The conversation at every mid-tier accounting firm we have worked with over the last two years starts the same way. "We know AI is going to change our profession; we are not sure what to do about it." The large firms have programmes; the mid-tier is often caught between imitation and paralysis. This article is the posture that, in our experience, actually works for firms between 100 and 5,000 staff.
Separate the three conversations
"AI at accountancy firms" is usually three distinct conversations mashed together:
- Client service automation. Using AI to deliver existing services faster (audit workpapers, tax preparation, advisory write-ups).
- Practice management. Using data to run the firm — utilisation, realisation, pipeline, margin.
- New service lines. Using AI to sell new things to clients — data strategy, AI readiness, automation.
Each has different economics and different risks. Conflating them is the first mistake.
Client service automation — the real picture
The vendor landscape is loud. The reality inside firms is more measured:
- Audit workpaper automation saves real time on well-scoped tests (sample selection, recalculations, cross-referencing) and near-zero on judgement-heavy areas (revenue recognition, provision analysis). Net audit time saving in our observations: 8–15%. Meaningful but not transformative.
- Tax research and memo drafting — useful, with citation and review discipline. Time saving 20–40% on routine memos.
- Document intelligence (contract reading, financial statement parsing) — a clear win, deployed carefully, in roughly half of our clients.
- Fully autonomous accounts production — not yet. Directionally yes, in three to five years; today, a junior-accountant replacement is overclaim.
The firms landing this well have an AI lead, an evaluation regime, a clear "use case" catalogue, and a culture that separates marketing from measurement.
Practice management — the more important conversation
Most mid-tier firms we work with have patchy data on their own operations. Utilisation, realisation, WIP ageing, pipeline, client profitability — there is usually a monthly pack, but rarely a timely, trustworthy, drillable view. The firms that fix this make better commercial decisions. The firms that do not are competing on instinct against firms that are not.
The pattern that works: the practice management system (usually Practice Engine, IRIS, CCH, or an adjacent platform) feeds a small data platform; the key metrics are under a semantic layer; partners have self-serve on client, engagement and team views. Nothing clever about this — it is overdue.
New service lines — the revenue opportunity
The mid-tier firms that are doing well in 2026 are the ones that have stood up a data and AI advisory capability, either organically or by acquisition. Clients are asking these questions of their accountant; if the firm cannot answer, the client goes to a consultancy. Several of our current engagements are partnerships with firms to co-deliver data and AI advice to their clients.
The regulatory posture
ICAEW, IFAC, PCAOB, FRC. Expectations on AI in audit are evolving rapidly. The posture that keeps regulators happy: document every AI tool used in the audit, evidence its evaluation, name a senior accountable for it, review annually. This is close to the evidence discipline any regulator will want across sectors; accounting firms are ahead on documentation, which helps.
Data protection — the common risk
Accounting firms handle significant client data. The common risk we see: partner-led AI experiments using ChatGPT personal accounts with client data. The cleanest fix is a firm-provisioned, tenancy-appropriate LLM capability with a clear acceptable-use policy and training. Partner education has to be the top of the stack; culture decides whether the policy holds.
A recommended twelve-month plan
- Stand up a firm-owned LLM capability with acceptable use, training and monitoring.
- Fix the practice-management data — one platform, one semantic layer, timely partner views.
- Pick three client-service use cases with measurable time-saving. Build evaluation. Roll gradually.
- Launch or partner on a data-and-AI advisory service for the firm's client base.
- Review quarterly. Decommission what is not working; double down on what is.
This is not exotic. It is the unglamorous sequence that, twelve months later, puts a firm in a materially stronger competitive position than its peers who chose one big bet.