schedule with money attached to each line.
Department-sorted registers produce three predictable failures. The first is duplication: the electrical risk appears once under facilities and once under equipment, with different owners and different mitigation dates, and neither entry is complete. The second is orphan lines: anything that crosses two departments — compressed air, network drops, the boundary between what ships and what is installed — tends to be recorded by nobody, because each department assumes the other one logged it. The third is that the register cannot answer the only question a steering meeting actually asks, which is what is due in the next three weeks and who is late on it.
Phase-sorted registers fix all three, because a phase has a start date and an exit criterion. When the exit criterion for the freeze phase is "utility survey complete and signed by the contractor", every risk that depends on that survey is visibly blocked until it happens. The sorting also makes the register short. A phase-sorted register for a single-line project typically holds twenty-five to forty live lines, which is a document someone will actually open, rather than the two hundred entries that a template generates and nobody maintains.
A risk line needs five fields to be usable: what could happen, what would trigger it, what it costs if it happens, who owns the response, and the date by which the response must start. The trigger is the field most often left out and the one that does the most work, because a risk without a trigger can only be reviewed, never detected. "Utilities late" is not a risk line. "Contractor has not confirmed the air pressure test date by the fifteenth, which pushes the acceptance test past the crew's departure" is a risk line, because on the sixteenth somebody knows to act.
The owner field should name a person rather than a company, and the response date should be earlier than the date the risk would bite — typically two to four weeks earlier for anything with an external supplier in the chain. The cost field is the one that turns a register into a budget conversation, and it is worth insisting that it is filled with a number even when the number is rough. A rough number that is wrong by half still supports a decision; a rating of "high" supports nothing.
Field | What it must contain | Common mistake |
|---|---|---|
Event | One sentence describing what happens, in the past tense as if it already did | Naming a category ("utilities") instead of an event |
Trigger | The observable signal and the date it would appear | Writing a condition nobody can see, such as "if the contractor is slow" |
Exposure | Currency figure: idle labour, expediting, lost output, penalty | Using high / medium / low, which cannot be summed |
Owner | One named person with authority to act | Naming a department, which has no hands |
Response date | The date the mitigation must start to still work | Setting it equal to the impact date, which leaves no runway |
The single most useful change to a risk register is to replace the colour scale with a currency scale, because currency adds up and colour does not. Two amber risks are not obviously worse than one red risk until both are converted to money, and the conversion usually reorders the list. The conversion itself is not hard: idle crew days, expedited freight, rework hours, lost production days at a stated contribution margin, and any contractual penalty all have a price that the finance team already knows.
As an illustration, assume a project with a four-person installation crew at USD 2,400 per day, a contribution margin of USD 900 per production day, and an expediting premium of 25 percent on any air freight. A two-week delay caused by a missing utility costs 14 days multiplied by 3,300, which is about USD 46,000. The same delay caused by a missing documentation set on a shipment costs the same in idle time plus whatever the air freight premium comes to. Once both lines read 46,000, the conversation about which one to chase first stops being a matter of opinion.
Band | Exposure (illustrative) | Review cadence | Escalation |
|---|---|---|---|
Register only | Under USD 5,000 | Monthly | Project manager |
Tracks the schedule | USD 5,000 to 25,000 | Fortnightly | Project manager, reported to sponsor |
Threatens a gate | USD 25,000 to 100,000 | Weekly | Sponsor, with a dated mitigation plan |
Threatens the business case | Above USD 100,000 | Every meeting | Steering committee, options paper required |
The freeze is where most PCBA factory setup project risks are created, in the sense that a decision taken in this phase determines whether a later risk can still be mitigated at reasonable cost. A layout that leaves 600 millimetres between a reflow oven and a wall has already decided that the oven's exhaust duct will need a custom elbow, that maintenance access will be from one side only, and that the next oven will not fit. None of that shows up as a risk during the freeze, because the layout is approved as a drawing rather than as a set of consequences. The register's job in this phase is to convert each frozen dimension into the risks it forecloses.
Freeze is not a signature on a drawing. A layout is frozen when three things are true: the equipment list is fixed to model and option level, the utility connection points are fixed to coordinates and capacities, and the ownership of every interface between parties is written down. If any of the three is still moving, the project is not frozen and should not behave as if it were — which means the register keeps the corresponding risks open and the procurement team keeps the option to change.
The practical test is whether a supplier could order long-lead parts from the frozen documents alone. If the answer is no, the freeze is partial, and the risks that depend on it should be dated from the real freeze rather than the announced one. Projects that announce a freeze and then keep editing the layout for six weeks usually discover the consequence in utilities, because the contractor sizes the pipe runs against a drawing that has already changed twice.
Freezing late is not free, and the cost has a shape worth writing into the register. Each week that the layout stays open delays the utility drawings, which delays the contractor's material order, which delays the installation window, which compresses commissioning. The compression is where the money goes, because commissioning that is compressed by two weeks usually loses the ability to run a full-load test, and the full-load test is what catches the issues that only appear when every machine runs at once.
As an assumption for planning, take a project that freezes six weeks late and then holds the original start-of-production date. Installation compresses from four weeks to three, commissioning from ten days to six, and the ramp-up absorbs the rest. If the compressed commissioning misses a defect that later costs two days of production and one rework batch, the direct cost is roughly 1,800 in lost contribution plus the rework. The visible cost is small; the invisible cost is that the register now contains a risk that was never mitigated and will be repeated on the next project. Writing the freeze date into the scope document that the freeze is measured against is the cheapest way to make the slip visible while it is still a slip and not a fact.
Most freezes produce a layout without a utility survey, and the two then disagree. The survey should record, at each proposed machine position, the available air pressure and flow, the available power at the nearest distribution board, the exhaust route and the floor's load capacity. Four numbers per position is enough, and the survey usually takes two days with an electrician and a fitter. Skipping it is the single most common way to convert a layout risk into a utilities risk three months later.
Where the survey and the layout disagree, the layout usually loses, and the cost of that loss is proportional to how much steel has already been ordered. A machine moved one metre to reach a power drop is cheap in the freeze phase and expensive after the conveyors are cut. The register line for this is simple and should exist on every project: "layout approved without a utility survey, positions to be revised after survey", with the exposure being the re-engineering of any conveyor or duct already ordered.
Banning changes after the freeze does not work, because the changes that matter are usually the ones a customer cannot avoid making. What works is a change budget: an agreed number of post-freeze changes, each with a stated cost in schedule days and a rule that the cost comes out of the same contingency. A budget of three changes, each costing five schedule days, is easier to defend than a policy of no changes, because it converts an argument about principle into an arithmetic question about capacity.
The budget also makes the register honest. Once changes are priced in days, the risk of "customer changes the layout after freeze" stops being a vague anxiety and becomes a line with a probability, an exposure and a named approver. It belongs next to the process route decision that determines which equipment appears on the layout in the first place, because a change of route — adding coating, adding through-hole — is the change that consumes the whole budget in one move.
Utilities carry the longest lead times on a typical project and the least tolerance for being discovered late, because they involve work that happens inside the building and therefore competes with everything else the building is doing. A transformer upgrade, a new exhaust riser or a floor penetration is not a purchase that can be expedited by paying for air freight. It is a sequence of site visits, inspections and permits, and its duration is set by parties who have no interest in your ramp-up date. This is the part of the register where the exposure is dominated by waiting rather than by spending.
Three separate risks hide inside the phrase "power is available". The first is capacity: the building may have a supply that is adequate for the offices and the existing line but not for a second line at full load, and the difference only appears when a load calculation is done at the point of connection rather than at the building's main breaker. The second is the queue: an upgraded supply or a new transformer is a utility company process with its own timetable, and in many locations the wait is measured in months and cannot be bought out.
The third is the earth arrangement, which is the one most often left off the register because it is invisible. Whether the building provides a protective earth that satisfies the equipment's requirement, whether the neutral is bonded at the transformer, and whether the line's own earth bar is connected by a conductor sized for the fault current are three questions that produce a fourth risk: the electrical inspection cannot pass until they are answered. All three should be written as separate lines with separate triggers, because they resolve through three different parties.
Compressed air generates two risks that are easy to under-price. The first is quality: a ring main that was fine for hand tools may deliver air with a dew point or an oil content that the machine's pneumatic valves and sensors will not tolerate, and the failure shows up as intermittent sensor faults rather than as a clear breakdown. The second is timing of the last meter: the machine needs a regulated, filtered connection at its inlet, and the party who installs the ring main rarely considers the last two metres to be part of the job.
The make-up air trap is the one that reorders budgets. Every cubic metre extracted by reflow and wave exhaust has to be replaced, and if it is replaced with untreated outside air, the hall loses its temperature and humidity control in the season when the outside air is worst. As an assumption for planning, two ovens extracting 500 cubic feet per minute each require around 1,000 cubic feet per minute of replacement, which is roughly 28 cubic meters per minute — a quantity that usually needs conditioning, and conditioning needs plant capacity that was not in the original budget. The register line should carry the plant capacity as the exposure, not the ductwork.
Utility risk | Trigger to watch | Lead time assumption | Cheapest mitigation |
|---|---|---|---|
Supply capacity insufficient at full line load | Load calculation not signed by the utility before layout freeze | 8 to 24 weeks for an upgrade | Load study in the freeze phase, before any steel is ordered |
Electrical inspection fails on earthing | Earth resistance measurement not scheduled | 2 to 6 weeks to remedy and re-inspect | Measure during installation, not at handover |
Air quality unsuitable for valves and sensors | No dew point or oil class stated in the survey | 2 to 4 weeks for dryers and filters | Specify air quality in the utility scope, not the equipment scope |
Make-up air capacity missing | Exhaust volume known but no replacement-air calculation | 6 to 14 weeks for conditioning plant | Calculate replacement air when the exhaust volume is fixed |
Floor point load above slab capacity | Machine weight list not compared to structural note | 4 to 12 weeks for reinforcement | Structural check at layout freeze, using crate and foot loads |
Floor loading is a risk that hides behind an average. A machine of 3,000 kilograms on four feet averages 750 kilograms per foot, but an uneven distribution can put 1,400 kilograms on one foot, and the slab has to be checked against the higher figure. The same applies to the route: a floor that carries the machine in its final position may not carry the crate that delivers it, because the crate spreads load differently and the pallet truck concentrates it. Both numbers belong in the register, taken from the installation drawing rather than from a quotation.
Access is the second half of the same line. Door width, corridor width, turning radius and lift capacity are four measurements that decide whether a machine enters the building through the door or through a roof opening, and the difference between those two outcomes is measured in weeks and in crane hire. The register line should name the person who walked the route with a tape measure, because a route that was never walked is a route that will be discovered on the delivery day.
Exhaust work usually triggers a permit, and the permit usually gates the inspection that gates energisation. That chain is worth drawing explicitly on the register, because each link looks small on its own and the chain as a whole is often the longest single path in the project. A coating room with solvent handling adds a fire review; a wave solder machine with flux adds an emission question; either can add weeks that no amount of supplier effort can compress.
The mitigation that works is to treat the permit list as a deliverable of the freeze phase rather than of the installation phase. Once the list exists with its lead times, the register can show which permit is on the critical path and which is not, and the project can decide to sequence installation around the long one. Projects that run beside an existing building rather than in an empty shell carry an extra version of this risk, and the way that changes the plan is set out in projects that run beside a live line rather than in an empty hall — the permit is the same, but the window in which the work can happen is not.
This is the phase where the register earns its reputation, because most of the PCBA factory setup project risks that get discussed in meetings originate here and most of the ones that actually hurt originate in the two phases before. Interfaces, customs and installation share a property: they are all about things crossing a boundary, and boundaries are where information is lost. A machine crossing a border, a board crossing from one conveyor to the next, a responsibility crossing from a supplier to a contractor — each is a place where the plan assumes a handover that nobody scheduled. The register's habit in this phase should be to ask, for every crossing, what evidence exists that it works, and to treat "we discussed it" as no evidence at all.
Every interface that has no named owner generates a risk line automatically, and the way to find them is to walk the equipment list against the utility survey and ask, for each connection, who confirms it and with what evidence. That exercise is described in detail in the interface rows that generate most of the technical risk lines, and the output is a list of rows that either have one accountable party or do not. The ones that do not are the risk lines; the ones that do are controls.
Two interface risks deserve their own lines on every project regardless of size. The first is the conveyor handshake between two machines from different vendors, which is standardised enough at the signal level — the SMEMA interface exists precisely to make that connection uninteresting — but which still depends on a height, a direction and a board-transfer confirmation that somebody has to check. The second is the earth and neutral reference between the line and the building, which produces faults that appear as random behaviour and are diagnosed by measuring rather than by reading a drawing.
Customs risk is almost never about duty. It is about a document that exists in the supplier's office and not in the broker's file: the certificate of origin, the conformity declaration, the packing list that matches the crates, the local-language manual the destination authority expects. Each missing document adds days, and the days are counted from the moment the shipment arrives rather than from the moment somebody notices.
As an illustration, assume two containers, a demurrage exposure of USD 120 per container per day after the free period, and a documentation gap that takes eight days to close because the certificate has to be reissued by a chamber of commerce. The direct charge is about USD 1,900; the indirect charge is eight days of a crew that has already travelled, which at an assumed USD 2,400 per day is USD 19,200. The mitigation is cheap and boring: a document checklist agreed with the broker before shipment, with the supplier's copy of each item attached.
The installation sequence is a risk source because it is usually written by the supplier for the equipment and by the contractor for the building, and the two sequences are rarely merged. The equipment sequence assumes a floor, power at the wall and a route in; the building sequence assumes the floor is finished before anything is placed on it. Where the two collide, the collision is predictable: machines arrive before the coating on the floor has cured, or the floor is finished after the machines are positioned and the finish is never applied under them.
Merging the two sequences into one dated plan, with the building activities shown alongside the equipment activities, is worth more than any amount of coordination meetings. Where the line arrives as a matched package rather than as loose machines, the equipment side of that plan gets shorter, which is the practical argument for an assembled line where the machines arrive already matched to each other — fewer joints to sequence means fewer joints to get wrong. It does not remove the building dependencies, and the register should not pretend that it does.
Commissioning has two distinct halves and only the first is usually planned. The first half is functional: does each machine do what it was bought to do, with the program and the material it will actually run. The second half is systemic: do all the machines do it at the same time, at the intended rate, with the conveyor handshakes and the exhaust and the air all working together. The second half is where the risks live that no factory acceptance test can catch, because a factory test runs one machine in a supplier's hall.
The mitigation is to schedule a full-load run as a dated event with its own exit criterion, rather than treating it as something that happens once production starts. A useful exit criterion is a stated number of consecutive boards through the whole line with no manual intervention, at a stated cycle time, with the utilities at their design values. If that event is not on the register, it is not in the plan, and the risk of discovering an interaction problem during the first customer order stays open by default.
Readiness is the phase that looks like it costs nothing and quietly determines whether the ramp-up is measured in weeks or in months. The equipment can be installed perfectly and still not produce, because the feeders are loaded for a BOM revision that arrived last Thursday, the program has the wrong panel offset, the stencil is at the wrong tension, or the operator who knows how to recover a placement fault has not been hired yet. Each of those is a risk line with a trigger and a date, and none of them appears in an equipment delivery schedule.
Material risk begins with the question of what will be run on the line during validation, because that answer determines the feeder count, the nozzle set, the stencil order and the amount of component stock that has to be on site. A validation build that uses production material is the most realistic and the most fragile, because any shortage on the customer's side stops the acceptance test. A validation build that uses a dedicated kit is more robust and less representative, because the kit was prepared with extra care by people who knew it would be inspected.
The register should carry both options and their consequences rather than choosing one by default. For a project that will commission multiple lines, an intermediate path is to validate each block with a kit and validate the hand-off between blocks with production material, which is a smaller and more forgiving use of the customer's stock. Where the through-hole block is short, the material risk is concentrated in a smaller number of part numbers, which is one of the quieter advantages of a simple DIP assembly line, where the through-hole block is short enough to commission in days rather than in weeks.
Program readiness is three deliverables that are usually produced by three different people at three different times: the stencil, the machine program, and the feeder list that tells the line where each part lives. The failures are predictable. The stencil arrives from a vendor who used the previous revision of the paste layer. The program is generated from a CAD export that has a different origin than the panel drawing. The feeder list is built from the BOM but the BOM was sorted by reference designator, so the 0402 capacitors are spread across eleven positions and two of them are duplicated.
Each of these is a risk line, and each has a cheap early trigger: the stencil artwork should be checked against the paste layer on the day the order is placed; the program should be verified against a known-good board before the line is released; the feeder list should be sorted by package and value and reviewed against the BOM by the person who will actually load the machine. A feeder list that was never reviewed by a loader is the most common cause of a first-build stoppage that lasts half a shift.
People risk has the longest latent period of any line on the register, because hiring and training cannot be compressed by spending money on expediting. An operator who has never run this class of machine needs weeks of supervised production before they can be left alone on a shift, and a maintenance technician who has never seen the platform needs longer, because the competence being bought is diagnostic rather than procedural. Projects that hire in week one and train in week nine are not training late; they are choosing to run the ramp-up with unqualified staff.
The register line should state the competence required at the first production order rather than the headcount required, because the two are different numbers. Three operators who can each recover a placement fault unaided are worth more than five who can start a job. The mitigation with the best return is usually to send one person to the supplier's factory for the acceptance test, so that the person who will train the others has seen the machine run before it arrives.
Readiness gate | Evidence that closes it | Owner | Typical slip if untracked |
|---|---|---|---|
Validation material on site | Kit list reconciled against the BOM revision in use, counted and stored | Customer materials | 2 to 4 weeks |
Stencil approved | Artwork checked against the current paste layer, tension and frame recorded | Process engineer | 3 to 10 days |
Program verified | First article placed at nominal coordinates on a known board, offsets recorded | Programmer | 1 to 2 weeks |
Feeder list reviewed | List sorted by package, checked against BOM, loaded and counted by the operator | Line leader | Half a shift per build |
Operators competent | Named operators signed off on start, stop, recovery and changeover | Production manager | 3 to 6 weeks |
Maintenance cover | Named technician trained on the platform, spares list stocked | Maintenance manager | Long, and invisible until a breakdown |
A gate only works if it can actually stop something, and the readiness gate is the one most often written as a checklist that nobody is empowered to fail. The version that works has three properties: it is scheduled before the first customer order rather than before the first shipment, it has a named person who can refuse to release the line, and its failure has a defined consequence — the order moves, not the gate. Without the third property the gate is a ritual, and rituals do not prevent ramp-up surprises.
The content of the gate is the six rows in the table above, and the useful discipline is that each row is closed by evidence rather than by assertion. A list that says "operators trained" is an assertion; a signed record naming four operators and the four tasks each was assessed on is evidence. The difference matters most in the first month of production, when the person who signed the assertion has moved to another project and the person running the line has to reconstruct what was actually verified.
Validation is where the register converts into evidence, and the trap is to treat acceptance as a single event. It is a sequence of three events with three different scopes, and each one proves something narrower than the project team usually assumes. Understanding what each event covers — and what it explicitly does not — is what allows the PCBA factory setup project risks around performance and capability to be closed honestly rather than administratively. A signed acceptance certificate that nobody can trace to a measurement is worse than no certificate, because it closes a line that is still open. The tests themselves are not complicated. What makes them useful is deciding, before the week they happen, exactly what each one is allowed to prove and what it is not.
A factory acceptance test proves that a machine, in the supplier's hall, with the supplier's material and utilities, does what the specification says. A site acceptance test proves that the same machine, in your building, connected to your services, does what the specification says. The gap between them is the entire reason both exist: everything that changes when the machine moves — the power quality, the air quality, the floor, the exhaust, the ambient temperature — is outside the scope of the factory test and inside the scope of the site test.
The register should therefore carry two separate lines for acceptance, not one, with the second owned by the customer rather than the supplier. The most common failure is to plan the factory test in detail and treat the site test as a formality to be completed on the last day, which means the site test is run without the material, the people or the time to do it properly — and a test that cannot fail is not a test.
Three boards cover most of what a site acceptance test needs to demonstrate. The first is a capability board: fine-pitch and small-chip features at the limit of what the product family requires, used to prove placement accuracy and print quality. The second is a coverage board: a board that exercises every feeder position, every nozzle and every conveyor path on the line, used to prove that the configuration matches the BOM. The third is a stress board: the most demanding real product, run in a quantity that exposes thermal and handling issues rather than a single pass.
Preparing those three boards is a small project of its own and belongs on the register with a date. Boards that are designed on the day of the test cannot test anything, and boards borrowed from another product family usually miss the one feature that matters. The useful discipline is to define the three boards at the same time as the acceptance criteria, so that the criteria are written against something that will exist.
A pass or fail result tells the project whether to sign, and it tells the process engineer nothing about how much margin exists. Capability indices — the Cp and Cpk of a measured characteristic such as paste volume or placement offset — tell you whether the process is centred and whether it has room before it starts producing defects. A process at a capability of 1.0 passes most acceptance criteria and will produce defects at a rate that makes the ramp-up painful; the same process at 1.3 gives the ramp-up somewhere to live.
As an assumption for illustration, take a paste volume target with a tolerance of plus or minus 10 percent and a measured standard deviation that puts the process at a capability of 1.0. Raising it to 1.3 requires either a 23 percent reduction in variation or a wider tolerance, and the first is usually achievable by changing the squeegee pressure and the separation speed rather than by buying anything. That arithmetic is worth doing during acceptance, because the same improvement is far more expensive to find later, under production pressure, with a customer waiting.
Ramp-up risks are unusual because they are the only ones that appear after everybody has declared success. The line is installed, accepted, staffed and producing — and the output is still not what the business case assumed, because yield is 6 percent below plan, because changeovers take twice as long as the demo, because the first shift of every week is spent recovering from the weekend. These risks have one thing in common: they scale with volume and with variety, so they are invisible during a single-board acceptance test and unavoidable at the five-thousandth board. They are also the risks that the project team has usually stopped looking for, because the milestone list says the project is finished and the language has already shifted from installation to operations.
A ramp curve is not a straight line from zero to full output, and pretending that it is produces a plan that is wrong in the first month and never catches up. The curve has three usable gates. Gate one is first article: the line produces one panel that passes inspection. Gate two is stable output: the line produces a stated quantity on consecutive days with no manual intervention beyond the defined allowances. Gate three is planned capacity: the line hits the output the business case assumed, at the yield the business case assumed, for a full week.
Each gate has an exit criterion and a slip cost, and both belong on the register. The useful discipline is to state the yield and the output together at each gate, because a line that hits gate three on output but misses on yield has not reached planned capacity — it has reached planned volume and is buying the difference with rework. Projects that separate those two numbers tend to discover the gap in month three, when the rework hours have already been absorbed into the cost of goods and nobody can say where they came from.
Gate | Exit criterion (illustrative) | Typical duration | Cost of slipping one week |
|---|---|---|---|
First article | One panel passes inspection against all criteria, offsets and paste volume recorded | 1 to 3 days after site acceptance | Low, unless it delays site acceptance |
Stable output | Stated quantity on three consecutive days, yield above the agreed floor | 1 to 3 weeks | Lost contribution on planned volume, plus overtime |
Planned capacity | Business-case output and yield held for a full week, changeovers at planned time | 2 to 8 weeks | Direct shortfall against the ramp assumption |
Two numbers determine whether a slow ramp is expensive or merely inconvenient: the yield gap and the available time. A ramp that starts at 92 percent yield against a plan of 97 percent is losing 5 percentage points of sellable output, which on a line producing 20,000 boards a week at a contribution of USD 4 per board is USD 4,000 per week. A ramp that also runs at 70 percent availability instead of 85 percent loses fifteen percent of its time before the yield loss is counted at all, and the two effects multiply rather than add.
As a further illustration, assume the same line at 70 percent availability and 92 percent yield reaches an effective output of 64.4 percent of its rated capacity, against a plan that assumed 85 percent and 97 percent, which is 82.5 percent. The gap of 18 points on a rated capacity of 20,000 boards is 3,600 boards per week — the number that should sit on the register as the exposure of the ramp-up phase, in the same currency as the utilities and customs lines. Putting it there is what allows a project to compare spending USD 8,000 on a training week against losing USD 14,400 a week for six weeks.
Closing the register is not the same as deleting it. The lines that were mitigated, the lines that materialised and the lines that never moved all carry information about how the next project should be planned, and the cheapest moment to capture it is while the people who lived through it are still on speaking terms with the project. A short close-out with three questions per line — did it happen, did the mitigation work, and what would have detected it earlier — produces a list of triggers that is worth more than the original risk assessment.
The practical output is a one-page "risks that actually bit" sheet, attached to the project file and to the next project's kickoff pack. Six to ten entries is typical for a single-line project. The next project's register can then open with those lines pre-populated and pre-dated, which is the only form of organisational learning that survives a change of personnel. Without it, every project rediscovers the same five surprises in the same order, and the PCBA factory setup project risks that were supposed to be managed become folklore instead.
Send the layout and the utility constraints before the freeze
If you are planning a line and the freeze is approaching, the most useful thing you can send is the layout with the utility connection points marked, plus the equipment list with model and option detail. We will review it against the risks we see most often at that stage — floor loading versus crate weight, exhaust against make-up air, the last meter of compressed air, and the conveyor handoff between blocks — and tell you which lines are still open. It is a desk review, not a survey, and it usually takes a few days. The output is a short list of questions to close before the drawing is signed, which is the cheapest point in the whole project to close them. What we ask for is deliberately small: one drawing, one equipment list, and a note of anything already ordered, so the review happens this week rather than after the steel arrives.
At the same time as the layout, not at the same time as installation. Most of the PCBA factory setup project risks that later cost money are decided during the freeze, so a register started after the equipment arrives can only record them rather than influence them. The first version should be short — ten to fifteen lines covering the freeze, the utility survey and the customs documents — and it should be reviewed against the layout before the drawing is signed. Expanding it later is easy; recovering the decisions it would have changed is not.
Between twenty-five and forty live lines is a practical range for one line, and it is deliberately smaller than what a checklist template generates. Above forty, the register stops being read weekly and becomes a document that is updated before audits. The number matters less than the ratio: every line should have one owner and one date, and any line without both should be deleted rather than carried, because a line nobody owns is a line that will be rediscovered as a surprise. Pruning is part of maintaining the register, not a sign that it is being neglected.
In money, with days shown alongside as the driver. Days are easier to argue about and money is easier to rank, and the conversion between them is where most of the useful conversation happens. A line that reads "six days of installation crew at 2,400 per day plus twelve days of lost contribution" can be compared with a line that reads "expedited freight at 25 percent premium", and the comparison usually reorders the list. Keeping both figures also makes the register readable by finance without translation.
The customer owns the outcome and the contractor owns the installation, and the supplier owns the consumption figures that both depend on. Most utility risks exist because those three responsibilities are not written down against the same values, so the same interface is designed twice or not at all. Putting the consumption figure in the purchase order as a named deliverable, and putting the acceptance evidence in the contractor's scope, closes most of the gap without changing any contract. Where a risk remains, name a single person rather than a company.
Make-up air. Exhaust volumes are specified because the equipment needs them, and the replacement air is rarely calculated by anyone, so the hall either loses pressure control or loses temperature and humidity control once the ovens run. The consequence looks like a paste printing problem and is diagnosed as a building problem, usually after several weeks of adjusting printer parameters that were never wrong. The mitigation is a one-page calculation at the freeze stage comparing extract volume with available make-up capacity.
By defining the ramp gates before handover and making handover conditional on gate two rather than gate one. Gate one is a single good panel, which almost any installation can produce; gate two is a stated quantity on consecutive days at a stated yield, which requires the material, the program and the people to all be ready at once. Projects that hand over at gate one transfer the ramp-up risk to the production team without transferring the budget to manage it, and the production team discovers the gap in the first month of customer orders.
It prevents the ones that are prevented by early action and documents the rest, which is a better outcome than it sounds. A risk that is detected at its trigger date can often be mitigated by spending a small amount — expediting one document, booking an inspection earlier, ordering a spare dryer. The same risk detected at its impact date can only be absorbed. The measurable value of a register is the difference between those two costs across the lines where the trigger fired while there was still runway.