The Enterprise AI Skills Map: building workforce readiness for generative AI

The paper argued that the scarce enterprise capability was machine fluency joined to domain judgement and responsible execution, and that organisations were buying the first of the three because it was the only one a vendor could supply.

What did The Enterprise AI Skills Map argue?

The paper argued that workforce readiness for generative AI had to be built in four layers, and that most employers were funding only the shallowest of them. Paper Three was written on 31 May 2023, in the weeks after two large surveys landed within nine days of each other, and everything in it describes the position as it stood that spring. The figures below were true of that moment. They are reproduced here as a record of what the evidence supported at the time, and the paper makes no claim about where any of it stands today.

• The World Economic Forum published its Future of Jobs Report on 30 April 2023, drawing on 803 companies across 27 industry clusters and 45 economies, together employing more than 11.3 million workers.
• Employers in that survey expected structural churn of 23 per cent of jobs over five years, made up of growth of 10.2 per cent and decline of 12.3 per cent.
• Those two movements gave 69 million roles created against 83 million displaced, a net fall of 14 million jobs, or about two per cent.
• Company training priorities put analytical thinking first at 48 per cent and creative thinking second at 43 per cent, with AI and big data third at 42 per cent and first among organisations of more than 50,000 people.
• The paper set out a capability stack of four layers, which were machine fluency, task craft, critical judgement, and governance and data discipline, and held that the first was the cheapest to teach and the least durable.

Why did the 23 per cent figure travel so badly?

Because churn counts jobs created and jobs eliminated in the same total, and the headline moved faster than its definition did. The Forum’s 23 per cent was the sum of growth of 10.2 per cent and decline of 12.3 per cent, which netted to a projected fall of about two per cent over the five years then ahead. Quoted as a forecast that a quarter of roles would disappear, it supported a conclusion the report never reached. The difference mattered because it changed the response. An organisation planning for a quarter of its roles to vanish would cut.

The second piece of evidence carried the same caution. Microsoft published a work trend report nine days after the Forum, based on a survey of 31,000 full-time employed or self-employed people across 31 markets, conducted by an independent research firm between 1 February and 14 March 2023. Both studies were large, both were employer-facing or vendor-sponsored, and the paper treated that sponsorship as a fact the reader needed in front of them while reading either number.

Which skills did employers say they wanted most?

Cognitive ones led the list. Analytical thinking headed company training priorities at 48 per cent, creative thinking followed at 43 per cent, and AI and big data came third at 42 per cent. Among organisations of more than 50,000 people the order changed, and AI skills came first. The paper set that ordering against what employers were actually buying, and found the money going to machine fluency, which was the one layer a supplier could package and sell.

That is the gap the paper was built around. Machine fluency could be taught quickly, priced clearly and refreshed whenever the tools moved, which made it the layer every training budget could reach. Task craft, critical judgement, and governance and data discipline had to be grown inside the organisation against real work, and the paper placed the scarcity there.

What did the paper propose?

It proposed four layers of capability, four workforce populations with different obligations, and a 90-day approach that started from tasks. Mapping tasks came before any rewriting of roles, and the reasoning was practical. A job description rewritten in 2023 to accommodate a technology six months old would be rewritten again within the year, and the exercise would consume political capital that the task work needed. Roles could wait.

Assessment followed the same logic. The paper held that course completion measured attendance, and that verified task quality, escalation behaviour and privacy compliance measured whether anything had changed in the work itself. Transfer was the thing to look for at the end of the 90 days.

How should the paper be read now?

As a record of a position taken in May 2023, and as a test of which parts of an argument survive their evidence. The survey figures belong to that date, and so do the training priorities, which reflected what employers expected of the five years then ahead of them. Conditions have moved a great deal since. What holds up for inspection is the structure, meaning the four layers, the distinction between churn and loss, and the order in which tasks and roles are addressed.

The claim at the centre of the paper was about supply. Employers in the spring of 2023 said they wanted analytical and creative thinking above technical AI skill, and the market they were spending into sold tool familiarity. The paper’s answer was to build the other three layers deliberately, starting from the tasks people actually did, and to judge the result by the quality of that work. That was the position as it stood in May 2023.

Frequently asked questions

What are the four layers of the enterprise AI capability stack?

The Enterprise AI Skills Map, written in May 2023, set out four layers of capability, which were machine fluency, task craft, critical judgement, and governance and data discipline. The paper held that machine fluency was the cheapest of the four to teach and the least durable, and that the remaining three had to be grown inside the organisation against real work. The paper mapped the four layers across four workforce populations with different obligations.

Why should tasks be mapped before job descriptions are rewritten?

Mapping tasks first avoids rewriting job descriptions that will need rewriting again. The paper, written in May 2023, argued that a job description rewritten that year to accommodate a technology six months old would be rewritten again within the year, and that the exercise would consume political capital the task work needed. Its 90-day approach therefore started from tasks, and left roles until later.

How do you know if AI training has actually changed how work gets done?

Measure transfer into the work itself. The Enterprise AI Skills Map held that course completion measures attendance, and that verified task quality, escalation behaviour and privacy compliance show whether anything has changed in how the work is done. Those three measures sat at the end of the paper’s 90-day approach, and they remain a practical test of whether training reached real tasks.

Why is machine fluency described as the least durable capability?

Machine fluency is the least durable layer because tool familiarity has to be refreshed whenever the tools move. The paper, written in May 2023, called it the cheapest of its four capability layers to teach and the least durable. It was also the layer a supplier could package and sell, which is why training budgets reached it first. The other three grow inside an organisation against real work.

How reliable were the 2023 surveys behind the AI workforce debate?

Both were large, and both carried a sponsor. The World Economic Forum’s Future of Jobs Report, published on 30 April 2023, surveyed 803 companies across 27 industry clusters and 45 economies employing more than 11.3 million workers. Microsoft’s work trend report, nine days later, drew on 31,000 employed or self-employed people across 31 markets. The paper read both as employer-facing or vendor-sponsored evidence and kept that in front of the reader.

Paul Forrest is a fractional Chief AI Officer and AI strategist working with UK boards, private equity portfolio companies and growing businesses. Read how the fractional CAIO engagement works, or see the full range of Ecaveo AI services.