Published 9 September 2026 · Updated 10 October 2026 · By Marie Castelli, Partner, Mentivis
How to build a business case for an AI training system
A business case for an AI training system compares, over three years, the full cost of the project (subscription, integration, configuration, internal time) with the expected gains on a defined scope: training time avoided, content production costs, repeated courses, compliance, effects on business indicators. Each gain rests on an assumption quantified by the company itself, with a low and a high estimate, rather than on market averages.
Pillar: Beyond the LMS: understanding and choosing your training tools
Why the finance department asks for a business case
For a finance department, training looks like a discretionary expense: accounting records the spending, not its return. Effects appear with a delay and their link with results is hard to isolate. The OECD also observes, in its study of 100 company cases (2021), that most companies assess training outcomes with basic methods, mainly questionnaires and observation at work.
An AI training system does not remove these difficulties. It can, however, produce data (time spent, level reached per skill, progress) that makes measurement easier. The business case should therefore present two things: a cautious calculation, and a measurement plan that will make it possible to check it.
Step 1: define the scope
A credible business case covers a limited, comparable scope: a population (a job, a team, a site), one or two priority skills, a time frame (usually three years). Too broad a scope dilutes the effects and multiplies the assumptions.
Preferably choose a scope where a business indicator is already tracked (quality, lead times, complaints, sales). Kirkpatrick Partners recommends starting from the result expected for the organisation, then working back to the behaviours and learning required.
Step 2: measure the current situation
This is the most neglected and the most important step. The business case compares the cost of the system with the total cost of the current situation, not only with the "training" budget line.
| Current cost item | Formula | Internal source |
|---|---|---|
| Content production | Design and update cost per hour of content | Invoices, designers' time |
| Employee time in training | Training hours × fully loaded hourly cost × headcount | HRIS, skills development plan |
| Repeated courses | Share of sessions taken more than once because learning did not stick | LMS history |
| Non-compliance | Expired certifications, incomplete funding files | Certification register, OPCO (the sector bodies that fund training in France) |
| Business effect | Current value of the chosen indicator | Operational dashboards |
Do not use an average "market" cost for these items. Your own invoices and your own time data are more defensible in front of a finance department than any benchmark.
Step 3: state the gain assumptions
Each gain is an explicit assumption, quantified with a low and a high estimate, and linked to a verification indicator.
| Lever | Assumption to fill in | Verification indicator |
|---|---|---|
| Training time avoided | % of time saved when employees skip what they already master | Time spent per learner before and after |
| Content production | % of production cost avoided | Cost per hour of content |
| Repeated courses | % fewer repeated sessions | Enrolment history |
| Compliance | Number of expiries avoided × cost of an expiry | Certification register |
| Business effect | Change in the indicator attributable to the programme | Comparison with a control group |
Worked example with fictitious values. A scope of 200 employees takes 14 hours of training per year, at a fully loaded hourly cost of €40. Training time represents 200 × 14 × 40 = €112,000 per year. With a low assumption of 5% and a high assumption of 15% of time avoided, the annual gain lies between €5,600 and €16,800. These percentages are illustrative: each company must set its own, ideally from a pilot.
Step 4: cost the full project over three years
The cost of the system is not limited to the subscription:
- Subscription over the commitment period, depending on the pricing model (per active user, per licence, per tier, flat fee).
- Integration with the HRIS, single sign-on (SSO) and existing tools. Ask for a contractual commitment rather than an indicative estimate.
- Configuration of skills frameworks and learning paths.
- Internal time of the training, HR and IT teams and of subject-matter experts.
- Training of administrators, trainers and managers.
- Upgrades and support over the period.
Cost items and pricing models are detailed in How much does an AI training system cost for a company?.
Step 5: calculate, test sensitivity and plan the measurement
The calculation follows the logic of the Phillips method, promoted by the ROI Institute:
- ROI (%) = (gains − costs) ÷ costs × 100, over the chosen period
- Benefit-cost ratio = gains ÷ costs
- Payback period: the month from which cumulative gains exceed cumulative costs
Present three scenarios (low, central, high) and identify the two or three assumptions to which the result is most sensitive. If the project only pays off in the high scenario, the business case should say so.
Finally, plan from the outset how each assumption will be verified: a pilot on one group, comparison with a group without the system, reviews at six and twelve months. See How to measure the ROI of a training plan.
Which risks should be included?
- Insufficient adoption. Without use, no gain materialises. Plan an active-use indicator.
- Data quality. Incomplete skills frameworks or inaccurate HR data distort personalisation and measurement.
- Vendor dependence. Check that data (results, learning paths, content) can be exported in a standard format.
- Data protection. See GDPR and training.
- AI regulation. See AI Act Article 4.
What to avoid
- Comparing the cost of the system with the current training budget alone, without including employee time or production costs.
- Presenting a single ROI figure with no range and no assumptions.
- Reusing "average" gains published by vendors without knowing the method, the sample or the comparison group.
- Adding up overlapping gains (for example counting the same time saved twice).
Frequently asked questions
Sources
- ROI Methodology, ROI Institute, accessed 2026
- Return on Investment in Training and Performance Improvement Programs, 3rd edition, Phillips, Phillips and Toes, Routledge, 2024
- The Kirkpatrick Model, Kirkpatrick Partners, accessed 2026
- Training in Enterprises: New Evidence from 100 Case Studies, OECD, 2021
Further reading
- What is an LMS? Definition, features and pricing
- LMS, LXP, adaptive learning, AI training engine: how to tell them apart
- The LMS is no longer enough: what an AI training system adds
- E-learning completion rates: what the research really says
Related products: MentivisOS Entreprise, MentivisOS Education