When KPMG pressed its own auditor for a 14% fee cut earlier this year, arguing that AI was reducing the cost to deliver, it handed every client of every firm the same script.

That lands differently on IT than on the partners. According to the Accountancy Age Top 50+50 report, median technology spend across the UK's 100 largest firms now sits just under £300,000, with the top quartile above £900,000. The same report names margin pressure and tighter pricing discipline as a defining force for 2026. The squeeze arrives from both directions at once: clients expect AI savings passed through, and the board expects the technology budget to show a return. Reconciling those two demands usually falls to whoever signed off the stack.

The question your stack can't answer

‍A 2025 Xero and Cebr study, cited in the Top 50+50 report, found 98% of UK accounting practices now use AI in some form, cutting time on routine tasks by around a third. Adoption is no longer the question. The uncomfortable one that follows: if AI has taken a third of the time out of routine tax and audit work, where did those hours go?

Most firms can't answer, and not because the AI underdelivered. The hours are real. But when a piece of work takes less time than planned, something, or someone, has to route the next piece of work and record that it happened. Otherwise, the capacity evaporates: the firm gave the fee discount and the freed time never became billable work. The same applies looking forward. If returns will take 30% less time next season, staffing plans, pricing and capacity forecasts should already reflect that. Both answers depend on the same underlying asset: task-level data on who is doing what work, for how long, at what cost.

Why it's an architecture problem
‍
Strip it back to the event that actually needs handling: Using AI means a piece of work finishes early, and someone has a few hours they weren't scheduled to have. In most stacks, nothing handles that moment, because the capability is split across multiple systems. Some practice management platforms do schedule at the task level, but the scheduling is manual: nothing detects the freed hours or proposes the next piece of work, so reallocation depends on a manager noticing or the staff member self-serving.

Resource management tools that do offer automated, proactive matching typically work at the engagement level, a unit far too coarse to catch hours recovered task by task. Either way, the answer to "where did the hours go?" is the same: nowhere anyone can see. And in most firms the underlying data is scattered across practice management, workflow tools, timesheets and departmental spreadsheets.

Firms keep buying AI to do the work faster. Meanwhile, the layer that would prove the work got faster, and redeploy the time it saved, doesn't exist in the stack at all.

What's missing is the combination: automated, proactive matching operating at the task level, integrated with practice management, workflow and time data. When a task completes early, the next one routes to the best available person, and the variance is recorded so the firm can prove where the hours went. One version of the truth, entered once, flowing through documented APIs rather than manual exports and spreadsheets.

This combination is what Beeye was built for, integrating with the systems firms already run, from CCH and IRIS to Outlook and Power BI. As Ben Brownson, Group Board Director & Head of Operations at RPG Chartered Accountants, put it: "We were inefficient in using two systems to allocate work and build reports. Beeye has given us a shared picture of upcoming work, statuses, capacity, scheduled hours, and budgets in one place."

Whatever tooling you choose, the requirement stands. When the fee pressure arrives, the defensible firms will be those whose IT leaders have supported the business with tooling to show which work got faster, where the recovered hours went, and the upside.