There is no shortage of commentary on what AI might eventually do to the tax function. What is harder to find, and more useful, is a frank conversation with the people already doing it.
At the
Here is what they said.
1. The Adoption Story Is Bottom-Up, Even When the Push Is Top-Down
At all three companies represented on the panel, AI adoption in the tax function was initially driven by executive mandate. CEOs and CFOs, many of whom were simultaneously building AI into their own products, set an expectation that finance and tax would find ways to apply the technology. But the panelists were candid about how slowly that translated into actual change on the ground.
"People are busy with their day-to-day jobs," said Rachel Luu, Senior Director of Global Tax at Autodesk. "It takes an investment of time to really learn AI and adopt it. We were not doing a lot in the tax group."
What finally moved the needle was not another directive from above. It was discomfort. At one company, a manager who handled the state tax notice mailbox left, and the work landed on an already-stretched analyst. The team used that moment to build an AI-assisted triage process that not only solved the immediate problem but demonstrated what was possible to the broader group. At another, a pilot of AI research tools produced an unexpected result: an indirect tax professional who, in the words of a colleague, "seems to abhor technology on a good day" became the highest-volume user on the team.
The recurring theme was that pressure and constraint, not enthusiasm, tend to produce the first real use case. "I get happy when people start complaining," said Duncan Dias, Associate Director of Finance Transformation at TripAdvisor, "because it's an opportunity to reevaluate how the job gets done."
2. Time Savings Are Real, But the ROI Is Showing Up Somewhere Else First
The panelists pushed back, gently but clearly, on the framing that AI ROI means headcount reduction in the near term. The more immediate and measurable return is showing up in two places: external advisor spend, and the quality of work the internal team is producing.
Jeremey Lees, The Head of Tax Transformation at Atlassian described cutting scope on two recent acquisitions: sales tax nexus and exposure analysis, and RSU deductibility analysis, because the team felt confident enough using AI research tools to handle that work internally. At Autodesk, Rachel described preparing substantial portions of the Pillar Two Globe Information Return herself and sending only a review copy to her advisors, rather than commissioning the full preparation. "Instead of them preparing the whole thing, I could ask follow-up questions all weekend," she said. "It's just like talking to a friend."
The honest accounting, though, was that the longer-term financial benefits are on a slower timeline. "Time savings are real, but you can't take them to the bank this quarter," said Jeremey. "We're raising the bar of quality before we're minimizing expense. The savings would be baked into our long-range plans."
A tax technology leader in the audience offered a useful reframe: ROI in tax does not always appear on the P&L. Avoiding a penalty, protecting against audit exposure, or reducing the time it takes to respond to a notice has real dollar value even when it is invisible to a CFO reading a financial statement. Getting alignment across finance on what ROI actually means, she argued, is as important as the technology itself.
3. The Advisor Relationship Is Changing, Not Ending
One of the most substantive discussions on the panel concerned the Big Four. The short version: advisors are not going away, but the way companies engage them is already different.
Jeremey Lees described using AI tools not only to cut scope from existing advisory relationships, but to lower the barrier to entry for firms that had never previously worked with the company. "We have a firm that has always gotten first dibs because they know us, and we don't pay for the learning curve," he said. "Now, another firm can come in and quickly get up to speed. We're not paying for that learning curve in the way we would have before."
Rachel Luu described an informal rule on her team: before calling or emailing an advisor, you run the research in Accordance first. The rule is not about distrust. It is about efficiency and professional development. "They've told me it's been wonderful," she said. "We have been saving fees."
Rachel also noted that the large firms are watching this shift and adapting. KPMG AI, she pointed out, now includes a button that lets users send their prompt and their issue directly to their engagement team. "They're smart," she said. "They're making sure they keep their client base."
The panel's broader conclusion was that the makeup of advisory spend is changing, not the total. The share going to Big Four firms for technical research and rote preparation is shrinking. The share going to technology partners is growing. And the work that does go to advisors is increasingly strategic, informed, and scoped more precisely because the in-house team now has a better baseline.
4. The Unexpected Power User Is the Most Important Data Point
Every panelist had a version of the same story. They expected early adopters to come from the people already comfortable with technology. They were wrong.
At TripAdvisor, it was someone in transfer pricing. At Autodesk, it was the M&A planning lead, who built her own persistent project environment to handle integration diligence questions without routing through an external advisor. At Atlassian, it was an indirect tax professional who ran more queries than anyone else on the team during the pilot. At Autodesk, indirect tax and withholding tax became among the heaviest users overall.
The pattern suggests something important about where these tools create value. It is not primarily in automating routine work. It is in the gray-area moments: the judgment calls, the novel fact patterns, the questions a junior team member would have previously had to escalate or sit on until they could get time with a more senior colleague.
"For the tax counsel or the attorneys on teams facing those out-on-a-limb judgment calls, it can be lonely," said Jack Moore. "That's where these tools earn their place."
Rachel Luu offered her way of working with tax AI: "I use it as a learning tool. I'm not a US International specialist. But we get questions. I don't want to keep bothering people, so I go into Accordance and ask. I love that."
5. What Monday Looks Like
The panel closed with a practical question: what should someone in the audience actually do next week?
The answers were consistent and concrete. Start with a pain point, not a strategy. Find the person on the team who is most frustrated with a recurring manual process and build something small for them. Create a shared channel where the team can post what they are trying. Make it safe to experiment. Don't wait for the data infrastructure to be perfect.
A memorable piece of advice came from an audience member who had been listening quietly.
"Treat the tool the way you would onboard a new senior associate," she said. "Have the conversation. Get to know it. Give it tasks. Review its work. If you would not hand work to a junior associate without checking it, do not do that here either. But if you would not refuse to work with a junior associate just because they are new, do not do that here either."