Nick | AI for Accountants

I'm a Big Four auditor who spent last year testing AI in audit work. What I learned: 90% of accountants are using AI wrong. Not because the tools are bad, because nobody showed them how. I wrote down my 50 best prompts. Free PDF: promptsforcpas.com/free/50-ai-pro…

Friday challenge for every CPA reading this: Copy 50 transactions into Claude. Add your chart of accounts. Ask it to categorize with confidence levels. 30 minutes. You'll know exactly where AI helps and where it breaks. Worth more than 10 articles about AI. Drop your results below 👇

A week into Claude Opus 4.7. What matters for accountants: → Better at long documents (contracts, 10-Ks, board packs) → More reliable multi-step tasks → Higher res vision (receipts, scanned docs) Still $20/month. Still doesn't matter if you haven't started.

China is 2.7% behind the US in AI. Despite 23x less investment. AI access is global now. Everyone has the same tools. The only advantage left: how fast you adopt. Starting today > starting with a better model next quarter.

my full AI stack as an auditor with Big Four experience: Claude Pro: $20 ChatGPT Plus: $20 Copilot: included Grok: free Perplexity: free tier + Intuit Assist in QuickBooks $40/month. 5-7 hours back per week. not complicated math.

OpenAI is building agent platforms for Fortune 500. Anthropic locked its most powerful model for cybersecurity only. For a 5-person CPA firm? Your AI stack is still: → $20/month Claude or ChatGPT subscription → Intuit Assist (already in QuickBooks) → Better prompts Don't let the enterprise arms race distract from the basics.

"We tried AI" = 3 prompts and gave up. "We use AI" = one workflow that saves 3 hours a week. Same tech. Different outcome.

@bradncpa the cost side before demand side is the part most founders skip. seen it in audit too, scaling a broken margin just scales the problem faster.

A home service business is ready to scale when two things line up. Their profit margin hits industry benchmarks and their unit economics work. And more leads actually come in if they spend on marketing If you spend more and the leads aren't coming, it could be a sign that the market might be tapped or the competition is too strong. Scaling at that point means going into new regions or adding new services. If the leads respond to spend, the next question is labor. Can you get enough guys to do the work without scheduling jobs six months out? Six months out for a standard job is a sign that something is broken. Either the cash is too thin to fund materials or the labor pool is too small. The businesses that scale well solve the cost side first. Then they figure out if the demand side will follow. Revenue is the last piece. Follow me @bradncpa for financial insights for home service businesses.

60% of finance teams are "piloting AI." 7% of CFOs see actual impact. Piloting ≠ implementing. That gap is the whole story.

100% agree. In accounting, AI multiplies the output of strong CPAs dramatically, especially on repetitive high-volume tasks like transaction testing, risk flagging and drafting client summaries. The biggest gains I see come when someone already knows what “good” looks like in audit or tax work.

hot take :) The biggest and most productive people in the AI era are the folks who are already good at their jobs. AI as a multiplier, not an equalizer/democratizer

This is the classic accounting-IT dance I see every week. The real fix isn’t faster tickets — it’s giving accountants simple AI tools so they don’t need to open a ticket for basic stuff like clearing cache, categorizing transactions or fixing common payroll errors. I built prompts that let CPAs handle 80% of these “urgent” issues themselves with Claude in under 2 minutes. Reply “PDF” if you want the 50 practical prompts that actually reduce these back-and-forths with IT.

I'm sipping coffee when a ticket pops in from accounting. 'Urgent: Can't access payroll system.' Sounds bad, right? But they didn't attach the error screenshot. Per protocol, I ping them for it. No response in 5 minutes? SLA says that's abandonment. Auto-close. 2 hours later, they freak out via Slack: 'Paychecks are late!' I ignore Slack; not official. They submit a new ticket, perfect this time. I 'investigate' and find they just needed to clear cache. Tell them that, close it. My average resolution time? Under a minute. Their payroll? Delayed by a day. But hey, metrics don't lie, and neither do I (usually).

This is exactly the direction accounting is heading – building custom Claude skills instead of doing everything manually. As a CPA focused on AI for accountants, I do something very similar: one main prompt acts as the “macro procedure” for audit workpapers or tax prep, with sub-agents handling data extraction, categorization, risk flagging, and client-ready summaries. Saved me dozens of hours already this year. Curious – what was the biggest time-saver in your Claude skill? The MCP connectors or the subskills? Reply “PDF” if you want my list of 50 practical prompts that work on real client files (audit, tax & quarterly close).

I save 6 hours of accounting every quarter thanks to this Claude skill I HATE accounting. Every quarter I had to spend 8 hours on accounting. A full day! Last year I started building this Claude skill and for the first time this quarter, it did most of the work! It only took me 2 hours to get the job done! This is how I built the skill: 1. At the center is the macro procedure. How do I get my quarterly accounting done. It lists out all the steps. 2. Subskills take care of downloading, renaming, and uploading invoices. 3. I have MCP connectors to my Google Drive and Gmail. These allow me to fetch most invoices and manipulate my accounting folder In the first quarter, I "wasted" 8 hours to write down the skeleton. Accounting took double as long. In the second quarter, I saved ~2 hours and invested 6 more on building the skill. In the third quarter, I saved ~4 hours and invested 4 more hours. After one year, I have managed to save 6 hours and did not need to develop the skill any further. We just went along and I said "ah and update your skill with X"

Spot on – multi-agent setups with Claude Opus 4.7 are game changers. As a CPA, I use a similar 4-agent workflow for audit workpapers: one agent extracts data, second categorizes transactions, third flags risks, fourth drafts the summary. Cuts review time from hours to minutes. CPAs – who else is already running agent swarms for tax or audit work? Reply “PDF” for my 50 practical prompts that actually work in real client files.

I BUILT REAL LIFE JARVIS WITH CLAUDE OPUS 4.7. Me: "Launch 4 Claude Code agents and 4 Codex agents." Jarvis: "Done." Me: "Give each one a different API security audit." Jarvis: "Sent all eight." Me: "Review the output and have them fix what they found." Jarvis: "Follow-ups sent." Ships tomorrow on Product Hunt.

I put together 50 prompts like these in a free PDF — organized by task (categorization, emails, tax research, reporting, standards). Reply "PDF" and I'll DM you the link. Tag a colleague who's still doing it all manually.

bottleneck was never the model tbh. it's adoption. workflow. actually starting. the accountants winning rn didn't wait for 4.7. started with Sonnet a year ago and compounded from there.

Unpopular opinion: Claude Opus 4.7 doesn't matter at all for 59% of accountants. Not because it's bad. It's incredible. Because they haven't used the model that came out a year ago.