AI for Accounting Firms.

For a lot of firms, a Team plan plus some training is genuinely the right starting point. The work I do is what comes after that — when you've realized your staff aren't asking ChatGPT, they're still asking the senior partner, because the answers live in your SOP doc and your email archive, not on the public internet. If your team isn't already using off-the-shelf tools, start there. Come back when the bottleneck is your own knowledge.

No. The CRA correspondence use case drafts a summary, suggests an approach, and flags the deadline. A partner reads it and decides what to do. Same for engagement letters and pre-auth-style narratives — drafted, never sent. The point of the build is to stop letters from sitting in someone's inbox for two weeks, not to remove the human from the legal and professional liability chain.

For a firm of ten to twenty staff, the build I described in the accounting note ran a fraction of the $30k another shop had quoted. Typical range for this kind of internal knowledge tool is $6k to $20k depending on how messy the source documents are. Cleaning the SOP doc is often the largest single line item — more than the AI part. Annual hosting and model costs after that are usually a few hundred a month.

I don't ship anything in February. Builds either go live in October-November so they're hardened by the time March hits, or in May-June after season ends. If something breaks during season, the tools are designed to fail closed — staff can still do everything by hand, the AI just stops drafting. The worst case is your March looks like last March. I'd rather build that fallback in than promise it'll never break.

Sometimes. If you're a one-partner firm with a few staff, the document classification and client onboarding pack drafts often pay back fastest. The internal SOP search is overkill — when there's one person who knows everything, you don't need a search tool over your own brain. The plain-English client summary use case tends to be the surprise win for sole practitioners with a lot of owner-operator clients.