Clients of management consultancy Accenture can be sure that the company to whom they pay handsomely for services that “reinvent how their organizations compete, grow and lead in an era of enterprise reinvention” have their best interests at heart when it comes to how to spend their technology budget.
Audio recordings of a high-level meeting at Accenture and leaked to campaigning website, 404 Media, reveal that Accenture is about to release a new product designed to help clients control what they spend on AI, money that flows to big US companies like Anthropic, Microsoft and OpenAI.
Accenture is, of course, all about AI, offering services that will let organisations develop “nation-building AI capabilities” and embrace “the practical value and strategic potential of AI”.
However, the US tech giants offering AI have begun changing their pricing models for access to the very latest in amusingly unpredictable word-guessing software. Until a few months ago, companies handed over anywhere from $200 and up, per month, per user, to use a large language model – something that’s now known in modern parlance as AI.
But under the old subscription model, each user was able to burn through, in the case of Anthropic for example, 8-13x more computing power than the subscription price covered. This, we assume, is one reason for the eye-watering losses that all AI companies suffer.
But that’s changing. Starting with the highest paying enterprise subscriptions, and gradually percolating down through the cheaper price tiers, users are now being charged on a ‘per-token basis’ to achieve vital tasks, like summarising emails and turning PDFs into PowerPoint presentations. With a token typically representing about 0.7 of a word, every question or document pasted into, or printed out by an AI now has a literal, charged cost. That means companies will have to pay up, even when the output of the AI is so bad it’s unusable. And as anyone who’s tried ChatGPT et al. will know, you normally have to ask the same question three or four times in different ways, just to get something sensible out of it.
The spreading change in charging policies, emanating from Silicon Valley, has come as something of a shock to many of Accenture’s clients, including taxi company Uber. It’s on record as saying it blew its entire 2026 AI budget in just the first four months of this year, and the company’s chief operating officer, Andrew McDonald, went so far as to openly blaspheme against the church of AI: “If you’re not actually able to draw a direct line to how many useful features and functionality you’re shipping, [AI] becomes harder to justify,” he told the Rapid Response podcast.
None of this should come as any surprise to a company that positions itself as one of the world’s leading technology consultancies. The industry has known for at least a couple of years that the subscription model of paying for AI was unsustainable, and that any organisation committing to the technology would sooner or later have to pay the real cost of the massive computing power needed to run the large language models at the heart of AI services.
Accenture clients will be safe from spiralling AI bills, however. According to the leaked recordings, the company plans to release “Token IQ”, a piece of software that monitors and limits the use of AI in an organisation in order to keep those pesky per-token charges down.
Accenture’s buzzword-laden messages on its website explain that its clients will be able to “reinvent platforms and align AI”, as after all, “gen AI is transformative”. So, it will be comforting to know that Accenture’s new (presumably paid-for) Token IQ service will help combat the financial effects of organisations acting on advice from, er, Accenture.