Every ERP project has a training phase. It sits on the plan with a start date, an end date, and a completion checkbox. When it is done, someone marks it green and the project moves on. That checkbox is one of the most expensive illusions in enterprise software delivery.
This article is about what that checkbox hides, why it matters, and what a fundamentally different approach looks like in practice — drawn from a real deployment with Constellation Cold Logistics (CCL), a multi-site Infor WMS rollout involving more than 200 users across four languages.
|
BEFORE YOU READ ON If you want to know how AI agents can reduce ERP training timelines by 50–90% — not in theory, but in practice, with real screenshots, real data, and a live client deployment — keep reading. This is not another AI overview. This is cutting-edge and implementation-ready. Knowing this puts you ahead of 95% of ERP practitioners working in the market today. |
Before looking at the process, it helps to understand the scale of what the industry already knows but rarely says out loud.
|
55–75% of ERP projects that miss objectives had insufficient training investment |
6–7 mo typical time to produce training content before go-live |
5–10% of project budget allocated to training vs. 15–20% recommended |
50% of ERP implementations fail on their first attempt |
These are not edge-case numbers. They describe the norm. And they point to the same root cause: organizations treat training as a logistics exercise rather than a competency exercise. Getting people into sessions is not the same as making them ready.
The train-the-trainer model is the industry standard for a reason. When you have a small team of SI consultants who deeply understand the ERP, and hundreds of end users who need to operate it, you need a cascade. You cannot fly consultants to every site for every user.
The model works like this. Consultants train key users — the people within the client organization who will become the internal ERP champions. Those key users receive documentation, attend sessions, watch recordings. They then train their teams. Their teams get trained before go-live. The training phase closes.
Here is what the project plan does not capture.
|
Stage |
What appears on the plan |
What actually happens |
|---|---|---|
|
Consultant → Key User |
Training sessions delivered |
Key users must absorb dozens of process flows — goods receipts, ASN creation, order management, inventory adjustments — across multiple modules, while simultaneously keeping warehouses, production lines, or finance functions running. There is no structured way to practise before go-live. |
|
Material production |
Training content created |
6–7 months of consultant hours to produce materials that may already be outdated by go-live due to system changes. |
|
Key User → End User |
Cascade training completed |
Knowledge degrades at every handoff. The consultant's deep understanding becomes a simplified summary by the time it reaches the warehouse floor. |
|
Competency check |
Training phase: ✅ complete |
No systematic check. Attendance is recorded. Whether anyone can actually process a goods receipt under pressure is unknown. |
|
Post go-live |
Hypercare period |
Gaps surface in production. Floor walkers overwhelmed. Consultants pulled back in. Business slows. |
The cost of this model is not just consultant time. Key users, IT staff, and project managers typically dedicate 25–50% of their working time for 6–18 months across an implementation — internal cost that rarely appears in any project budget, peaking during the training phase.
This is the context in which most WMS go-lives happen. It's expensive, slow, and structurally inefficient — and so normalised that most organisations don't question it. CCL did.
|
|
That three-week window is where the model collapses under its own weight. You have six months of content production, a cascade of decreasing fidelity, and then a narrow window in which large numbers of users across multiple sites need to be trained — often in multiple languages — with no reliable way to verify that any of it has actually embedded before the system goes live.
This is the part nobody discusses in steering committee meetings. The training dashboard shows completion rates. Sessions attended, materials distributed, sign-offs collected. What it cannot show — because traditional ERP training has no mechanism to capture it — is competency.
Competency means: can this user process their day-to-day tasks in the live system, under real conditions, without assistance? Completion means: did they show up?
|
What is tracked today |
What actually determines go-live readiness |
|---|---|
|
Training sessions attended |
User can independently complete their role-specific workflows |
|
Materials distributed |
User has practised the process, not just observed it |
|
Sign-off by manager |
User scores consistently across repeated attempts |
|
Training phase: green |
Department-level readiness is quantified before go-live |
|
Hypercare team standing by |
Users can self-serve guidance without escalation |
Gartner projects that 70% of ERP initiatives will fail to meet business goals by 2027. The competency gap — the space between "trained" and "ready" — is a significant driver of that figure. It is baked into how training has always been delivered, and it has been accepted as a cost of doing business.
The conventional response to poor training outcomes is to spend more on training — more consultant hours, more materials, more sessions. The AI-driven approach starts from a different question: what if the training content cost almost nothing to produce, was always current, worked in any language, and came with a built-in mechanism to verify competency?
|
|
|
|
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This is also why AI-powered testing and AI-powered training are the same capability, not two separate ones. The process recording that drives test automation is the same recording that generates training content. The two phases, historically separated by weeks of re-work, collapse into a single workflow.
CASE STUDY · INFOR WMS
Constellation Cold Logistics (CCL)
CCL is a leading cold storage and logistics operator. Their Infor WMS implementation involved a large, distributed workforce operating across multiple sites, with teams working in four different languages — a scenario that exposes every weakness in the traditional training model.
Under a conventional approach, this would have required separate training streams per language, localized materials, on-site consultants at each location, and months of content production. The scope would have absorbed a significant portion of the project budget and pulled key operational staff away from running cold storage facilities for an extended period.
With Infomind's AI training platform deployed as part of the implementation, the approach changed entirely. SI consultants recorded their sessions — as they would have done regardless. The platform generated role-specific training videos using actual CCL screens and actual CCL operational data. Not generic WMS walkthroughs. The exact workflows CCL staff would use from day one.
Those videos were generated across all four languages automatically. Users across every site accessed the same quality of training content, in their own language, without any additional production effort or re-recording.
From the same source material — and from the test automation scripts produced during the testing phase — the platform generated interactive assessments for each workflow. Each user could watch a video, then enter a guided practice environment. The AI tracked every step: completed correctly, skipped, or done out of sequence. Users received specific, actionable feedback. They could repeat any assessment as many times as needed.
|
users trained across all sites |
languages, zero re-recording |
to full training library, not months |
competency visibility before go-live |
CCL's leadership did not receive a training completion report at the end of the training phase. They received a competency report. The distinction is significant.
For each user: videos watched, assessments attempted, scores per attempt, steps missed, improvement trajectory. For each department: overall readiness score, users who had not yet completed practice, average score against the benchmark threshold. For the project board: a genuine, data-backed answer to the question — are our people ready to go live?
Below is the Driver's License Report from Infomind's platform — a live view of every user's assessment name, top score, and number of attempts. Not a sign-off sheet. Not an attendance register. A per-user, per-workflow competency record built entirely by AI, tracking every click and every data entry the user made during practice.
Driver's License Report — per-user assessment scores & attempt counts
Each row is a real user. The AI records every attempt, every score, and every step — not whether they showed up, but whether they can do the job.
The second view is the Group Driver's License Progress dashboard — a department-level readiness summary that rolls up every user's status into a single report the CIO or project board can act on. Finance at 17% passed. Inventory Management at 17%. Manufacturing, Order to Cash, Production — not yet started. This is the kind of visibility that makes go/no-go decisions based on facts, not assumptions.
Group Driver's License Progress — department-level readiness dashboard
Every department. Every user. Pass rate, attempt count, and readiness status — visible before go-live, not discovered after.
No traditional training programme produces this. There is no spreadsheet, no sign-off sheet, no post-session survey that can tell you Dhinagar attempted the Create WMS ASN assessment twice and scored zero both times, or that Ranjeet has attempted it four times and still has not passed. The AI knows because it tracked every interaction — every field entered, every step skipped, every sequence error. That is not just better reporting. It is a fundamentally different signal about whether an organisation is actually ready.
|
The training phase has never been a tick-box problem. It has always been a measurement problem. The tick-box is what you get when you cannot measure anything better. |
That visibility had not been available before. Not because organisations did not want it, but because the training model could not produce it. When training is delivered through a cascade of human sessions, there is no systematic record of what each individual user can actually do. There is only what they attended.
The cost of poor training is not just the hypercare budget. It compounds.
|
Cost category |
Traditional model |
AI-powered model |
|---|---|---|
|
|
6–7 months consultant and SME time |
Days, from existing recordings and test scripts |
|
|
$500–$2,000 per employee for role-based sessions |
Fixed platform cost, scales to any user count |
|
|
Linear cost increase per language |
140+ languages, no marginal cost increase |
|
|
Key users pulled away from operations for training periods |
Self-paced, on-demand — no operational disruption |
|
|
High — users unable to self-serve, consultants pulled back in |
Low — users have practised, AI provides in-context guidance |
|
|
Re-engage consultants, schedule sessions |
Existing training library, immediate access |
Industry best practice recommends allocating 15–20% of total ERP budget to training and change management. Most organizations allocate 5–10%. The AI model does not just close that gap — it changes the economics entirely. Training content that previously required months of specialist time to produce is now generated in days from artefacts already in the project.
For the system integrator, the shift is significant. The training phase has historically been a delivery risk — content production is slow, quality is variable depending on who produces it, and there is no reliable signal that users are ready until the system goes live and problems surface. AI removes the production bottleneck entirely and creates a readiness signal that the SI can monitor in real time before cutover.
For the client, the change is more fundamental. Training moves from a project phase to a live capability. When a new module goes live, when a process changes, when a new cohort of users joins, the training library updates and is available immediately. The competency tracking that was previously impossible becomes routine.
The consultants who built the system do not spend fewer hours on the engagement — they spend those hours on what actually requires their expertise. Process design, configuration decisions, validation, change management. Not content production.
|
"ERP is 90% about people, process, culture and politics — and 10% about IT." — ERP Experts Online Forum, widely cited across the industry |
If that is true — and the failure statistics suggest it is — then the most important thing an ERP implementation can do is get the people side right. Traditional training has never been built to do that reliably. The approach described here is.
The fear at every go-live is the same. You have spent months building a system. You have a go-live date. You have a hypercare plan because you know users will struggle. You hope the struggle is manageable.
With CCL, that fear was replaced by evidence. Before cutover, the project board had a report showing, for every department and every user, who had practised their workflows, how many times, and how well they scored. That is not a confidence booster. It is a fact-based answer to the question every CIO asks and almost nobody can answer honestly: are our people ready?
The numbers speak for themselves: organisations using AI-powered training report a minimum 50–80% reduction in training production time compared to the traditional model — and in complex, multi-site deployments like CCL, the reduction is closer to 90%. That is not rounding up. That is the difference between six months of content production and a training library ready in days.
The real story is what happens at go-live when users who have actually practised — not just attended — start using the system for the first time. That is where the investment pays out. And that is the part the traditional model has never been able to guarantee.