This article is a companion to my review of the book Accrual Intentions by Alexis Kingsbury which you can find here. I list 10 things for your leadership team to discuss.
Purpose To help firm leaders consider how artificial intelligence may reshape their firm’s business model, service delivery, people, client relationships and risk management—and agree on a practical response.
The stimulus In Accrual Intentions, Alexis Kingsbury created an experimental accounting firm with eleven AI “team members” and one human. The project revealed striking capabilities, but also the limits of AI in quality, compliance and judgment. It was a creative experiment, not a real firm This document draws strategic questions from that experiment.
Central proposition The opportunity is not merely to do existing work faster. It is to redesign how work is performed, how quality is assured, how team members develop, and what clients receive. Human judgment and accountability remain essential.
01 What is our ambition for an AI-enabled firm?
AI should be a strategic decision, not an accumulation of unrelated software subscriptions. Kingsbury’s experiment raises the possibility of redesigning the firm around a smaller amount of routine production work and a greater contribution from people in judgment, relationships and advice.
DISCUSS In three years, do we want primarily faster preparation, greater capacity without matching headcount growth, a richer advisory offering, or a different client experience? Or some combination of these?
AGREE Agree a one-paragraph AI ambition, a partner sponsor and two or three measurable strategic outcomes.
02 Which work should humans do—and which should AI do?
The most useful distinction is not between jobs that disappear and jobs that survive. It is between activities suitable for automation, activities requiring professional review, and activities where human judgment and client relationships should lead. A tax compliance engagement, for example, involves data gathering, drafting, reconciliations, interpretation, explanation and final approval; each has a different a different skill and risk profile.
DISCUSS For our key service lines, which tasks are repetitive and rules-based, which involve difficult technical judgment, and which create distinctive client value.
AGREE Map two high-volume services into “automate / AI-assist with review / human-led” activities.
03 Are our processes sufficiently standardised to scale AI?
AI can amplify inconsistencies as readily as efficiencies. Before automating an engagement, a firm needs consistent inputs, documented procedures, appropriate access rights, clear quality criteria, escalation points and ownership. “One partner does it this way” is a weak foundation for an automated process. Be like the water taxis of Spetses! LINK TO THAT ARTICLE
DISCUSS Where do inconsistent workpapers, client onboarding steps, templates or hand-offs currently generate rework? How much variation is genuinely required by client circumstances?
AGREE Select one process to standardise and document before automating it end-to-end.
04 How will we verify accuracy and retain professional judgment?
The experiment demonstrated that plausible AI output may still be wrong. In tax and accounting, correctness is not established by fluent wording. Reviews should test source data, calculations, legislative references, assumptions, client-specific facts and the reasonableness of the final result. Higher-risk work warrants stronger independent verification and documented approval.
DISCUSS Which AI outputs can be checked through deterministic controls, and which require technical review? What must never go to a client without an accountable professional signing off?
AGREE Adopt risk-tiered review rules, including evidence of checks and mandatory human approval for client advice and lodgements.
05 Are client confidentiality, privacy and data controls fit for purpose?
AI tools may transmit data to third parties or store it in unfamiliar locations. In July 2026 the TPB issued AI guidance explaining that permission is needed before a practitioner divulges client information to a third party, which can include an AI provider, depending on its configuration. Tool selection must therefore consider permissions, confidentiality, retention, training uses, access controls and contractual protections. The OAIC also recommends careful due diligence and caution with personal information in public AI tools.
DISCUSS Do we know which AI services team members use and what client information they enter? Do engagement terms and client permissions reflect actual data flows?
AGREE Approve an AI-tool register, vendor assessment process, data-handling rules and a review of client permissions.
06 Who owns AI governance and what is our risk appetite?
Responsibility must not be diffused between partners, IT providers and individual enthusiasts. AI risks belong within the firm’s existing risk and quality arrangements, with ownership for approval, access, testing, incident response, monitoring and retirement of tools. APES 325 requires firms in scope to establish, maintain, monitor and document a risk management framework. AI should be explicitly addressed within it.
DISCUSS Who can approve a new tool or workflow? Who resolves a disputed AI answer? What is our escalation process when an output is wrong or a confidentiality breach is suspected?
AGREE Nominate an accountable leader, appoint workflow owners and add AI to the firm’s risk register and review cycle.
07 How should we redesign roles, supervision and career development?
As AI handles more first-draft and routine work, accountants may increasingly supervise systems rather than only junior team members. This shifts the premium towards instruction, technical reasoning, scepticism, exception handling and communication. It also creates a developmental risk – if graduates skip foundational tasks without learning the underlying principles, future reviewers may lack the knowledge to detect mistakes.
DISCUSS What work must graduates still perform to learn? Which new skills will managers need to lead AI-assisted teams? How can experienced team members transfer judgment rather than merely approve output?
AGREE Redesign one role’s development plan to cover technical foundations, AI literacy, review skills and client conversations.
08 How will clients experience the benefits?
Most clients do not purchase a set number of processing hours. They purchase outcomes and they value accuracy, responsiveness, reassurance and useful insight. AI may make more timely reporting, faster turnaround, proactive exception alerts and accessible explanations commercially viable. But a faster production process is not automatically a better client experience, and some clients may prefer human interaction for sensitive matters.
DISCUSS If compliance work takes less effort, what will improve for clients? Which clients would value more regular management insight, cash-flow conversations or decision support?
AGREE Design one measurable client-service improvement and test it with a small client cohort.
09 What happens to pricing, capacity and profitability?
AI could reduce the cost of some tasks while increasing the volume of work that can be produced. However, review time, software costs, exception rates, rework and client demand may offset headline savings. Billing approaches should reflect scope, risk and value rather than assume that faster work automatically warrants lower fees. Firms also need to decide whether released capacity is invested in growth, better service or reduced workload.
DISCUSS Which services will face fee pressure? Where might fixed-fee packaging or new advisory services be appropriate? Are we measuring actual net capacity rather than optimistic time savings?
AGREE Build a before-and-after business case tracking margin, turnaround, review time, rework and client outcomes.
10 How will we experiment responsibly and manage change?
The strongest approach is to start with useful, bounded workflows, learn quickly and expand when evidence supports it. Team members need clarity about what is approved, what is changing in their work and how their skills will develop. Vendor capabilities also change rapidly, so the firm should preserve control of its procedures, knowledge, records and exit options.
DISCUSS Which two or three pilots can produce measurable benefits with acceptable risk over the next 90 days? What evidence would justify expansion—or stopping a pilot?
AGREE Approve a 90-day pilot portfolio, with owners, success criteria, review gates and a communication plan.
This article was prepared with help from Chat GPT and was prompted by Rob’s reading of Accrual Intentions by Alexis Kingsbury. It’s a must read book for accounting firm leaders.