1. Why onboarding fails
Picture three people starting on the same Monday. A site engineer joins a contractor mid-project. A BIM coordinator moves from another firm, fluent in a different authoring tool. A maintenance technician starts at a plant that is replacing half its controllers this year. All three get a laptop, a safety video, a binder and a tour. Six weeks later, one is doing real work, one is quietly redoing everything the old way, and one is waiting for someone to tell them what to do.
The pattern is familiar because onboarding is usually designed around the organization's administration rather than the job. Forms get completed; capability doesn't get checked. The most common failure is that nobody defines what "ready" means, so nobody can tell whether a new person has got there. The second is that the knowledge the person most needs lives in the heads of a few experienced colleagues and is never written down. That second problem is getting sharper: nearly 40% of facility managers expect to retire within ten years, and the same pattern runs through trades, plants and engineering offices. FAM-WRK-01 · #18
A useful definition: onboarding ends when a person can do the core of their role to the organization's standard, without someone beside them, and the organization has evidence of it. Everything in this guide serves that finish line. It applies to new hires, internal transfers, people promoted into supervision, and the partner firms who join a project halfway through. #25 · #28 · #62
Three ideas run through the rest of the guide. First, design for the person's starting point: a graduate and a fifteen-year veteran need different things, and giving both the same programme makes it worse for both. Second, practise the real work, with the real templates, close to when it will be used. Third, verify before you trust, and measure what people can do rather than whether they liked the course. LSC-CLT-02 · LSC-TRF-02 · LSC-MSR-01
2. Before day one
Most of what makes the first month go well is decided before the person arrives. Start with the role, not the course catalogue. Break the role into its recurring tasks and decisions, and for each one ask what the organization needs to observe when it's done well. A site engineer's tasks might include setting out, reviewing shop drawings, raising requests for information, checking a pour against the inspection plan and briefing a subcontractor. A plant technician's might include isolating equipment, changing a setpoint under change control and reading an alarm history. That list is the backbone of the plan. #13 · #34
Next, find out what the person already knows. A short diagnostic task, done in the first days or even before starting, tells you which parts to skip. The evidence here is unusually clear: heavy guidance that helps novices actually hinders experienced people, a finding first shown with electrical apprentices and PLC programmers. So build two tracks. Newcomers to a skill get demonstrations and worked examples. Experienced hires get problems and reference material, and a conversation about where their old habits will clash with your ways of working, such as drafting habits carried from 2D CAD into object-based modelling. LSC-CLT-02 · LSC-CLT-03 · #49
Then handle the practical set-up so day one isn't spent waiting:
- Accounts and access, named to the person, with only the rights the role needs. For contractors and subconsultants, give named accounts that expire when the engagement ends, never a shared login. NIST-SUP-01 · AGT-03
- Equipment and personal protective gear sized and ready, and site inductions booked rather than "arranged when they get here".
- A named buddy for everyday questions and a named supervisor for decisions. Tell both, in writing, what they're responsible for.
- The real artefacts the person will use: company templates, standards, the project's information requirements, a sample of good finished work. People learn faster from their own organization's materials than from generic examples. #16 · #56
Finally, write down what success looks like at day 30, 60 and 90 and record today's baseline. If you can't say what someone should be able to do in a month, you won't be able to tell whether they can. The 30-60-90 guide and builder in this series turns this into a template. #14
3. The first four weeks
The first weeks should lead with the person and the real challenges of the role, not with a week of slides. Introduce the people they'll depend on and the decisions they'll face, then get them working on supervised versions of real tasks as early as possible. Onboarding programmes that open with the role's actual problems, rather than generic company content, give people a reason to learn the rest. GL-15 · #26
For tasks the person hasn't done before, use the sequence with the strongest evidence in technical training: show a complete worked example, then give partly completed tasks, then independent ones. In a review of 55 studies, learners who studied worked examples outperformed those left to solve problems unaided, with the typical example-learner doing better than about two in three of their peers. For a BIM coordinator that might mean watching a federation set up, then finishing a half-built clash test, then running one alone. For a site engineer it might mean reviewing a completed inspection plan, then completing one for a simple pour, then drafting one from scratch. LSC-CLT-01 · #59
Some things must be clear in week one regardless of role:
- Who can stop work, and for what. New people are the least likely to speak up and the most likely to see something odd. Tell them the stop conditions and that they'll be backed for using them. SOP-USE-03 · CS-06
- Which steps can't be undone: hold points before concrete or grout hides the evidence, safety interlocks, published models, money leaving the company. These are where supervision is mandatory. QMS-ITP-02 · PRM-MFG-01
- Where to find the answer at the moment of need: short, searchable guidance inside the tools people use, not a course they have to leave the work to find. #22
Keep the pace humane. Don't put a newcomer through everything in week one; spread it out, because spacing improves retention and because people need time to use what they've learned. And design the environment so that asking questions is normal. Psychological safety isn't a soft extra: it's the condition under which a new person tells you they don't understand the lifting plan before the lift. #3 · CS-06
4. Practice that sticks
The single most reliable finding in learning research is that retrieving something from memory strengthens it more than reviewing it again. Across 118 studies, practice testing beat restudying by a moderate margin, and by more when learners got feedback. For technical roles, retrieval doesn't mean a multiple-choice quiz. It means doing the task from memory: setting up the model, isolating the pump, completing the permit, then getting feedback on what was missed. LSC-PRC-01 · #19 · #60
Spread that practice out. Short refreshers at roughly one, three and eight weeks after first learning a skill, timed where possible to when the person actually uses it on a project, are far better than one long session. Don't pay extra for "optimal spacing algorithms": evenly spaced refreshers work as well as expanding schedules. LSC-PRC-02 · LSC-PRC-03
Expect the best methods to feel harder. Practising from memory, spacing sessions and mixing problem types all lower how well people perform during training while improving what they remember and transfer. Fluent, easy, massed practice feels better and works worse, which is why people left to pace themselves tend to reread rather than practise. Build the retrieval and spacing into the programme; don't leave them to choice. LSC-PRC-05 · LSC-PRC-06
Two cautions about technology. First, AI assistants can do the work instead of the learner. In a field experiment, students who practised with an unrestricted AI assistant performed better while using it and 17% worse once it was taken away; a guarded tutor that gave hints instead of answers avoided the harm. For new staff, use AI that prompts and checks, and periodically check that people can do the core task with it switched off. LSC-MOD-03 · PRM-LRN-01
Second, VR is a tool for specific jobs. Across studies with real-world outcomes, extended-reality training worked about as well as conventional training, not better, with a small edge for physical tasks. Use it where real practice is dangerous, expensive or impossible: confined spaces, fall protection, crane familiarization, a building that doesn't exist yet. Don't use it to teach software. LSC-MOD-01
Finally, protect the early-career experiences that used to come from routine work. As AI and automation absorb drafting, takeoffs and first-pass checks, juniors lose the repetitions that built judgment. Replace them deliberately with supervised cases, peer review and explanations of why. #54 · DM-WRK-04
5. Verify, then trust
Before someone works alone on anything consequential, confirm they can do it and interpret the result. That means watching them do a representative task to the organization's standard, not checking that a course was completed. For model-based work, it includes whether they understand what the model is telling them and who's accountable for it, not just whether they can operate the software. #28 · BM-ROL-04
The person who most shapes whether training turns into practice is the direct supervisor. Supervisor support and the climate of the team consistently predict whether skills transfer to the job, especially for judgment-heavy skills such as coordination and site leadership. Give supervisors a short guide: what the person has practised, what to give them to do next, and what to watch for. #53 · LSC-TRF-04
Let new people make mistakes where mistakes are safe. Error-management training, where learners are encouraged to make and learn from errors in a safe setting, improves transfer to new situations. A sandbox model, a test controller, a mock tender: give them somewhere to break things before anything real depends on them. LSC-TRF-05 · #44
Pair them with experience, in both directions. Mentoring is a retention lever as well as a learning one, and an experienced colleague recording how they actually do the work, while they're still there to do it, is the cheapest knowledge capture available. Newer staff often bring digital fluency the veterans lack; pair them so judgment and new methods both move. FAM-WRK-02 · #18
6. Knowing it worked
Most onboarding is evaluated with a satisfaction survey. Unfortunately, how much people liked a course barely relates to what they learned: ratings mostly reflect the instructor's style and how pleasant the experience was. High satisfaction can even accompany weaker designs, because the methods that work best feel harder. LSC-MSR-01 · #10
Measure what matters instead:
- Performance on new, job-like tasks without help. Can the site engineer complete an inspection plan for a pour they haven't seen? Can the technician isolate an unfamiliar pump correctly? LSC-MSR-03
- Delayed checks. Re-test at 30 to 90 days. Immediate tests mostly capture what people can do in the moment. LSC-MSR-04
- Time to independent work, measured against the baseline you set before day one. #14
- Objective records where you have them: quality nonconformances, rework, model audit results, near-miss reports. Self-reported confidence inflates results. LSC-MSR-05 · QMS-ARC-04
- If you ask people anything, ask how useful it was, not how enjoyable: usefulness ratings track on-the-job use better. LSC-MSR-02
Set the measures before the first person starts, not after. If time to independent work currently averages five months for site engineers, write that down. A programme that brings it to three months has paid for itself several times over, and you can say so. A programme without a baseline can only report that people attended. For small firms that hire a few people a year, keep it simple: a one-page record per starter of when they first completed each critical task alone, and what went wrong on the way. #14 · LSC-MSR-03
Then close the loop. Each cohort's results should change the next cohort's programme: what to drop, what to move earlier, what new people kept getting wrong. Onboarding is a process you improve, not a document you publish. #64 · QMS-ARC-01
The checklist. Define ready. Diagnose the starting point and run two tracks. Use real templates and real tasks. Show, then fade. Practise from memory, spaced over weeks. Make stop authority and irreversible steps explicit. Verify before trusting. Give supervisors a role. Measure performance, later. Improve the next cohort.