AI in accounting is no longer a forecast. Our four-part Beyond Accounting Bitesize series asks what comes next: how to control AI, prove its value, build with it, and develop the skills to use it well.
Each year, the date has moved closer. When we began our AI and accountants presentations in 2023, our speakers thought AI wouldn’t really impact the profession until 2035. In 2024 they said 2030, and last year 2027. This year, Chris Easton put it simply: “This is happening now.”
Our audience was already there. Earlier in the session, 79% of poll respondents said AI’s significant impact on audit and accounting is already happening, and 85% said their company uses AI.
AI & Accountants 2026 is a four-part Beyond Accounting Bitesize series drawn from the Chartered Accountants Worldwide Network USA (CAW Network USA) webinar of the same name, our follow-up to last year’s AI & Accountants session. Host David Powell, Chief Executive of CAW Network USA, was joined by Campbell Robertson, founder of Boxwood Strategy Group and a former global data and AI practice leader at IBM; Jannie Wentzel, Chartered Accountant and partner and chief operating officer at Cential; Chris Easton, CPA and Chartered Accountant, CEO of Applied Logic, and a past president of CAW Network USA; and moderator Nancy Chakabuda, Chartered Accountant, US CPA, and Vice President of CAW Network USA.
Part 1: From Adoption to Value
Campbell opens with a gap. On the figures he cites, 97% of finance departments report some AI in use, yet only 6% of finance leaders call their adoption mature or enterprise-wide, and only 18% of professional service firms track ROI at all.
The shift that matters is from assistant to agent. An assistant answers and a person decides; an agent carries out multi-step tasks across systems, so the control point moves, and often nobody has said where.
He walks through three client examples. In month-end close, the bottleneck was a dirty chart of accounts, not the model. In exception handling, an agent found anomalies nobody had had time to look for, and an efficiency project became a disclosure question overnight. His summary: “Every one of these started as a productivity project. All three ended up as a controls project.”
His four questions for your own firm:
• Inventory: what AI is in use, including unofficial shadow AI?
• Approval: where is the human approval gate, and is it documented?
• Evidence: what trail could you show a regulator or auditor next quarter?
• Value: what was the measured benefit against the measured cost?
With no binding AI-specific auditing standard as of the webinar, he says, build that documentation trail now.
Part 2: The Risk of Not Using AI
Jannie takes governance a step further. To govern is to oversee direction as well as control, he points out, yet AI governance has focused largely on control. Of three AI risks (using AI, external threats from others who use it, and not using it), he believes the profession tends to neglect the last.
His firm of about 20 people uses AI agents to deliver a risk assessment with about half the team it once needed. Analyzing the last couple of years of financial statements for trends used to take a day or more; with an agent, it takes about 20 minutes of his time. Smaller organizations, in his view, may have the edge: they are more nimble, and integration hurdles are fast disappearing.
He warns against leaving AI governance to IT alone; bringing in strategy and operations lets governance enable innovation as well as control risk. And AI isn’t necessarily a separate risk, he says, because it shapes how almost every risk is managed: “don’t think of AI in isolation. It’s pervasive in everything we do.”
Part 3: Building Vertical AI – From Idea to Launch in Under a Year
Chris turns from governing AI to building it. While the largest technology companies spend billions on the models, he argues, the only defensible ground for application developers is vertical AI: deep expertise and data integration applied to one industry’s problem.
His case study is Ketelo, the AI platform for governance, risk, and compliance that his own team built from an idea Jannie raised after last year’s session. Work began in late January 2026, with a minimum viable product by April, a customer deployment in May, and a launch at a San Diego conference.
The large language model itself, he says, is “a small part of it.” Most of the work sits around it: grounding it in company data, guiding agents through processes that can run to 200 tasks, connecting to business systems through MCP servers (connectors that let AI discover a system’s tools and data), and keeping decisions recorded and auditable.
The aim is proactive risk management that monitors controls continuously and responds to risk signals: rising interest rates, in his view, should have triggered a risk assessment at Silicon Valley Bank a year before it collapsed. For accountants who find the right niche, he concludes, AI is an opportunity, not a threat.
Part 4: Panel Q&A – Asking the Right Questions
Nancy opens the panel with a poll result: 65% of respondents agreed that AI is not poised to replace bookkeepers and accountants. Campbell’s view: AI replaces tasks, but the human remains accountable, and knowledge workers are becoming intuition workers.
On control, Chris describes building an agent to control the agents, and Jannie advises starting with as many human-in-the-loop steps as practical, then removing them as comfort grows. Campbell is blunt about accuracy: a hallucination “is an error, and an error still needs to be captured by the human.”
On cost, Campbell describes token maxing, a Silicon Valley practice of rewarding staff for heavy AI use, which left some organizations paying more for AI than for the person they would have hired.
On skills, 90% of poll respondents said AI and data analytics proficiency will become a requirement for new hires. Jannie challenges the data analytics part: “You need to have the ability to ask questions. You don’t need to necessarily analyze. The agent can do that for you.”
In closing, Campbell urges firms to watch the cognitive impact of overusing AI and to plan a human fallback if AI goes down or changes in the middle of month-end. Jannie’s message: don’t be afraid, but embrace AI in a controlled manner. Chris calls this a time to define the opportunity.
The common thread
Across the series, the AI itself is no longer the hard part. The hard part is everything around it: the controls Campbell describes, the direction Jannie wants governance to set, the infrastructure Chris built, and the questions only an experienced professional knows to ask. In the final poll, 91% of respondents said they embrace AI to enhance how they do their job. These four episodes show how to do that without losing control.
Listen, watch, and join us
Listen to all four episodes of AI & Accountants 2026 on Beyond Accounting Bitesize, and follow the show for the next series.
Or watch the whole event on demand: Watch the full session »
Join us for our next Beyond Accounting webinar, Women in Finance – Empowering Digital Leaders for the Future of Finance, on Thursday, October 22, at noon Eastern. Register now »
How we make this podcast
Beyond Accounting Bitesize episodes are edited from our webinar recordings. For this series, the introductions and closing notes were written with AI assistance, and reviewed and approved by our team.