1. Learning as a system, not a catalogue
A catalogue answers "what courses do we offer?" A strategy answers "what must our people be able to do in two years that they can't do today, and how will we know?" The second question starts with the business: a move to model-based delivery, a new plant line, a safety record that has stalled, a generation of experienced staff about to retire. #13 · #14
Three diagnostic layers keep the strategy honest. At the organization level: which outcomes matter, and which gaps in capability stand between you and them? At the task level: which recurring tasks and decisions make up each role, and which are changing? At the person level: who can already do what? Requests that arrive as "we need a Revit course" often turn out, after this diagnosis, to be a standards problem, a workflow problem or a staffing problem that training won't fix. #13 · #34
Learning is also socio-technical. A new tool, a new procedure or a new delivery method changes roles, decision rights, data, incentives and support all at once. Training that ignores those changes teaches people to work in a system that doesn't exist. Plan the operating change and the learning together, and expect a temporary dip in productivity while people adjust. #63 · #11
Finally, a strategy is a loop, not a launch. The same closed loop that quality systems use applies here: plan, deliver, check what changed, fix the cause of what didn't, and learn. Most learning programmes run the first two steps and skip the rest. #64 · QMS-ARC-01
Consider a typical request: "Our site teams need training on the new inspection app." Diagnosed at the organization level, the goal is fewer missed hold points and faster closeout. At the task level, only three steps actually change: recording an inspection, attaching evidence and releasing a hold point. The person level shows that most supervisors already use the app's photo features and few know the release workflow. The result is a 40-minute practice session on real inspections, timed for the week before go-live, plus a one-page job aid. It isn't a day-long course, and it costs a fraction of what was asked for. #13 · QMS-ITP-02 · #22
2. What works, and what to retire
The strongest evidence for training adults in technical work comes from cognitive and instructional psychology, not from neuroscience. Five findings are robust enough to design around:
- Retrieval: practising from memory, with feedback, beats reviewing material again. For technical skills, that means doing the task, not answering questions about it. LSC-PRC-01
- Spacing: practice spread over days and weeks beats one long session; the gap should grow with how long the skill must last. LSC-PRC-02
- Worked examples that fade: novices learn more from studying complete solutions, then partly completed ones, than from solving problems unaided. LSC-CLT-01
- Guidance matched to prior knowledge: the support that helps beginners hinders experienced people, so diagnose first and run separate tracks. LSC-CLT-02
- Practice that matches the job: for fixed procedures, training that uses the real files, templates and equipment transfers best. LSC-TRF-02
Two cautions keep these honest. Most of the evidence comes from students learning words or mathematics, so effects for complex field and software skills are probably smaller. And the methods that work best often feel harder, so satisfaction ratings will reward the wrong designs. LSC-EVD-02 · LSC-PRC-05
Several popular claims should come out of your materials and your vendors' pitches:
- Learning styles. Matching instruction to "visual" or "hands-on" learners doesn't improve learning. Choose the format the task needs. LSC-MYTH-01
- "Rewires the brain." All training changes the brain, including bad training; it isn't evidence of anything. LSC-MYTH-02
- "Only 10% of training transfers." A rhetorical opener from 1982, never a measurement. LSC-TRF-01
- The 10,000-hour rule. An average from one study of violinists, turned into a threshold. In professions, accumulated practice hours explain under 1% of performance differences: experience isn't the same as deliberate practice. LSC-EXP-03 · LSC-EXP-01
What does a design built on the robust findings look like? Take a two-week programme on model coordination for project engineers. Day one: study two complete clash-review records from past projects, then a third with the decisions blanked out. Day two: run a live clash review on a real model, from memory, with a coach. Then, instead of a second day in the classroom, three 20-minute refreshers at the end of weeks one, three and eight, each a new review on the engineer's own project. Experienced coordinators take a short diagnostic review on day one and skip the worked examples if they pass. None of this needs new technology; it needs a different schedule. LSC-CLT-01 · LSC-PRC-01 · LSC-PRC-02 · LSC-CLT-03
The practical consequence: build your method on the robust findings, label weaker claims for what they are, and ask every vendor for learning results on real tasks rather than vocabulary. LSC-EVD-01 · CM-EVD-01
3. Building capability across disciplines
The skills that matter most rarely sit inside one discipline. A BIM coordinator needs information management, contracts and coordination. A plant technician now needs controls, data and security. A project manager needs estimating, procurement and people. A learning strategy that mirrors the org chart will miss the seams where work actually fails. #20 · #47
Four capability shifts show up across the sectors in the Database:
- From tool operation to information management. BIM job ads still over-weight software proficiency, while the scarce skill is running the process: requirements, acceptance, the common data environment. Build competence in information management and carbon accounting alongside the classic credentials. BM-ROL-04 · CM-WRK-05
- From operator to technician. Automation removes routine operator tasks and raises demand for people who combine controls, data and troubleshooting. Put that training into every automation business case, not after it. DM-WRK-04 · DM-WRK-01
- From doing to verifying. As AI drafts more of the work, the critical skill becomes finding what's wrong. Train people to find planted errors and assess whether they check, not how they prompt. PRM-LRN-02 · PRM-LRN-05 · #45
- From experience in heads to knowledge in systems. With retirement waves across facilities, trades and engineering, capture how experienced people actually do the work while they're still there, and pair them with newer staff. #18 · FAM-WRK-01
Map these into a competency framework that goes beyond technical operation: judgment, communication, explaining decisions, ethical use of AI and data. Tool proficiency alone doesn't show readiness for professional work. #36 · #42
A practical way to find the seams is to trace one recurring failure back through the teams it touched. A late steel change, for example, may involve a design team that issued without coordination, a procurement team that didn't flag the lead time, a site team that found out at delivery, and a commercial team that missed the notice deadline. Each team's own training covered its own step. Nobody was trained on the handoffs. Cross-disciplinary learning means building scenarios, rotations and joint reviews around those handoffs, so a site engineer has sat through a procurement review and a buyer has walked a pour. #20 · VCM-ADM-01 · CM-RSK-02
4. Choosing formats
Format follows the task. A few rules from the evidence:
- Decision simulations with structured debriefs suit judgment-heavy capability: coordination, escalation, go/no-go calls, crisis decisions. The debrief is where the learning happens; a game without one is entertainment. #61 · GL-01
- VR and other extended reality work about as well as conventional training, with a small edge for physical tasks. Use them where real practice is dangerous or impossible. LSC-MOD-01
- Short modules are a good delivery format for spaced refreshers and job aids, not a proven method in themselves. LSC-MOD-02 · #22
- Step-level feedback is what makes tutoring, human or software, effective. Systems that only mark the final answer are much weaker. LSC-MOD-04
- AI tutors should hint and check, never hand over the answer during learning. LSC-MOD-03 · GL-17
- Gamification has small, conditional effects. Points and badges aren't a learning mechanism; wrap them around real practice if you use them at all. LSC-MOD-05 · GL-12
- Communities of practice keep fast-moving skills current between formal updates, especially for AI tools that change monthly. #33
Whatever the format, build it on your own standards and templates, and time it close to when people will use the skill. Software training delivered months before a rollout largely evaporates. #16 · LSC-TRF-03
Most strategies end up with a blend: a short, shared foundation for everyone, simulations for the judgment-heavy roles, on-the-job practice with coaching for the procedural ones, and a community that keeps it current. The mix is right when each piece can name the task it serves. #61 · #53 · #33
5. An 18-month roadmap
A roadmap turns the strategy into sequenced commitments. This one is deliberately simple; adapt the timing to your size.
- Months 0–3: Diagnose. Agree the three to five business outcomes learning must serve. Map roles to tasks and identify the critical and changing ones. Assess current capability with job-like tasks, not self-ratings. Record baselines. Identify single points of expertise and start capturing their knowledge. #13 · #14 · #18
- Months 3–6: Pilot. Pick one bounded, representative team and one or two programmes. Build them on the robust methods: worked examples, retrieval by doing, spaced refreshers, real templates. Put training ahead of any related rollout. Measure against a comparison group. #1 · #12 · #52 · LSC-MSR-05
- Months 6–12: Scale what worked. Expand only the programmes with evidence. Train managers to coach and reinforce. Build communities of practice. Add onboarding tracks using the same design, so new staff enter the system rather than around it. #53 · #33 · #25
- Months 12–18: Sustain and renew. Assign owners to curate, update and retire content. Scan 12 to 36 months ahead for technology and regulatory change and update skill maps before demand arrives. Connect learning paths to career progression so people can see where development leads. #57 · #21 · #55
Fund it properly. Communications, coaching, governance and reinforcement cost money, and programmes budgeted only for content tend to stall at completion. Visible leadership commitment, including protected learning time, is a recurring difference between programmes that change behaviour and those that don't. #9 · #7 · QMS-LSS-02
Two warnings for the roadmap. First, resist scaling before the pilot has a result; enthusiasm from the pilot team is not a result. Second, expect the roadmap to change. A new owner requirement, a plant expansion or an AI tool that changes a workflow can reorder priorities overnight. Review the roadmap quarterly against the business outcomes you set in the first three months, and drop work that no longer serves them, however much has been invested. #1 · #57 · #14
6. Measuring and governing
A learning strategy is credible only if it reports results the business recognizes. Use the four levels as a taxonomy, but stop at reaction only if you have to: performance on job-like tasks after 30 to 90 days and observed use on the job are what matter. #10 · LSC-MSR-03 · LSC-MSR-04
Prefer objective records to self-report: model audits, inspection results, rework, nonconformances, system logs, supervisor observation. Use comparison groups or staggered rollouts where you can, because pre-post designs can't separate training from everything else that changed. LSC-MSR-05 · QMS-ARC-04
Be clear about learning data. Tell people what is collected, who sees it and whether it affects assignments, and keep AI-generated recommendations explainable and contestable. Trust in the system is a precondition for honest participation. #48 · #50
Governance doesn't need a large structure. A small learning council works well: an operations leader who owns the business outcomes, a technical lead from each major discipline, the person who runs learning, and someone who speaks for the frontline. It meets quarterly, reviews results against the baselines, approves what scales and what stops, and holds owners to their content reviews. Between meetings, each programme has a named owner who answers for its results, not just its completion rate. #7 · #57 · LSC-MSR-01
Finally, publish what you find, internally at least. Rigorous evidence on training in construction, engineering and manufacturing is scarce. An organization that measures its own programmes properly knows more than its competitors, and can prove it. LSC-MSR-06
The short version. Start from business outcomes. Diagnose at three levels. Build on retrieval, spacing, worked examples, two tracks and real templates. Retire the myths. Develop capability at the seams between disciplines. Pick formats by task. Pilot, measure, then scale. Fund the follow-up. Report performance, not satisfaction.