The Accounting Firm Business Model Is Being Repriced — Here Is the Evidence

Four converging lines of evidence — the adoption/value gap, what the Big Four actually shipped, the capacity math of AI versus offshore versus hiring, and the emergence of AI assurance as a service line — say the accounting firm business model is being repriced. The evidence, assembled.

Summary

The profession has spent three years arguing about whether AI will change its work. That was the wrong argument. The work is changing at the margins; the pricing of the work is changing structurally.

Five forces are converging. Adoption is near-universal but value delivery is not. The Big Four have industrialized agentic audit. Private equity has consolidated a thousand-plus firms and is now funding AI-native challengers. The talent pipeline that fed the leverage model is contracting. And the first audit client has successfully demanded that its auditor's AI savings be passed through in the fee.

Together they reprice compliance work and force every practice to answer a question it has been able to defer: what, exactly, are clients paying for? This paper sets out the evidence, including the parts that cut against the narrative, and closes with what a mid-sized practice should do in the next twelve months.

---

1. The adoption/value gap is the real story

Thomson Reuters surveyed 1,816 professionals across 62 countries in March–April 2026. Seventy-four percent use AI weekly. Ninety-one percent say their organization falls short of AI's potential. (Thomson Reuters, Future of Professionals 2026)

Adoption is no longer the constraint. Translation into delivered value is.

The same research puts a number on the commercial consequence. Seventy-eight percent of corporate clients regard AI-enabled quality improvements as essential, but only 6% believe most providers deliver AI value. Thirty-two percent are reconsidering provider relationships within twelve months, and a third of those represent $1M+ in annual work — approximately $143 billion of combined US legal and accounting revenue under active reconsideration.

Clients have priced AI into their expectations. Most firms have not priced it into their offer.

---

2. What the Big Four actually shipped

Agentic AI in audit moved from demonstration to production in 2026. The scale is not in dispute; the outcomes are.

| Firm | What shipped | Scale |
|---|---|---|
| EY | Multi-agent framework embedded in EY Canvas (7 April 2026), built on Microsoft Foundry, Fabric and Azure | 160,000 audit engagements, 130,000 assurance professionals, 150+ countries; Canvas processes 1.4 trillion journal entry lines annually; full end-to-end support targeted 2028 (EY) |
| Deloitte | "Connected Agentic Intelligence" in Omnia (24 June 2026), including a Trustworthy AI framework and in-platform Tutor agents | ~85,000 Audit & Assurance professionals globally (Deloitte) |
| PwC | Simplified Audit for Private Business, Advanced Walkthrough Assistant, and Evidence Match for agent-led evidence extraction on cash, AR and AP | Targets end-to-end AI across the audit cycle in calendar 2026, global agentic platform ~18 months later (Accounting Today) |
| KPMG | Minority equity stake in Fieldguide, co-building agents that perform audit procedures (23 October 2025) | No quantified outcomes disclosed (KPMG) |

Three caveats belong alongside those numbers.

No firm has published independently audited productivity or audit-quality data. Every figure above is firm-supplied.

PwC's timeline is a hedge, not a delivery. US Assurance Transformation Leader Shawn Panson on the 2026 target: "For calendar '26, that is our hope. We have work to do…" Colleague Kyle Maryanski conceded the engineering difficulty — "We had to ingest 20 different kinds of documents…"

At smaller firms, technology is currently degrading audit quality, not improving it. ICAEW's 2026 Audit Monitoring Report names rapid adoption of software-based audit systems as "the most significant technology risk resulting in poor audit quality since the pandemic" — staff working outside the system, tick-boxes substituting for documentation. Across 375 monitoring visits in 2025, 73% of audits were good or generally acceptable and 6% needed significant improvement (down from 10% in 2024). The report does not mention AI at all: the damage is coming from ordinary audit software rollouts. (ICAEW, PDF)

The lesson for a mid-sized practice is not "adopt what EY adopted." It is that tooling without methodology discipline makes quality worse, and the regulator has already documented it.

---

3. Capital has arrived, and it is consolidating

Private equity has been the largest structural change to firm ownership in a generation, and AI is now its investment thesis rather than a footnote.

2015–2025: 177 direct PE investments triggered 875 roll-up acquisitions, affecting more than a thousand firms.
2025 alone: fewer than 200 direct investments spawned roughly 900 follow-on deals. The consolidation multiplier has risen about 4x since 2021.
Roughly half the US Top 30 carried PE money or an alternative practice structure by early 2026.
Thrive Holdings / Current committed $1B to acquiring local practices and has raised approximately $2B from SoftBank, Altimeter and D1. It has acquired close to 50 practices, including Larson Gross. OpenAI took an ownership stake and provided an AI training team, with the stake growing as the accounting portfolio expands.

Source: Footnote.news, 14 July 2026

That last item is the most structurally significant fact in the profession right now. It vertically integrates model provider, proprietary firm workflow data and practice ownership in a single chain — and it is why the IESBA opened a workstream on private equity investment in accounting firms in March 2026 (IESBA).

Note the evidence quality: Thrive/Current's reported outcomes — 31% time saving on returns and accuracy "far above the typical manual error rate" after processing 7,000 tax returns in the 2025 season — are vendor-supplied and explicitly flagged as independently unverified by the publication reporting them.

On the software side, Basis raised $100M at a $1.15B post-money valuation in February 2026, led by Accel and GV with Lloyd Blankfein participating, up from $34M in December 2024. Founded 2023. Named firm customers include Boulay PLLP, Clark Nuber PS, MarksNelson LLC, Pinion LLC and UHY LLC. Its architecture matters more than its valuation: LLM agents wrapped in rules-based controls, the design pattern that makes agent output auditable. Its claim that ~30% of the Top 25 firms are deploying its agents is vendor-stated and unconfirmed. (SiliconANGLE)

---

4. The talent pipeline: the constraint nobody has solved

| Metric | Figure |
|---|---|
| Accounting bachelor's + master's degrees awarded, 2023–24 | 55,152, down 6.6% YoY |
| Bachelor's / master's decline | −3.3% / ~−15% |
| New CPA exam candidates | 42,626 (2023) → 28,082 (2024) → 16,448 (H1 2025) |
| Spring 2025 enrolment | 266,506, up 12.4% — highest since 2020 |
| Firms expecting to hire at least as many new grads in 2025 | 75% |
| New graduate hires, 2024 | 11,985 |

Source: Journal of Accountancy, October 2025, reporting AICPA data.

Be balanced about this. Enrolment rebounded 12.4% and AICPA's Jan Taylor notes that "the rate of decline has slowed year over year," attributing the improvement to higher entry-level pay and reduced licensure hurdles. The pipeline is contracting, not collapsing.

But AI is now both cause and cure, and firms are being judged on it. Sixty-five percent of job candidates evaluate a firm's tech stack before accepting an offer (Accounting Today). Sixty-two percent of professionals say access to professional-grade AI influences whether they accept a job, and 24% are considering leaving within two years because their firm is not delivering on AI (Thomson Reuters).

The unresolved paradox: if AI automates a meaningful share of entry-level work, the profession loses the training ground that produces senior reviewers — precisely the people required to supervise AI. No firm and no standard-setter has published a credible answer. The frequently quoted claim that AI firms can automate "approximately one-third of new hires" is a survey respondent's assertion with no disclosed methodology; treat it as directional, not measured.

---

5. Offshore versus AI: the consensus flipped

In 2025 the working assumption was that AI would kill offshoring. By 2026 the practitioner view had inverted: AI removes the friction that limited offshoring — time zones, language, email culture, accent — making it cheaper and more effective rather than redundant. A new role, the "AI work reviewer," is emerging offshore. The cited economics: roughly $500 per month offshore against ~$1,500 for a comparable US accountant, with projected 50% savings from offshoring and 25% from outsourcing after a one-to-two year ramp. This appears in an opinion column and the savings figures are the author's projections, not measured outcomes (Accounting Today opinion).

The planning implication stands regardless of the exact numbers: AI, offshore and hiring are now substitutable inputs to the same capacity decision, and most firms are modeling them in separate spreadsheets owned by separate partners.

---

6. Fee compression has its first documented case

KPMG demanded that its own auditor, Grant Thornton, pass along AI-driven savings — and threatened to switch firms. Grant Thornton's audit fee for KPMG fell from $416,000 (2024) to $357,000 (2025), a 14% cut worth $59,000. (TheStreet)

This is the first documented case of an audit client explicitly pricing the auditor's AI productivity into the fee. It will not be the last, and it will not stay confined to audit. Every procurement team that reads it now has a template.

Meanwhile firm spending is shifting: 64% of firms will invest in AI in 2026, up from 57% in 2024; average technology investment moved from $19,000 in 2025 to a planned $20,000 in 2026; and technology has held steady at ~21% of overall firm budgets for three years (Accounting Today). Spending is being reallocated within a fixed envelope, not expanded. Firms are buying AI by not buying something else.

---

7. CAS is the demonstrated exit — and it is crowding

The AICPA / CPA.com benchmark survey (n=206 CAS practices) remains the most-cited data in the profession on advisory economics — though it reflects CY2023 data collected in 2024, i.e. pre-agentic, and a refreshed benchmark is overdue. Read the figures below as the pre-AI baseline this section argues is now being repriced, not as current economics.

| Metric | Figure |
|---|---|
| Median CAS growth rate | 17% (three-year projected median growth 99%) |
| Net client fees per professional | $156,250 median, up 29% over 2022 |
| CFO-level advisory premium | 30%+ higher monthly recurring revenue |
| Industry-niche practices | 38% higher median CAS revenue; 51% higher net revenue per client |
| Firms with a formal CAS business plan | ~$10,000 more median annual revenue per client |
| Practices still billing primarily by the hour | only 10% |
| Practices investing continuously in technology | 51% — and they serve 50% more clients (100 vs 67) |
| Practices with dedicated CAS-only staff | 78% |

Source: CPA.com · Full survey PDF

Two honest qualifications. First, this is CY2023 data collected in 2024 — entirely pre-agentic. A refreshed benchmark is the single most useful missing data point for 2026 planning, and every current CAS pricing model is an extrapolation from it.

Second, the lifeboat is full. Ninety-three percent of firms already offer advisory services and 79% of accountants anticipate increased competition for high-value advisory clients. "Advisory" is not a strategy. The 38% / 51% niche premium is the only durable differentiator visible in the data.

---

8. The new service line: assurance over AI

The most interesting revenue opportunity is not defensive. It is assurance over AI systems.

KPMG launched AI Trust Services on 25 September 2025: AI model risk assessments and control testing, model validation, real-time systems assessments, and AI assurance and attestation — covering model accuracy and assumptions, agent behavior within guardrails, and AI-related system changes affecting financial reporting controls, mapped to SOC, FedRAMP, SWIFT, HITRUST, FDA and HIPAA (KPMG). In December 2025 KPMG International became the first professional services network to attain ISO/IEC 42001 certification for AI management systems (KPMG).

There are two problems with the product.

There is no generally accepted attestation standard for AI systems. Firms are mapping to SOC 2 and ISO 42001 by analogy. Until the AICPA's assurance services executive committee or the IAASB issues criteria, AI assurance reports are not comparable across providers.

The credibility exposure is live. The sharpest critique of 2026 is that "the Big Four are selling AI governance while their own reports hallucinate" (Startup Fortune). Deloitte Australia delivered a A$440,000 independent assurance review to the Department of Employment and Workplace Relations containing a fabricated quotation from a federal court judgment and references to non-existent academic papers. Errors were identified in August 2025 by Sydney University researcher Chris Rudge; a revised document was issued and Deloitte repaid the final contract instalment in October 2025 (CFO Dive · OECD.AI).

That incident is worth more to a mid-sized practice than any vendor case study: the control that failed was review, not technology.

---

9. What regulators have and have not done

Practices planning around a coming rulebook should stop.

| Body | Position |
|---|---|
| IAASB | Decided December 2025 to pursue non-authoritative guidance only — practice notes and staff publications, not new standards. Anchored on ISQM 1 and ISA 220; action plan discussed March 2026. Evidence base: 240+ stakeholders across 8 roundtables on six continents (Accounting Today · IAASB) |
| FRC (UK) | First national AI-in-audit guidance, 26 June 2025. Requires proportionate documentation of AI tools, context-appropriate explainability, and controls over automated tools — but introduced no new regulatory mandates (FRC) |
| IESBA | 2023 technology revisions effective December 2024: accountants must understand, explain and evaluate the technology they use, with automation bias named explicitly as a threat to objectivity. Technology non-authoritative material planned Q3 2026; AI-specific material Q4 2026 (IESBA) |
| PCAOB | Its most recent GenAI position is a Spotlight dated 22 July 2024, finding GenAI use concentrated in administrative and research tasks. No comparable update since (PCAOB) |

Enforcement capacity is falling as AI use rises. Combined SEC and PCAOB audit enforcement actions fell to 39 in 2025 from 58 in 2024 (−33%), with penalties of $17.9M against $52.2M (−66%) (The Brattle Group via Thomson Reuters). The PCAOB's 2026 budget is $362.1M, down 9.4% (SEC).

Aggressive AI deployment plus shrinking oversight is the genuine systemic risk of this period, and it means the profession's own quality control is for now the only control. Note also that the PCAOB lowered its liability standard from recklessness to negligence in 2024 (Forbes / Yvonne Hinson, AAA CEO) — materially higher exposure for AI-assisted work, arriving precisely as enforcement resources fall.

---

10. What a mid-sized practice should do in the next twelve months

Six moves, in order.

One: fix the review standard before buying anything. ICAEW's finding and the Deloitte Australia incident point to the same failure — review, not technology. Write down what constitutes adequate human review of AI-assisted work, what evidence of that review must exist, and who signs. This costs nothing and is the control that would have prevented every documented failure in this paper.

Two: inventory your tools, including the ones nobody approved. Thirty-three percent of professionals use unapproved AI tools, rising to 41% where their organization moves slowly, and 41% lack access to professional-grade tools meeting security standards (Thomson Reuters). Shadow AI is an audit-trail problem before it is a security problem, and IESBA's technology expert group is now monitoring it explicitly.

Three: reprice compliance work now, on your terms. The KPMG/Grant Thornton case will reach your clients. Decide what your compliance fee buys and where the value moved before a client tells you.

Four: pick a niche and commit. The benchmark data is unambiguous — industry-niche CAS practices earn 38% higher median CAS revenue and 51% higher net revenue per client. With 93% of firms offering advisory, niche is the only differentiation the data supports.

Five: model capacity as one decision. Build a single twelve-month capacity plan in which AI, offshore and hiring compete against one another with explicit assumptions. Most firms are running three separate plans and calling the sum a strategy.

Six: assess the assurance-over-AI opportunity honestly. If you have a technology-risk capability, ISO/IEC 42001 is the anchor and this is the highest-margin new service line available. If you do not, the realistic version is an AI governance readiness assessment for your existing clients — tool inventory, use policy, human-oversight evidence, remediation roadmap — which is a genuine advisory product you can sell in 2026 rather than a standard you are waiting for.

The firms that come through this repricing will not be the ones that adopted earliest. They will be the ones that can evidence what their people did, what the machine did, and who checked.

---