Most Accounting Firms Are Stuck at Level 1 AI — Here's What Level 2 Looks Like

On a Tuesday morning in April, a managing partner at a 14-person CPA firm sat down to a problem she'd seen every month for the past year. Her team had been using AI tools for eight months. She'd invested in ChatGPT Plus licenses for everyone, gotten the Copilot feature in QuickBooks turned on, and scheduled two all-hands training sessions.

And every Tuesday morning, two staff accountants were still spending half their day manually pulling transaction data into a spreadsheet before anyone could start the real work.

"I thought AI was supposed to handle this," she told her operations manager. "Why are we still doing this step by hand?"

The answer isn't that her AI tools are broken. It's that her firm is operating at Level 1 — and Level 2 requires something she hasn't built yet.

Wolters Kluwer published a framework in early 2026 that defines where small accounting firms actually are, not where vendors say they should be. It maps four distinct levels of agentic AI — each requiring different infrastructure, producing different outputs, and enabling different business models. According to their research, 70% of U.S. accounting firms now use AI weekly. But the firms that have crossed into Level 2 are 87% more likely to report revenue growth compared to firms still running fragmented technology stacks.

That gap is exactly where the leverage is for a 10-25 person CPA firm in 2026.

---

The Four Levels of Agentic AI for Accounting Firms

The Wolters Kluwer framework — Taskers → Automators → Collaborators → Orchestrators — describes not just what AI does at each level, but what your firm's infrastructure has to support to get there.

Level 1 — Taskers (What You're Already Doing)

Level 1 AI automates discrete, repetitive tasks: document sorting, data entry, basic classification, drafting a first pass of a client email. This is where most small accounting firms live in 2026.

ChatGPT for draft emails and research. Copilot in Word or Excel for formatting. QuickBooks AI suggestions for transaction categorization. These tools reduce friction on specific tasks — they are genuinely useful. But they don't eliminate process steps. A human still moves data between systems. A human still initiates each task. A human still reviews and routes each output.

The defining characteristic of Level 1 AI: it replaces keystrokes, not workflows.

Level 2 — Automators (Where the Gains Actually Start)

Level 2 AI runs complete processes end-to-end without human intervention at each step. The defining example for accounting firms is the transactions-to-trial-balance workflow.

A Level 2 implementation doesn't just categorize transactions — it ingests the bank feed, categorizes transactions against the chart of accounts, flags exceptions for human review, and generates the trial balance. The accountant reviews the exceptions. They don't touch the process steps at all.

This is where Wolters Kluwer documents 3-4x efficiency gains. A workflow that took an accountant three hours per client per month now takes 20-45 minutes of review time. The reclaimed hours shift to advisory work, additional client capacity, or reduced overtime pressure during peak season.

Level 2 is the crossing point. It's also where most small firms stall — not because the tools don't exist, but because the infrastructure isn't there.

Level 3 — Collaborators (The Review Layer)

Level 3 AI provides intelligent guidance during complex workflows — where judgment still matters but can be augmented. Exception flagging with explanation. Anomaly detection with historical precedent. Review assistance that surfaces the two items most likely to need attention before the partner even opens the file.

A Level 3 accounting firm has AI embedded not just in process steps, but in the review conversation itself. The accountant isn't just approving a trial balance — they're working with an agent that has already identified the revenue recognition discrepancy from last quarter and surfaced the three transactions that don't match client patterns.

Few small accounting firms operate at Level 3 today. It requires both the infrastructure of Level 2 and a layer of firm-specific historical data — engagement patterns, client benchmarks, firm-level risk parameters — that agents can reason against.

Level 4 — Orchestrators (The Integrated End-to-End)

Level 4 is where multiple agents coordinate across the full client lifecycle. Client intake through filing-ready return. Engagement kickoff through billing. No manual handoffs between systems.

This is where the most sophisticated firms are headed in the next three to five years. For a 10-25 person CPA firm, Level 4 is not a 2026 objective — it's a 2028-2029 planning horizon. Understanding it matters for infrastructure decisions you make now. Trying to build it now is a distraction from the more immediate opportunity.

Your firm's 2026 mandate is specific: get to Level 2.

---

Why Most Small Accounting Firms Stall Between Levels 1 and 2

The Ohio managing partner from the opening scenario has the tools, the intent, and staff who are willing to use AI. And still, her Tuesday morning looks the same.

The problem isn't the tools. It's three structural foundations that Level 2 requires — and that most firms don't have in place before they start evaluating AI agents.

The Three Foundation Problems

Wolters Kluwer's analysis identifies three gaps that block the Level 1 → Level 2 transition:

1. API-first platforms. Level 2 automation requires your core tools to exchange data automatically — not through CSV exports, manual copy-paste, or human-mediated transfers. An AI agent can't run a transactions-to-trial-balance workflow if it has to ask a human to export the transaction file first. This is the most common Level 2 blocker for small CPA firms: their accounting platform doesn't talk directly to their document management system, practice management tool, or AI layer.

2. Role-based access controls (RBAC). AI agents operate on data. The question of which agent should access which client's data — and with what permissions — must be defined before you deploy agents, not after. Most small firms have never explicitly mapped this. When they try to deploy an agent, they discover they don't have a clear answer to a basic question: should this AI tool see all clients, only active engagements, or only the specific client file it's currently working on?

3. Clean data governance. Agents work on what they're given. If clients send documents in seven different formats, using three different naming conventions, arriving through four different channels, the agent spends its cycles on reconciliation — not accounting. Firms making the fastest gains from Level 2 automation are often the ones who, six months earlier, standardized their client document intake process. That's not an AI project. It's a process discipline project that makes AI effective.

API-First Platforms: What This Actually Means for a 10-Person Firm

"API-first" sounds like an IT requirement. For a small firm owner, it means one practical thing: can your core tools share data without a human carrying it between them?

QuickBooks Online: Yes. Robust API with thousands of integrations — AI agents, practice management platforms, document management systems.

Xero: Yes. Same profile as QBO — API-first, widely supported by AI tooling and workflow automation platforms.

QuickBooks Desktop: Generally no. Desktop accounting software was not designed for API connectivity and requires middleware or manual exports to share data with external systems.

Older Sage versions: Often no. Enterprise-grade integrations exist, but they require configuration complexity out of scope for most small firms.

If your core accounting platform doesn't offer native API connections, that's your Level 2 blocker — not your AI tool selection. No AI agent will fix a data handoff problem. The agent will just fail faster.

Data Governance: The Unglamorous Pre-Condition

A firm owner once described her data problem this way: "Our clients call the same expense category four different things. So when the AI tries to categorize it, it makes a different choice every time. Then we spend more time correcting the AI than we saved."

This is the reality in most small CPA firms. Transaction categories are named inconsistently across staff. Chart of accounts templates vary by client relationship or the partner who set up the account. Client-provided documents arrive labeled "March Invoice" or "Final_v3_REAL_USE-THIS" or nothing at all.

You can't automate a process built on inconsistent inputs. The investment in standardizing client document intake, naming conventions, and chart of accounts templates is not glamorous. It's also not optional if you want Level 2 to deliver the efficiency gains the Wolters Kluwer data documents.

The good news: this is a one-time cleanup, not ongoing overhead. Firms that complete it report that the process standardization itself — even before any AI agent is deployed — reduces monthly errors by 30-40%.

---

What Level 2 Looks Like in Practice: The Transactions-to-Trial-Balance Workflow

Here is the Level 2 accounting workflow in concrete, before-and-after terms for a firm on QuickBooks Online:

Before Level 2 (current state for most firms):

1. Staff accountant opens QBO, manually reviews uncategorized transactions
2. Exports transaction data to Excel for further review
3. Formats trial balance spreadsheet manually
4. Reviews for anomalies by eye
5. Generates report and routes to partner for review
6. Total staff time: 2-4 hours per client per month

After Level 2 (agentic workflow):

1. Bank feed syncs automatically overnight via QBO native connection
2. AI agent categorizes transactions against historical patterns and chart of accounts
3. Agent flags exceptions — transactions it can't categorize confidently, anomalies vs. prior periods — with context
4. Trial balance generated automatically
5. Staff accountant receives a review queue: exceptions only, with the agent's reasoning for each
6. Total staff time: 20-45 minutes of review per client per month

The efficiency gain is real. Firms in the Wolters Kluwer data reporting the 87% revenue growth correlation are running workflows like this across their entire client base. The accountant's work becomes review and judgment — the part that actually requires a CPA.

But notice what made this possible: the API connections were configured before the agent was deployed. The access permissions were set. The chart of accounts was consistent enough for the agent to categorize accurately. The Level 2 outcome was built on the three foundations — not on the AI tool itself.

---

The Self-Assessment: Which Level Are You?

Answer three questions honestly:

Question 1: When your team uses AI tools, does a human have to manually move data into the AI tool to start each task?

• Yes → You're operating at Level 1. AI is assisting individual steps, not running workflows.
• No → You may have Level 2 infrastructure. Keep going.

Question 2: Do your core platforms — accounting software, document management, and practice management — connect via API, or do staff export and re-import data between them?

• Export/re-import → Level 1 ceiling. You'll hit the API problem before Level 2 automation delivers value.
• API connections → Level 2 foundation exists. Now check your data governance.

Question 3: Are your client document intake process, transaction categories, and chart of accounts templates consistent across all staff and all clients?

• Inconsistent → Level 2 will underdeliver. The agent will spend its time on reconciliation, not accounting.
• Consistent → You're ready to evaluate Level 2 agent tools with confidence that they'll work.

If you answered "Level 1" or "export/re-import" to any question, that's your first action — not evaluating new AI products.

---

Your One Action: Audit Your Foundation Before Adding More AI

Open a blank document. List your three core platforms: accounting software, document management system, and practice management or communication tool.

For each one, answer a single question: does it offer native API connections to other tools?

• QuickBooks Online → Yes
• Xero → Yes
• QuickBooks Desktop → No (without a third-party connector)
• FreshBooks → Limited
• Google Drive with manual uploads → No
• SharePoint via M365 → Yes, with configuration

If any platform doesn't offer API connections, you've found your Level 2 blocker. That's the conversation to have — with your current vendor about their integration roadmap, with a replacement option, or with a practice management tool that bridges the gap.

This audit takes 30 minutes. It will tell you more about your actual AI capacity than any AI tool demo. The firms in the Wolters Kluwer 87% cohort aren't running better AI tools — they built the foundation first and then let the agents run.

---

Frequently Asked Questions

What are the four levels of agentic AI for accounting firms?

According to Wolters Kluwer's framework: Level 1 (Taskers) automates repetitive work like document sorting and data entry. Level 2 (Automators) runs complete processes like transactions-to-trial-balance end-to-end. Level 3 (Collaborators) provides intelligent guidance during complex workflows — flagging exceptions, assisting review. Level 4 (Orchestrators) coordinates multiple agents from client intake through a filing-ready return. Most small accounting firms in 2026 operate at Level 1.

Why do most accounting firms stall between Level 1 and Level 2 agentic AI?

Wolters Kluwer's analysis identifies three missing foundations: API-first platforms (tools that connect via open APIs rather than requiring manual data export or import), role-based access controls (defining which AI tools can access which client data), and clean data governance (consistent, structured data that agents can query without manual correction). Most small firms are missing at least one. Agents can't run end-to-end workflows if a human has to hand-carry data between steps.

What does a Level 2 agentic AI workflow look like for a small accounting firm?

The most common Level 2 implementation for small CPA firms is the transactions-to-trial-balance workflow: bank feed ingestion, then transaction categorization, then exception flagging, then trial balance generation. With an API-first platform like QuickBooks Online or Xero, an agentic AI can run this workflow overnight and deliver a review-ready trial balance each morning. The accountant reviews exceptions rather than performing each step — reducing a 3-4 hour monthly task to 20-45 minutes of review.

Is the 87% revenue growth statistic for integrated accounting firms accurate?

Yes — this figure comes from Wolters Kluwer's 2026 research on accounting firm technology integration. Firms where 75% or more of their technology stack is fully integrated reported revenue growth at significantly higher rates than firms with fragmented tools. The causal mechanism isn't just AI — it's the integration foundation that makes AI effective. When tools share data cleanly, time spent on manual reconciliation decreases and capacity shifts to advisory work.

What should a 10-person accounting firm do to move from Level 1 to Level 2?

Audit your three foundations before adding new AI tools. First: identify which of your core platforms offer API connections. Second: define which staff roles should have access to which client data in AI-assisted workflows. Third: standardize your input data — if client documents arrive in inconsistent formats, clean that process before automating the review. The Level 2 upgrade path is 80% infrastructure and 20% AI tool selection.

---

The Crossing Report covers AI adoption for professional services firm owners every Monday. Subscribe for the weekly intelligence brief.