From BYOAI to Business Value: channelling bottom-up adoption into enterprise impact

Bottom-up adoption of artificial intelligence becomes enterprise value where the organisation channels it, and the approved route has to be open before the unapproved one is closed.

Nothing in this article constitutes legal advice. It is offered to support discussion and further review with qualified advisers.

Why were so many employees using their own artificial intelligence tools at work?

Because demand for the tools had outrun official supply for roughly 18 months, and people working under delivery pressure solved their own problem in the way people always do. Paper Seven in this series, From BYOAI to Business Value, closed on 31 May 2024 and reports the position as it stood at that date. Its starting point was a study published on 8 May 2024 by Microsoft and LinkedIn, drawing on a survey of 31,000 full-time employed or self-employed knowledge workers across 31 markets, conducted by an independent research firm between 15 February and 28 March 2024, alongside labour market data, product telemetry and research with large corporate customers. Three quarters of those surveyed reported using artificial intelligence at work. On the paper’s reading, the organisation held the risk without the visibility and without the benefit, and the task it set itself was to convert that into something governed.

• Of the artificial intelligence users surveyed, 78 per cent brought their own tools to work, rising to 80 per cent at small and medium-sized companies.
• The 78 per cent figure took artificial intelligence users as its denominator, and quoting it against knowledge workers generally overstated the position considerably.
• Use of generative tools at work had nearly doubled over the preceding six months, and 46 per cent of users had begun within that same window.
• The paper argued for opening the approved route before closing the unapproved one, since blocking first bought an organisation months of invisible use and a workforce that had learned to keep quiet.
• Licence activation measured procurement while weekly repeat use on a named task class measured adoption, and the paper held that the two differ by more than most sponsors expect.

How should the 78 per cent bring-your-own figure be quoted?

With its denominator attached. The study reported that 78 per cent of artificial intelligence users brought their own tools to work, rising to 80 per cent at small and medium-sized companies, and the group missing from that fraction was everyone who used nothing at all. Quoted against knowledge workers generally, the same number becomes a considerably larger claim than the study made. Two qualifications belonged permanently beside the headline. The sponsor sold products in this market, and the sampled population was knowledge workers, which left out employees outside that category.

The doubling reported in the study was specific to generative tools. Use of those tools at work had nearly doubled over the preceding six months, and 46 per cent of users had begun within that same window. Taken together, the figures described a change in behaviour that was both fast and recent, which is worth highlighting for anyone tempted to read the numbers as a stable picture.

How does an organisation turn unofficial use into something governed?

By opening an approved route and making it the easier one to take. The paper proposed a discovery method that draws on more than self-report, on the reasoning that a workforce asked to declare unapproved tools has an obvious incentive to stay quiet. It then set out a channelling strategy with a sequencing rule at its centre, which was to provide the sanctioned route faster than the unsanctioned one is closed. An organisation that blocks first and provides later was likely, in the paper’s view, to buy itself six months of invisible use and a workforce that had learned to keep quiet about what it was using.

Access alone was treated as the weaker intervention. The paper’s fourth section argued for redesigning the workflow around the tool, so that the gain shows up in how the work is actually done. A value tree then connected each redesigned workflow to a business outcome the sponsor already recognised, which is how a pilot earns the right to a second year.

How should adoption be measured?

As repeat use on real work. Licence activation measures procurement, weekly repeat use on a named task class measures adoption, and the paper’s position was that the two differ by more than most sponsors expect. The dashboard it proposed was built to survive a change of sponsor, which in practice means reporting a small number of measures that stay meaningful once the original enthusiasm has moved on.

What had changed in Europe by the end of May 2024?

The Council adopted the artificial intelligence regulation on 21 May 2024, completing the legislative procedure. Signature and publication had still to follow when the paper closed ten days later, so on the paper’s account the regulation had yet to become law at that date. Anyone reading this now should treat that section as a snapshot, since the European position has moved on considerably in the time since.

The seventh section deals with reciprocity, and it is the part that has aged best. An organisation asking its people to move off tools they chose and onto sanctioned ones is asking them to give up something that was working for them, and it owes them something in return, whether that is a better tool, time saved that they keep, or credit for the workflow they worked out themselves. Read now, the paper is a record of one moment in enterprise adoption, when three quarters of a surveyed population had already started while official provision was still catching up. The figures have moved. The argument about sequencing, measurement and what the organisation owes back is the part worth carrying forward.

Frequently asked questions

What does BYOAI mean?

BYOAI stands for bring your own artificial intelligence, meaning employees using tools of their own choosing to do their work. A Microsoft and LinkedIn study published on 8 May 2024 found that 78 per cent of the artificial intelligence users it surveyed brought their own tools to work, rising to 80 per cent at small and medium-sized companies. The 2024 paper read that behaviour as a response to demand outrunning official supply.

How can an organisation find out which AI tools its staff are already using?

Through a discovery method that draws on more than self-report. The 2024 paper made discovery its first step, on the reasoning that a workforce asked to declare unapproved tools has an obvious incentive to stay quiet. Discovery therefore sets what people report alongside other evidence the organisation already holds, and it comes before any decision about which tools to sanction and which to close.

What is a value tree in an AI adoption programme?

A value tree connects each redesigned workflow to a business outcome the sponsor already recognises. In the 2024 paper it followed the workflow redesign step, on the argument that the gain from an artificial intelligence tool shows up in how the work is actually done. The paper’s position was that this link is how a pilot earns the right to a second year.

What does an organisation owe staff it moves onto approved AI tools?

Something in return for what they give up. The 2024 paper called this reciprocity and named three forms it can take, whether a better tool, time saved that the employee keeps, or credit for the workflow they worked out themselves. Someone moving off a tool that was working for them is being asked to surrender something, and the paper held that the organisation should answer that.

Was the Microsoft and LinkedIn survey on AI at work reliable?

It was a large study with real findings, and two qualifications belong permanently beside its numbers. The sponsor sold products in the market it surveyed, and the sampled population was knowledge workers, which left out employees outside that category. The survey covered 31,000 full-time employed or self-employed knowledge workers across 31 markets and was conducted by an independent research firm between 15 February and 28 March 2024.

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.