Why aren't accounting firms making progress with AI despite trying?

The pattern is consistent across surveys: most accounting firms that have 'tried AI' are stuck in Stage 1 — one or two people experimenting, no firm-wide workflow integration. The CPA Practice Advisor March 2026 piece identifies the most common blockers: tools introduced without explaining why, no dedicated time for staff to learn, and no human oversight protocol that makes AI output trustworthy. The result is that AI feels risky and inconsistent, so it doesn't get used consistently, so it doesn't produce the results that would justify using it consistently. The way out is deliberate and sequential — not more tools.

What does 'starting small with AI' actually mean for an accounting firm?

Starting small means identifying one specific pain point — not 'we want to use AI' but 'we spend 40 hours per month manually categorizing client transactions and we want that to be 10 hours.' Then deploying one tool against that specific problem, measuring the time before and after, and using that result to justify the next step. The specificity is what matters. Broad AI initiatives fail at small firms because there's no measurement that proves they're working. One workflow, one tool, one time metric is enough to build from.

How should accounting firms communicate AI adoption to their clients?

Transparent and proactive is the right posture. Tell clients you're using AI to improve the accuracy and timeliness of their work, and explain what the human oversight looks like (e.g., 'our team reviews all AI-assisted bookkeeping before it's final'). Clients increasingly expect their accountants to use AI efficiently — what they want to know is that a qualified professional is still responsible for the output. The firms that are struggling with client communication are those that either haven't told clients anything (creating surprise when clients discover AI is in use) or have oversold AI capabilities in ways that create unrealistic expectations.

What's the difference between AI augmenting accounting work versus replacing it?

The practical distinction: AI replaces tasks, not roles. Transaction categorization, report compilation, and document request follow-up are tasks that AI handles better and faster — the time saved goes back to the accountant for higher-value work, not to a hiring freeze. The roles that are safe in the near term are those that require professional judgment, client relationships, and regulatory accountability — exactly what licensed CPAs do. The roles under pressure are those where the majority of time is spent on repeatable, rules-based tasks without significant judgment involved. Most accounting staff roles involve both; AI shifts the ratio.

How do you measure whether AI is actually working at an accounting firm?

Start with time per task, not general impressions. Before deploying AI on a specific workflow, baseline the time: how many hours per month does your team spend on client transaction categorization? On monthly report compilation? On new client onboarding documents? After AI deployment, measure the same metric. If you can't measure it, you can't manage it — and you can't make the case to staff that AI is helping rather than creating more work. The secondary metrics to track over time: realization rate on advisory engagements, capacity per employee (number of clients served), and write-off rate on revision-heavy deliverables.

AI Adoption for Accounting Firms: 4 Steps

Published: March 15, 2026 | By: The Crossing Report | 6 min read

Summary

The AICPA/CIMA found that only 19% of accounting professionals use AI daily — 17% have never used it at work at all. CPA Practice Advisor (March 10, 2026) published a practitioner framework from Andrea Wynter, VP at ADP Canada, identifying four specific strategies for accounting firm leaders who have tried AI and aren't getting traction. Here's what each strategy means in practice and where to start.

The Real Problem Isn't the Tools

The AI tools available to small accounting firms in 2026 are genuinely useful. Transaction categorization tools with 90%+ accuracy exist and are accessible. Automated reporting tools that generate client-ready monthly statements exist and are affordable. AI-assisted client communication exists and is built into tools accountants already pay for.

So why are only 19% of accounting professionals using AI daily?

The AICPA/CIMA global survey from February 2026 quantifies the gap: 88% of accounting and finance professionals call AI the most transformative technology of their careers. Only 8% feel their organization is well-prepared for it. That's not a technology availability problem.

CPA Practice Advisor's March 10 follow-up analysis — featuring Andrea Wynter, VP at ADP Canada — identifies what's actually blocking adoption at the firm level. The blockers are organizational and structural, not technical. And they're fixable.

The 4-Step Framework

Strategy 1: Start Small — One Pain Point, One Tool

The pattern at firms where AI adoption fails: the initiative is too broad. "We want to use AI to improve efficiency" is not a project. It's an aspiration. Aspirations don't get completed; projects do.

Wynter's framework starts with a specific pain point — not AI as a category, but a specific workflow that takes more time than it should. Examples:

"We spend 25 hours per month on client transaction categorization that shouldn't require a senior accountant's judgment."
"Monthly reporting packages take 3 hours each to compile — that's 60 hours per month for 20 clients."
"New client onboarding document collection takes 3 weeks of back-and-forth."

One of those statements is a project. Pick it. Deploy one tool against it. Measure the time before and after. That result — hours saved, not "AI is going well" — is what justifies the next step.

What to do this week: Identify the single workflow in your practice that consumes the most staff time for the least complexity. Write down the hours per month. That number is your baseline.

Strategy 2: Invest in Learning — Time, Not Just Access

Only 17% of workers globally believe their employers are investing in the skills they need for AI, per ADP's 2026 workforce research. At small accounting firms, this typically manifests as: "We gave people access to the tool. They didn't use it."

Access is not learning. A license to Fathom, Keeper, or Claude doesn't produce results if staff don't know how to integrate it into their actual workflow. The firms seeing real adoption results have built structured learning time — not a one-hour webinar, but dedicated time to experiment with real work.

The minimum viable learning investment for a small firm:

Two hours per staff member per month designated explicitly for AI tool experimentation (not squeezed around billable work)
One shared "what worked" document where staff document the prompts, tool settings, and workflows that produced useful results
One person designated as the AI practice lead — not a new hire, just whoever is most curious and gets first access to new tools before firm-wide rollout

What to do this week: Block two hours of protected time for each staff member this month, specifically for AI experimentation on real (non-client-critical) work. Make it explicit that output quality doesn't matter — learning the tool is the goal.

Strategy 3: Communicate Transparently — Why and How, Not Just What

The second most common failure mode: AI tools introduced without explaining why they're being introduced, what the firm expects staff to do differently, and how AI fits into the firm's values and responsibilities.

When this context is missing, staff fill it in themselves — usually with anxiety about job security, uncertainty about quality standards, and confusion about what "AI-assisted" means for their professional responsibility.

Wynter's framework specifies three things firm leaders need to communicate explicitly when introducing AI:

Why this tool, why now — what problem it solves and why solving it matters for the firm's direction
The ethics policy — which tasks AI handles, which require human judgment, and what "human review" means specifically (not "review it carefully" — who does what)
AI as augmentation, not replacement — with specifics: "We're using AI to handle transaction categorization so you can spend more time on the analysis clients are actually paying for." Not abstract, not reassuring platitudes.

What to do this week: Before deploying any new AI tool firm-wide, write a one-paragraph statement of purpose that answers: what will change, what won't change, and what the oversight protocol is. Share it before rollout, not after.

Strategy 4: Maintain Human Oversight — The Non-Negotiable

This is the strategy that's easy to agree with in principle and easy to skip in practice when everyone is busy.

Human oversight means a qualified professional reviews AI-generated output before it reaches a client — not to rubber-stamp it, but to catch errors and apply judgment. For accounting work specifically:

Transaction categorization: a human reviews flagged exceptions (not every transaction, but the ones AI is uncertain about or that fall outside established rules)
Monthly reports: a licensed accountant reviews the numbers and edits the narrative before delivery
Tax prep inputs: AI-generated summaries are reviewed against source documents by the preparer
Client communications: AI-drafted emails are reviewed for accuracy before sending

The firms that have run into trouble with AI aren't the ones that deployed it — they're the ones that deployed it without defining where the human review step is. The oversight protocol doesn't need to be complex. It needs to exist, be documented, and be followed.

What to do this week: For every AI tool currently in use at your firm, write one sentence describing the human review step. If you can't write that sentence, the oversight protocol isn't defined yet.

The Sequence Matters

Wynter's framework presents these as sequential, not parallel, for a reason.

Without starting small (Strategy 1), you can't measure whether the investment in learning (Strategy 2) is working. Without transparent communication (Strategy 3), the learning investment produces anxiety rather than capability. And without human oversight (Strategy 4), the capability you build creates liability rather than results.

The accounting firms that have moved past pilot mode and into real adoption — the 19% using AI daily — didn't get there by deploying more tools. They got there by treating AI adoption as a process improvement project: specific problem, specific tool, specific measurement, and clear human accountability.

The 81% who aren't using AI daily aren't behind because they're lazy or risk-averse. They're behind because no one told them what to do next, in order, with specific enough steps to actually do it.

Your Next Step

Answer these four questions in writing before deploying your next AI tool:

What specific workflow problem are we solving? (one workflow, measurable hours)
What learning time are we committing? (hours per month, per person, protected on the calendar)
What have we communicated to staff? (the why, the ethics policy, and what changes)
Where is the human review step? (who does it, for what output, before what goes to a client)

If you have written answers to all four, you're ready to deploy. If you don't, the deployment will produce the same result as the last one that didn't take hold.

The Crossing Report covers AI adoption for professional services firm owners every Monday. Subscribe here.

Stuck in Pilot Mode? The 4-Step AI Adoption Plan for Accounting Firm Leaders