Bytes sells AI licenses and funds a study saying licenses are not enough

On 29 October Bytes Technology Group published The AI Ambition Gap, a report it commissioned from The Productivity Institute at Alliance Manchester Business School. Its authors, Bart van Ark, professor of productivity studies at the school, and Dr Vassilis Karountzos, drew on a review of published research and eight expert interviews. Their finding is that AI makes single tasks much faster, and that the gain shrinks at every step from the task to the job, the team and the company.

Bytes is a FTSE 250 software reseller and services company. Its interim results on 13 October put gross invoiced income for the six months to 31 August at £1,605.2m, up 19.6%, and gross profit at £97.1m, up 17.8%. Microsoft accounts for around half of that gross profit. The same statement said the combination of advice, technical depth and managed services is what lets Bytes take a greater share of customers’ technology spend as AI, cybersecurity and cloud adoption accelerate.

So a company that lives on licenses has paid for research whose central conclusion is that a license, on its own, delivers little. Sam Mudd, Bytes’ chief executive, wrote the foreword. “AI takes extraordinary capital to build and almost nothing to try,” she said in the release, so it arrives whether or not leaders have a plan. Across Bytes’ customer base, she said, “companies in the same sector are making completely opposite bets”, from a license for every employee to nothing at all. “Neither guarantees results.”

The report is a careful, sourced account of why AI pilots stall, built from published studies rather than Bytes’ own customer data. It is also a prospectus for the work that follows a license sale: the management, training and process redesign that integrators charge for.

Where the gains leak

At task level the numbers are large. A study the report leans on found customer support staff with a generative AI assistant handled about 14% more queries an hour, with the biggest gains for less experienced staff. Professionals using ChatGPT for routine writing and analysis finished 40% faster, with quality rated 18% higher. Both measure single tasks, and the authors say experimental studies of this kind show what AI can achieve under favorable conditions rather than what organizations get in practice.

The widest gap the report shows is in software. Across more than 500,000 GitHub developers, adoption of successive generations of AI coding tools raised coding activity by about 240%. Completed projects rose by around 80%. Software releases rose 30%. Reviewing, integrating, testing and shipping the extra code still ran on people, and those people absorbed much of the gain.

Wide adoption, shallow use

The adoption figures tell the same story from the other end. McKinsey Global Institute research cited in the report found 89% of organizations use AI in at least one business function, but only 46% had moved beyond pilots, and one in four had redesigned workflows around AI rather than adding it to existing processes. ‘AI high performers’, firms attributing at least 5% of earnings before interest and tax to AI and reporting a significant overall impact, were 6% of the sample. A separate four-country survey of senior executives found 69% of firms use AI in some form, and more than two-thirds of those executives reported average use of around 1.5 hours a week.

The report’s clearest dividing line is management. It cites evidence that firms in the top quartile of management practice scores adopt technology, including AI, at substantially higher rates than those in the bottom quartile, and that firms with stronger management three years earlier were more likely to adopt AI later, after size, age, sector and location were taken into account. Its reading is that well-run firms adopt AI first because they already have the measurement systems and change-management experience, “rather than the effect of AI alone”.

Van Ark put the conclusion this way: “The ambition gap lies between those individual efficiencies and sustained business results. Closing it requires firms to invest in management, skills and the redesign of workflows alongside the technology.”

What the reseller gets out of it

That prescription doubles as a product line: the report’s four business enablers read like a statement of work. The four are a clear goal for what AI is expected to achieve, proper training, technical feasibility tested against economic viability and a readiness to change how work is done. It adds that trusted intermediaries – it names Catapults, growth hubs, chambers of commerce, colleges and universities – can reduce vendor risk by helping firms test whether systems work as promised, and that cost per successful outcome says more than the model or token price, because per-query charges can rise with use. Bytes, meanwhile, is buying CyberOne, a managed security provider, in a deal it expected to complete by the end of October.

One of the eight experts interviewed for the report was Andrew Holden, Bytes’ chief financial officer. The acknowledgements also record that Microsoft Copilot was used for language editing and drafting support, with the authors reviewing and fact-checking its suggestions.

None of that makes the findings wrong: the report is explicit that the 89% and 69% adoption figures measure whether AI is used at all, not how deeply, and it sets out the limits of the main studies it compares. Its J-curve point is an honest one: retraining, reorganization and data clean-up show up as cost first and output later, so measured productivity can dip before it rises.

The 6% figure will be quoted in sales decks for a year. The report’s own line belongs beside it: “licensing a tool does not mean that a firm can integrate it productively.” Bytes sells the license. It has now paid for the study on everything the license needs next.