Starting today, I'm sharpening the focus of this newsletter. Count On AI is now Applied AI for the accounting profession — Claude-forward, multi-tool when it matters, tested against real client work instead of vendor decks. Tuesday cadence stays. Skeptical eye stays. What's new: a weekly Claude Column with prompts, security guardrails, and workflows you can run before lunch. Today's lead is exactly the kind of piece you'll keep getting.
Most firms don't have an AI strategy. They have an AI receipt pile.
The 50-person regional firm down the road is spending $4,200 a month on AI tools. The solo CPA next door is spending $25. Both are seeing real ROI.
The gap between firms with a real AI stack and firms with one wishful subscription has gotten wide fast. Here's what five different firms are actually paying for in May 2026.
Firm 1: Solo CPA, Texas. Total stack: $25 a month.
Just ChatGPT Plus. Uses it for client emails, first-pass tax research, and engagement letter drafts. No firm management software, no AP automation. The whole firm fits on a personal credit card and a Calendly link. Margin is fine because there's no payroll except his own.
Firm 2: Small tax firm, 5 staff, Arizona. Total stack: roughly $680 a month.
Karbon Team at $59 per user, ChatGPT Team at $25 per user, plus 1040SCAN through their tax software. The Karbon line item is the big one — $295 for the team, plus another $125 in workflow add-ons. Seasonal staff get dropped onto ChatGPT Team during March and April only.
Firm 3: Mid-size CAS firm, 20 staff, Ohio. Total stack: about $2,100 a month.
Canopy Standard at $150 plus Workflow and Document Management modules (around $1,000 loaded across 20 users), Vic.ai for the AP side (custom, roughly $800 a month for their invoice volume), and Microsoft 365 Copilot at $21 per user for the leadership team. They were the first firm in their county to move bookkeepers from manual entry to AI-assisted review. Bookkeeper hours per client are down 40%.
Firm 4: Audit-focused firm, 15 staff, Pennsylvania. Total stack: $1,400 a month.
MindBridge for audit risk scoring, Blue J Tax for technical research, Microsoft Copilot for general productivity. They don't pay for general practice management AI. Their 15-year-old engagement platform works and the partners don't want to retrain. Pitches from new vendors get ignored unless the tool touches sampling or analytics.
Firm 5: Multi-service regional, 50 staff, Colorado. Total stack: $4,200 a month.
Karbon Business at $89 per user across leadership (about $1,800), Blue J Tax for the tax department, SurePrep 1040SCAN, Vic.ai for AP, MindBridge for audit, and Microsoft 365 Copilot deployed firmwide. The CFO can give you ROI by department because she set it as a KPI before signing the first contract.
What none of them are paying for: dedicated "AI compliance assistants," voice-cloning client check-in tools, generic "AI for accountants" Chrome extensions, or any product whose website is more polished than its case studies. Firms that buy on hype don't make it into stacks like these. They make it into next quarter's churn report.
Canopy announced an AI-powered bookkeeping module set for summer 2026, designed to plug into their existing practice management stack. The play here is obvious: stop losing CAS work to bookkeeping-first competitors. Worth a hands-on demo before you commit to Botkeeper or a Bench-style outsource, especially if you're already on Canopy.
Their joint platform went GA in the US, Canada, and UK earlier this year. If your firm touches expats, foreign holdings, or transfer pricing, three-hour treaty research sessions are over. Same pricing tier as standard Blue J, no incremental charge for international queries.
That's the documented number from a single mid-market customer — an 84% reduction in labor cost per invoice. Year-one savings around $250K, growing to $600K by year five. The math only works above 500 invoices a month. Below that, the subscription eats the savings.
Copilot at $21 per user beats ChatGPT Team at $25 per user, but only if you're already on a qualifying Microsoft 365 plan. For most tax firms on Microsoft, Copilot is cheaper once everything's added up. For firms living in Google Workspace, ChatGPT Team is still the simpler call.
The labor cost per invoice reduction one mid-market firm saw moving from manual AP to Vic.ai. From $3.30 to $0.53. The catch: it took 18 months for the AI to fully calibrate to their chart of accounts. Year one was the smallest win. Year three is when the line item disappears entirely.
What it is: AI-powered tax research platform combining GPT-4.1 with a proprietary library of authoritative primary sources, Tax Notes commentary, and (as of Q1 2026) the IBFD international tax database. A real RAG system, not a ChatGPT wrapper.
What it does well: Predicts case outcomes with 90% claimed accuracy. The diagramming feature for ownership structures and tax flow is faster than anything you can do in Word or Visio. Cross-border queries now return cited authority in seconds rather than hours.
What it doesn't do well: Not a replacement for a tax partner's judgment on novel facts. You still own the position. Pricing is opaque — published tiers don't exist, so expect a sales call before you see a number.
Pricing: Custom. Plans for sole practitioners, regional firms, and national firms. CPA.com members get preferred pricing through the partnership. Schedule a demo at bluej.com/get-started.
New section. Every Tuesday from here on out.
Most accounting newsletters cover AI like sports commentary: talking about it, not from inside it. The Claude Column is different. Each week you'll get one of four things: a Claude prompt you can paste and use today, a security guardrail you can't afford to skip, a Claude workflow walkthrough, or a real-world case from inside a firm. Claude leads because it's the AI I lean on hardest for accounting work. I'll still call out when ChatGPT, Copilot, or Gemini does a thing better.
This week: a prompt.
Prompt of the Week — The Email Thread Reconciler
Paste any long client email chain into Claude. Use this prompt:
"Read this email thread and give me three lines: (1) what the client originally asked, (2) what was agreed to, and (3) what's still open. Quote dates if relevant. No commentary."
Use it when:
A client is asking why their refund hasn't come through and the thread is 14 emails deep.
You inherited a thread from a colleague mid-engagement.
You need to brief a partner on a client situation in under 60 seconds.
Why it works: Claude is unusually good at long-context summarization (200K context window), and "no commentary" kills the unnecessary preamble most AI tools add. "Quote dates" forces grounding in the actual text instead of inference.
Where it fails: if attachments hold the real answers, Claude only sees the email body. Paste the attachment content too, or note that pieces are missing.
Try it once this week. Reply and tell me what you used it for. I read every reply.
Pull last month's firm credit card statement and audit every AI subscription in 30 minutes. List the tool, the monthly cost, and which staff member actually uses it at least once a week. Most firms find 20% of the line items haven't been opened in 60+ days. Cancel those before the next renewal cycle.
P.S. — Reply with one Claude question you want answered in a future Claude Column. The next four weeks rotate through prompts, security, workflows, and real-world cases. Subscriber questions move to the front of the line.
