The "should we use AI" conversation is over. Ninety-eight per cent of accounting professionals already use AI in some form, and sixty-three per cent believe a firm loses value if it doesn't. Most of my days are spent talking to firm leaders and their teams about how AI actually shows up in the work, and I walked through what I'm seeing at the AI in Practice Summit. If you would rather watch the demos, the recording is here.
Here is the part that should give you pause. While almost everyone uses AI, only twenty-one per cent of firms have a written AI strategy, and only twenty-one per cent have an AI policy. One in five. A strategy is the shared picture of where AI fits in the firm. A policy is the guardrails: what goes into AI, what stays out, which tools you use and how AI shows up in client work. Most firms have neither.
Everyone is in the boat. Almost nobody is rowing together.
The picture I keep coming back to is five people in a boat on rough water. Two are rowing in sync and clearly have a handle on it. One is rowing backwards and creating drag for the rest. Another is paddling hard with the wrong tool. The last sits idle in the back. Everybody is getting splashed, everybody is in the same turbulent water, and they are pulling in different directions.
That is most firms right now, and the cost of it is measurable. Teams with an AI strategy save twenty-four per cent more time than teams without one. Firms with a policy save twenty-three per cent more. Same tools, same Claude. What separates them is whether they wrote down what they were doing and gave their teams some guardrails. Intention is the whole game.
The value lives where your tools overlap
Think of AI in your firm as three zones. Your core is the tech stack you already run on, the Karbon, Xero, QuickBooks, payroll and tax software your team lives in. Your lab is where you experiment, so that is Claude, ChatGPT and Lovable, the tools sitting off to the side. The zone that matters most is the overlap in the middle, where connectors, MCP and native AI functionality bring the two together.
The firms doing this well lean into that overlap. Replacing your core stack with AI backfires, and running AI off to the side leaves you with party tricks that stay one-offs. The value shows up when Claude can reach into the systems your team already uses.
The most boring thing you can do is also the most strategic
So start with SOPs. That is the least exciting place to begin, which is exactly why so few people talk about how much rests on it. Forty-three per cent of firms already use AI for SOPs, the fastest-growing use case in the profession, and the reason is simple. SOPs are where the gap between what a firm says it does and what it actually does is widest. The quiet stretch around a new financial year is the right time to close it.
Most firms have a messy library: outdated documents, three different formats, a PDF here and a Word doc there. Drop that whole folder into Claude and ask it to bring the SOPs up to date and tell you what is missing. Then ask it to walk you through the gaps one question at a time, because a wall of questions overwhelms and a single question gets answered. Better still, record the people who actually run a process talking through it, the way we did for a payroll run with a junior, a senior and a manager, then hand Claude the transcript. It produced a structured six-page SOP and flagged the genuine gaps the call left open, like a missing approval step and an undefined cutoff time.
This carries more weight than it looks. As AI agents arrive, your SOPs are what they will follow when they run parts of your workflow on their own. Clean documentation turns your operating system into something an agent can act on, and it shifts your team from doing the work to reviewing it.
Your best advisory context is already in your transcripts
A client is more than their profit and loss. The themes that should shape an advisory call are usually sitting in last quarter's meeting transcript and across your practice management system, and they are the first things a busy adviser forgets. So let Claude read the lot before you prep.
For one client I combined a profit and loss, a balance sheet and the transcript from last quarter's meeting, then let Claude pull the live records from Karbon through a beta MCP connection: the emails, the work items, the back and forth between calls. Without being asked, it applied the firm's brand-voice skill and produced both a partner prep doc and a client deck. The prep doc surfaced what I would want to know before walking in: a slipping gross margin, a rising line of credit and growth that had stalled.
Two building blocks make that repeatable. Projects are folders that hold the context for a client or a topic and can be shared across the team, so everyone touching a client works from the same place. Skills are reusable prompts for anything you do more than once, so the way I prep an advisory call can become a skill my next great adviser uses, which gives the firm a consistency that people rarely manage on their own.
The same context turns a mess into something a client can read. One demo takes two five-hundred-line general ledger exports from QuickBooks and builds an interactive, on-brand dashboard with a cash-flow waterfall and a plain-language summary. When a client asks why a number looks off, you drill straight into it. Less formatting, more conversation.
Start inside the systems you already run on
Land where the numbers point. The firms winning with AI are the ones using it with the most intention. Begin inside the systems your firm already runs on, build alongside and within your core rather than replacing it, then extend with Claude for what those systems don't do natively yet. Go back to your firm with a strategy that compounds rather than a drawer full of one-off tricks. You can watch the full session and every demo here.
