AI Won't Fix a Broken Accounting Firm: Process First, Then Tools

If your accounting firm's onboarding process is inconsistent, your tools don't share data, and your review process involves chasing down information, adding AI won't resolve these issues; it will only exacerbate them by speeding up the same problems and reducing visibility. Firms that see a return on their investment standardize their workflows first, connect their data second, automate their processes third, and finally apply AI to tasks that require judgment at volume. Skipping directly to AI implementation often creates noise instead of increasing capacity.

---

What the Research Says

• Workflow redesign showed the strongest correlation with financial returns from AI among 25 factors studied by McKinsey in its 2025 State of AI survey, surpassing budget, talent, and model selection. Approximately 70% of organizations overlook this crucial step.
• High-performing firms were nearly three times more likely to redesign workflows rather than automate existing ones.
• About 95% of enterprise generative AI pilots demonstrated no measurable impact on the bottom line in a 2025 MIT study, primarily due to brittle workflows and poor alignment with daily operations.
• Only 18% of firms actively track AI ROI; 40% are unsure if it is being measured (Thomson Reuters, 2026).

What “Broken” Looks Like in an Accounting Firm

• Onboarding varies depending on the person managing it, with no standardized checklist or consistent “ready to start” definition.
• Document collection is a manual process: sending an email with a list, receiving half the documents, chasing the rest, and consequently losing time.
• Tools don't communicate with each other, requiring data to be re-keyed between practice management, portals, ledgers, and spreadsheets.
• Preparation is inconsistent, causing reviewers to spend time fixing formatting and locating missing pieces instead of conducting thorough reviews.
• Status updates rely on individuals' memories, requiring ongoing inquiries to determine what's behind.

Adding AI to this mix results in faster but still inconsistent onboarding, automated reminders based on incomplete data, and an additional layer that cannot be audited during the review process.

Why AI Makes It Worse

• Scales the error rate. If preparation is inconsistent in one engagement out of six, AI-accelerated prep will also be inconsistent in one out of six cases, but at a higher speed and volume.
• Hides failure points. A manual chase at least indicates someone is struggling. An automated chase fails silently until a deadline is missed.
• Entrenches the mess. Integrating AI into a flawed process makes it much harder to resolve the underlying issues later.
• Costs goodwill. Teams observe the “AI project” producing rework, leading to resistance against subsequent automation efforts that could genuinely help.

The Sequence That Works

1. Standardize

   Establish one standardized method for each high-volume engagement type, including steps, owners, checklists, and criteria for “ready for review.”

2. Connect

   Integrate your core tools so that client data flows seamlessly from onboarding through preparation to billing, allowing visibility into engagement status without manual compilation.

3. Automate the rules

   Implement document-collection checklists and reminders, task routing, milestone-based client updates, and draft invoices on a consistent schedule. This step does not involve AI, but it can save significant time.

4. Apply AI to judgment-at-volume

   Utilize AI for data extraction from source documents into workpapers (with review), initial anomaly detection, drafting client communications and memos, and summarizing guidance and prior files. Now AI is working with clean inputs within a defined process, allowing you to measure its effectiveness.

How to Measure Whether It Worked

Before starting, select two or three metrics to track and compare afterward:

• Time per engagement (including preparation hours and review hours).
• Cycle time from “ready to start” to submission.
• Capacity per staff member, the number of engagements handled without compromising quality.
• Rework rate, how often engagements need to be redone.

If these numbers do not improve, the issue likely lies with the underlying process, not the AI itself.

Bottom Line

AI is a valuable tool for small firms to manage demand that exceeds their hiring capacity. However, it should be the final step in the process, not the foundation. Start with standardization, connection, and automation, then apply AI to build a robust system that delivers measurable value.