AI Training Has Become
a Solution Looking for a Problem
For the last two years, I have been running AI training for large companies.
Thousands of employees have attended sessions on ChatGPT, Claude, Gemini and Copilot.
They have learned prompting, research, data analysis, presentations and image generation.
They have seen Projects, Skills, agents and automation.
Yet when I speak to leaders a few months later, many say the same thing.
“We are not seeing the value we expected.”
I see this from the inside. The main issue is not always the quality of the training. It is how companies decide to commission it.
The usual process is simple.
A company decides that it needs to do something about AI. It calls a trainer, books a session and asks employees to attend.
When I ask what they want the training to achieve, the answers are often vague.
- Everyone else is doing it.
- Some early adopters are moving ahead.
- Other employees feel left behind.
- Leaders want everyone to reach the same level.
These are valid reasons for creating awareness.
If the company buys awareness training, it should expect more awareness. It cannot later complain that employees have not automated their work or improved productivity.
Most AI training is built around features.
But knowing that a feature exists does not tell an employee how to use it in their work.
- Here is how to write a better prompt.
- Here is Deep Research.
- Here is how to analyse a spreadsheet.
- Here is how to generate a presentation.
- Here is an agent.
These capabilities are useful.
The next morning, that employee returns to meetings, emails, deadlines and existing systems. They may know that AI can analyse data, but they still do not know which data they can upload, what security rules apply, who needs to check the output or how the result fits into the next step.
People are using free versions with usage limits.
Most companies I work with in India have not bought enterprise AI licences for all employees. The tools may not connect with the company’s email, HR platform, CRM or internal data.
Many leaders say that they want employees to “adopt AI.”
Employees are asked to use AI so the company can report that employees are using AI.
- Does it mean that employees should send more prompts?
- Should everyone use AI every day?
- Should a process take fewer hours?
- Should recruitment become faster?
- Should customer complaints close sooner?
- Should the company reduce errors or costs?
Until the company answers these questions, AI usage becomes the objective.
There is also a more difficult question.
- Will they move to more valuable work?
- Will targets increase?
- Will the company reduce headcount?
Employees notice when leaders avoid this discussion. If people believe that explaining their work will help the company remove their jobs, they will not actively help to automate it.
There are two practical ways to start.
Both approaches give the AI initiative a clear direction.
The first is outcome-led.
The company knows the current performance and has a target.
For example, reduce the time taken to close a customer complaint from five days to two.
The second is issue-led.
The company has identified an area of pain but has not measured it properly.
For example, recruiters spend too much time screening CVs, or the sales team spends too much time preparing proposals.
Once the outcome or issue is clear, the company needs to understand the workflows behind it.
“Reduce time to hire” is not one activity.
It includes creating the job requisition, writing the job description, sourcing candidates, screening CVs, communicating with candidates, scheduling interviews, conducting interviews, collecting feedback and preparing the offer.
Each workflow has people, systems, information, decisions and handovers.
Companies need to identify who performs each step, which systems they use, where information is copied manually, where delays happen and which decisions require human judgment.
Most companies have never documented their workflows at this level.
Once the workflow is clear, assess each step.
The right solution may be a prompt, a Project, a reusable Skill, a small automation or an agent.
- Can AI assist the employee?
- Can part of the step be automated?
- Can the whole step be removed?
- Does a human need to make the final decision?
- Is the technology ready now, or should the company review it again in six months?
AI may improve one workflow by 5 percent and another by 50 percent. A few workflows may eventually become almost completely automated.
Training should come after the company has decided how the work will change.
Instead of teaching twenty generic features, teach employees how to use AI in the two or three workflows that matter.
Use the company’s actual processes, systems, policies and examples. Show employees what information they can use, where human review is required and how their role will change.
Using AI well requires practice. Employees need to provide context, review outputs, correct mistakes and improve their instructions.
Measure results at three levels.
- At the business level, measure cost, revenue, customer outcomes or progress against the original objective.
- At the workflow level, measure time, turnaround, errors, rework and manual effort.
- At the behaviour level, measure whether employees are using AI regularly in the selected workflows.
Take a baseline before training. Check application after 30 days. Check the results and barriers again after 90 days.
- Successful prompts can become reusable Skills.
- Repeated workflows can become automations.
- Stable automations may later become agents with humans checking important decisions.
Companies will not become AI-native by training everyone once.
They will become AI-native by improving one workflow at a time.