By Ritvars Mētra, Founder of ReadinessCompassThe training environment has forty customers, twelve materials and three suppliers. Every record is complete, every name is sensible, and nothing is blocked, duplicated or half-finished.
Production has 60,000 customers, four of whom are the same company under different spellings, several hundred materials that should have been discontinued in 2019, and a supplier whose payment terms nobody can explain.
People were trained in the first environment and go live in the second. The gap between them is not a data issue. It is a capability issue that will surface on day one as though it were a system fault.
Transfer requires resemblance
The reason this matters is definitional rather than a matter of preference. In their review of transfer of training in Personnel Psychology, Timothy Baldwin and J. Kevin Ford define the conditions of transfer as the generalisation of material learned in training to the job context, plus its maintenance over time.
Generalisation to the job context is precisely what a sanitised environment prevents. A person who has only ever created an order for a clean customer record has not practised the job; they have practised a simplified version of it that omits the part where judgement is required.
The evidence on practice reinforces this. William McGaghie and colleagues found simulation-based education with deliberate practice outperformed traditional clinical education at a pooled effect size of 0.71 — but the mechanism is repeated attempts at a realistic task. Simulation that removes the difficulty removes the benefit along with it.
What the sandbox quietly teaches
| In training | On day one | What the user concludes |
|---|---|---|
| Search returns 3 results | Search returns 400, several near-identical | “The search is broken” |
| Every customer record is complete | Half are missing a tax field | “The migration failed” |
| Nothing is ever blocked | Credit blocks, quality holds, expired contracts | “It will not let me do my job” |
| One plausible material per scenario | Three obsolete variants of the same part | Picks the wrong one, silently |
| Screens respond instantly | Loaded system, real volumes | “It is too slow to use” |
The third column is the expensive part. Each of these arrives at the service desk as a defect, gets investigated as a defect, and is closed as working-as-designed — consuming support capacity in the fortnight when it is scarcest, and teaching users that raising things achieves nothing.
The fourth row is worse, because nothing is raised at all. Someone picks the wrong obsolete material, the order goes through, and the consequence appears three weeks later in a stock reconciliation.
Why it happens, and what is negotiable
Nobody chooses a bad training environment. It happens because refreshing a training client with realistic data is genuinely difficult: it needs anonymisation for data protection, it is heavy to maintain, it breaks when configuration changes, and it competes for the same infrastructure and the same people as testing.
So the realistic goal is not a full production copy. It is a small amount of deliberately awkward data, which is far cheaper and delivers most of the benefit:
- Twenty customers, not forty — but with two near-duplicates, one on credit hold, and one missing a mandatory field.
- Enough volume in one search path that people have to use filters rather than scrolling. This single change fixes the most common day-one complaint.
- The five real exceptions the team hits weekly, built as scenarios with the data pre-staged so they can be practised repeatedly.
- One genuinely broken record so that finding and reporting a data problem is itself a trained behaviour.
Ask the process owners which five cases cause the most trouble today. They will answer immediately, and those cases should exist in the training environment before any course is delivered.
Two adjacent traps
Refresh timing. If the training client is refreshed after training has been delivered, everything people learned about navigation and data may no longer hold. Freeze the training environment for the training window and communicate any change explicitly.
Shared environments. Twenty people practising in one client will collide — consuming the same stock, locking the same records, completing each other’s half-finished documents. Either give people their own data ranges or accept that the session will teach frustration rather than the process.
Both are small logistical decisions that determine whether an expensive training programme produces capability or only attendance — which is the difference a rehearsal at realistic volume will expose whether or not anyone planned for it.
More on capability before go-live in the ERP Readiness Hub, or start with building the curriculum from roles and tasks.