--- title: "AI for Accounting: Why Your Accountant Isn't Using It Yet (And What To Do About That)" description: "Most SMBs think their accountants are already using AI. They're not. Here's what actually works, what doesn't, and the three workflows that move the needle for a 20-person finance team." canonical: https://richardbatt.com/blog/ai-for-accounting-smb-playbook date: 2026-04-27 author: Richard Batt tags: [AI Implementation, Finance Operations, SMB AI, Automation] type: blog_post --- # AI for Accounting: Why Your Accountant Isn't Using It Yet (And What To Do About That) _Most SMBs think their accountants are already using AI. They're not. Here's what actually works, what doesn't, and the three workflows that move the needle for a 20-person finance team._ **Richard Batt** — AI implementation specialist. 120+ projects across 15+ industries, serving SMBs (5-200 employees) worldwide from Middlesbrough, UK (working globally). Contact: richard@richardbatt.com · https://richardbatt.com Your accountant got a demo of some AI tool last month. She opened it, ran a test. It worked fine on the test data. Then she closed it and went back to the spreadsheet. This happens in 8 out of 10 finance teams I talk to. The reason isn't that AI doesn't work. It's that the tool didn't fit the workflow. Or it needed data in a format nobody had. Or the integration took six weeks to scope, and she had invoices due Friday. Most accounting AI tools are built for accountants who have time to set them up. They're not built for the CFO or the bookkeeper who has three hours on a Tuesday morning before the bank reconciliation needs to happen. Here's what I've seen work in practice across 15+ finance teams in the last nine months. ## The three workflows that actually move the needle Finance teams I've automated fell into one of three camps: invoice processing, expense categorization, or reconciliation prep. Not all three. Almost always one. ### Invoice processing (the quick win) Invoice processing is the easiest win and the one that sticks. You get emails with PDFs. The team extracts the vendor name, amount, due date, and cost code. That data goes into an invoice register or into accounting software. Right now that's manual. Someone opens the PDF, reads it, and types the data into the system. A 30-person company with 200 invoices a month does this 6-8 hours a week. AI can pull that data out of the PDF and drop it directly into your accounting software. Not perfectly. Maybe 95% accurate. But enough that the one person who checks it spends 20 minutes fixing the 5% instead of 6 hours extracting it. A logistics company I worked with did this in February. They handled 150 invoices a month from 40 different vendors. The extraction took their accountant about 4 hours a week. With the AI workflow, it takes 25 minutes. They redeployed those 3.5 hours to supplier relationship work and cost analysis. That's the stuff the supplier invoices are actually telling you if you have time to listen. The cost? About $400 a month for the tool, $1,200 for setup, $200 a month for someone to monitor quality. Payback on that $1,400 setup was three weeks. ### Expense categorization (the scaling problem) Expense categorization gets harder the bigger you grow. A 10-person company categorizes expenses as they come in. The CFO sees the credit card transaction, knows the project, codes it herself. Takes about 30 seconds per transaction. A 50-person company has 300 transactions a month, often from people who don't know the correct cost code. The finance person spends 2-3 hours a week on recoding. AI can learn your cost structure from historical data (your existing categorized transactions) and auto-code new ones. Not perfectly. But well enough that reviewing auto-coded transactions takes 10 seconds instead of 60. A tech agency I set this up for had 45 employees and 180 expenses a month. About half were miscoded. The AI model learned from their historical data and caught 89% of the categories on first try. The finance person now spends 15 minutes a week reviewing exceptions instead of 90 minutes a week doing the coding. This one is trickier because it needs training on your actual expense history and your actual cost structure. It requires a bit more setup than invoice processing. But the payback is real if you have the volume. ### Reconciliation prep (the hidden efficiency) Bank reconciliation is not fun. You get a bank statement. Your team matches it to the accounting software. The account never reconciles perfectly. Someone spends 4 hours tracing the differences. AI can do the matching part. It is pattern recognition, not judgment. AI looks at the transaction amount, the date, the description, and the memo, and it finds the corresponding entry in your general ledger. Most reconciliations in finance teams that don't use AI involve one person pulling the bank statement and manually clicking through transactions one by one. It takes 3-4 hours to reconcile a single account for a month. With the AI matching, the exceptions (transactions that don't match) are much smaller. The team still has to investigate them. But they're not scrolling through 400 transactions to find 12 that are actually unmatched. A 25-person professional services firm I worked with had one accountant who spent Tuesday morning on bank reconciliation every month. That Tuesday now takes 45 minutes instead of 3 hours. The remaining 2.5 hours went to analytical work: understanding cash flow, flagging anomalies, building a forecast. ## What doesn't work (and why your accountant is right) Before you buy a tool, here's what I've seen fail. **AI that needs clean data.** Most accounting AI tools are built on the assumption that your data is organised. Your invoices are scanned well. Your bank transactions have consistent formats. Your expense descriptions follow a standard. In reality, they don't. You've got hand-written vendor names, you've got payments to "ACME" and "ACME INC" and "Acme Supplies." AI fails quietly on garbage data. Your accountant knows this. She doesn't trust tools that need perfect inputs. Fix the data first, or pick a tool built for messy inputs. **AI that's slower than doing it manually.** Some tools require so much setup that the payback never happens. You need to upload the data in a specific format. You need to build a training set. You need to map your cost codes to the tool's framework. A month of setup for 20 minutes a week of savings is a bad trade. Your accountant is right to close the browser. Pick a tool that works with your existing data structure. **AI that doesn't integrate with your accounting software.** If the AI extracts data but you still have to copy it into your system manually, it's not actually AI. It's just a very slow data entry person. The real wins come when the AI output feeds directly into Xero, QuickBooks, Netsuite, or whatever you're running. That integration is worth more than the AI itself. **AI that doesn't let your accountant override it.** The best AI tools have a human-review loop. The AI makes a guess (95% confident on an invoice category), the human confirms or corrects, and the AI learns from the correction. The worst ones lock the AI output and require a support ticket to change it. Your accountant wants agency. Give it to her. ## How to actually implement this Pick one workflow. Not all three. Most teams pick invoice processing because the ROI is clearest and the setup is straightforward. Here's the sequence: 1. Measure the baseline. How many hours a week is your accountant spending on the one thing you're automating? For invoice processing, count the invoices and time the extraction on five of them. Do the math. "We process 120 invoices a month at 3 minutes each = 6 hours a week." You need that number. 2. Pick a tool that works with your accounting software. Don't pick the best AI. Pick the one that integrates with what you're already using. If you're on Xero, use a Xero-approved AI partner. If you're on QuickBooks, same thing. 3. Run a pilot. Process 50 invoices through the tool. Let your accountant review the output. Get feedback on accuracy and on workflow. Does the AI need a different invoice format? Does the tool's output match your data structure? 4. Build the integration. This is where people usually get stuck. The tool vendor will want to do a demo. Then they'll scope it. Then they'll send a proposal that assumes six weeks. Don't let them. Most accounting integrations can go live in one week. You usually need one file export from your accounting software and one integration rule (if vendor is X, create invoice in account Y). 5. Go live with one person. Don't roll it out to the whole team. Let one accountant use it for a week. She'll find the edge cases. She'll show you what needs to be fixed. Then you expand. 6. Monitor quality. The most common failure mode is the tool working fine for 80% of invoices and catastrophically failing on the other 20%. Make sure someone is checking the output, at least for the first month. ## The thing your accountant hasn't said out loud Your accountant isn't resisting AI. She's protecting you. Most of the AI tools designed for accounting are sold to accountants who don't understand their own cost of labour. A tool vendor says "this saves you 20 hours a month," and the accountant thinks "20 hours is not a full week, this is noise." But in a 25-person company, if your accountant saves 20 hours a month, that's either someone new you didn't have to hire, or 20 hours a month of analytical work that wasn't happening before. Your accountant wants to do analysis. She doesn't want to extract data from PDFs. But she's seen enough failed "automation" projects that she's skeptical. The burden of proof is on you and the tool vendor. That's actually reasonable. ## FAQ **Q: What if we use a virtual CFO or outsourced accounting service?** A: The same workflows apply. The difference is you're paying a service provider hourly, so the ROI is much tighter. If your virtual CFO is spending 4 hours a month on invoice data entry and you can cut it to 20 minutes, that's a $300+ a month saving on their fees. Start with them. **Q: Does this work for construction accounting or project accounting with cost codes?** A: Yes. Actually easier. Construction teams have clear cost structures and high invoice volume. The AI learns the cost code mapping and applies it consistently. The main thing: you need 3-4 months of historical invoices (100+) for the AI to learn from. If you're just starting, the ROI takes longer. **Q: What about AI doing the analysis, not just data entry?** A: That's the next frontier. But it's not here yet. Current AI can highlight that accounts receivable is 15% higher than last month. It can flag that one vendor's invoices jumped in price. It cannot read a board meeting discussion from six months ago and connect it to a change in supplier spend. That requires context AI doesn't have. Your accountant still does the analysis. **Q: Does the AI tool need to learn our specific data, or does it work out of the box?** A: Depends on the tool. Some (like invoice extraction) work out of the box. Some (like expense categorization) need training. The out-of-the-box tools are easier to implement but sometimes less accurate. Training-based tools are more accurate but take longer to set up. Pick based on your tolerance for setup time vs. accuracy. **Q: What if we're a one-person finance team?** A: Invoice processing still works if you get 50+ invoices a month. Reconciliation prep is less valuable (you're probably spending 4 hours a month, not 4 hours a week). Expense categorization doesn't apply unless you're processing 100+ expenses a month. Be honest about your volume before you buy a tool. **Q: How much does this actually cost?** A: Depends on the tool and the volume. Invoice processing: $300-800 a month for the tool, $1-2K for setup, maybe $200-400 a month ongoing support. Bank reconciliation: usually built into accounting software (Xero and QuickBooks both include it), $0-200 a month. Expense categorization: $200-600 a month depending on volume. Total cost is usually under $1,200 a month, and almost always pays back within two months on a single workflow. --- **The real move:** Pick invoice processing. Measure your current time spend. Run a pilot. Get your accountant to sign off on accuracy. Go live. Most teams are live and running in three weeks. The businesses that did this in February are already trialling expense categorization. That's the pattern you want: one win, then another. --- ## More about Richard Batt Richard Batt is an AI implementation specialist who helps businesses deploy working AI automation in days, not months. 120+ projects across 15+ industries. ### Products and pricing - **[The AI Ops Vault](https://richardbatt.com/vault) — $97/month:** 100+ AI prompts, 50+ automation blueprints, 20+ workflow templates from real projects. Copy, paste, deploy. - **[AI Revenue Roadmap](https://richardbatt.com/roadmap) — $2,500:** Fixed-price operational audit. Top 5 AI opportunities ranked by ROI, implementation roadmap, $50K+ savings guarantee. - **[AI Ops Accelerator](https://richardbatt.com/accelerator) — $4,000-$6,000:** 6-week done-with-you sprint. 3-5 working automations deployed in your business. ### Key pages - [Home](https://richardbatt.com/) - [About Richard](https://richardbatt.com/about) - [The AI Ops Vault](https://richardbatt.com/vault) - [AI Revenue Roadmap](https://richardbatt.com/roadmap) - [AI Statistics (2026)](https://richardbatt.com/ai-statistics) - [Blog](https://richardbatt.com/blog) - [Contact](https://richardbatt.com/contact) ### Contact - Email: richard@richardbatt.com - Location: Middlesbrough, UK (working globally) - Website: https://richardbatt.com