Finance — Lindsay Sailor, MBA

Republished with permission from the FrUn SMB AI Guide. Canonical chapter: Superhuman Docs → · All chapters

Lindsay Sailor, MBA · Fractional CFO, Sailor Financial · LinkedIn

1. Mindset — What an SMB Should Believe Before Spending a Dollar on AI

The belief that costs you money

Here’s the belief that costs SMBs the most money: that AI means instant accuracy. It doesn’t. AI is fast and it is confident, and when you first start using it, speed plus confidence inspires the feeling that it must be right. But feeling right and being right are not the same thing.

The businesses that get burned take an output and move on without checking it past a gut feel. In accounting, the money rarely leaks out in a dramatic blow-up. It leaks out in a hundred small coding errors nobody caught because the machine sounded sure of itself. The fix is simple to say and harder to do: treat it like what it is, a fast, capable junior you would never let sign off on the books alone.

Where it helps vs. where it just feels productive

Frankly, it helps across most of accounting: categorization, reporting, forecasting, reconciliation, making business plans, and answering questions. Just not as well as the marketing wants you to believe. The work that feels productive is the work where you skip the review step, because the output arrives clean and formatted and you get that little hit of done. The work that actually pays off is the boring version where the tool does the first pass and you check it. Same tool. Different habits. The habit is the whole game.

What has to be true first

Clean data and a consistent process. AI reads your data, so if your data is a mess, you get a confident mess back in return. A consistent process gives the tool context: where things go, why they go there, what a normal month looks like. Without both, you’re just automating chaos faster. That’s not a win.

2. The SMB Tool Stack — Start Cheap, Stay Simple

Start with these three

Your accounting system (QuickBooks Online), ~$40 to $275/mo. Built-in bank feed coding, suggested reconciliations that flag errors, expense analysis, budgeting off your history, and automated reminders. Is it AI the way the breathless LinkedIn posts mean it? Maybe not. But it saves time and gives you consistency, and that translates to real value.

A general AI assistant (Claude, ChatGPT, Gemini, Copilot), $0 to As much as you want to spend/mo. Augments your judgment, not your books. Pull data out of QuickBooks and hand it over for a sober second thought: does my chart of accounts fit my industry, do my margins make sense for my size, where are the weak points, what are some options to improve my accounts receivable process. It’s a great place to ask the questions you’re a little embarrassed to ask a pro before you go pay one. Think of it as a competent junior controller. Useful, fast, and it will occasionally tell you something wrong with total confidence.

An AP capture tool (Dext, Bill, or Hubdoc), ~$25 to $100/mo. Reads the vendor, date, amount, and tax right off your receipts and invoices, then creates the entry in QuickBooks for the bank feed to match. Replaces manual data entry and the receipt shoebox. I’m a Dext fan, but you really can’t go wrong with any of the three.

Skip or postpone

Dedicated FP&A tools. Overkill unless you’re rolling up multiple entities or currencies. And if you genuinely need them, you also need a pro to set them up, so it’s not a DIY starting point anyway.

AI bookkeeping replacements. Fine for a genuinely simple business. I mean only one business line, no split vendors, no complex inventory, no multicurrency, no sales tax sort of simple business. They get shaky the moment you have client recharges, foreign sales tax, or vendors that don’t always code to the same place.

Heavy custom accounts payable/accounts receivable automation. Basic receipt capture, absolutely. But once you start to need department routing and multi-step approvals, get professional help. Build it wrong and you’ll find out two months later the whole thing has to be redone.

Zero budget

Start with what’s already inside QuickBooks: bank feed coding, suggested reconciliations, dynamic budgeting, automated reports. Then add a free LLM tier to analyze the reports you export and answer questions about your numbers. Costs you nothing beyond the subscription you’re already paying for, and it’ll take you further than you’d expect.

3. The First 30 Days — Where to Start

Highest-ROI workflow: bank rules

Set up vendor and category rules for the transactions you see most. QuickBooks will suggest where things go, but if you know your Facebook ads always hit digital advertising, make the rule. Set it to auto-add and QuickBooks codes and books it without you in the system at all. Your bookkeeping starts doing itself. This is maybe the least sexy thing in this chapter, but has the biggest payoff. (Most of the real wins are like that.) And if you don’t have accounting software yet, get some. Please don’t run a real business out of Excel. You already know that’s a bad idea.

Week by week

Real talk before you start - I wanted to give you a lot of ideas for implementation, but doing all four of these weeks across your whole accounting function in 30 days, on your own, isn’t realistic. So don’t. Pick one module (AP, accounts receivable, the bank feed, reporting) and run these four weeks on that one thing. Next month, pick the next module and run it again. Same playbook, one piece at a time. If you’re looking for guidance on which one to start with - it’s the bank feed.

Week 1, audit. Get the lay of the land for the module you picked.

Map how it (accounts payable/accounts receivable/Payroll/etc.) actually works: how data gets in, who touches it, and where it slows down.

Don’t like writing? Get a voice to text app and rant the process outloud. It’ll save you time and energy.

Hand it to Claude and then ask for it to turn it into a clean, usable SOP.

Ask Claude how it would set this up or improve it inside QuickBooks (or your tech stack)

Ask what other software is worth a look for this module, so you know your options before you commit to anything.

Ask how it would implement them.

Week 2, clean and investigate. Nothing downstream works if the data underneath is off, so look hard before you build.

Export your chart of accounts and your transaction list, give them to Claude

Ask whether the COA actually fits your business and your industry.

Ask it to flag coding issues and inconsistencies it can see.

If you deal with sales tax, ask specifically about likely coding problems to review.

Then go into QuickBooks and take inventory: what bank feed rules already exist, which accounts receivable (customer) workflows are turned on, same for bills and AP, and what automated reports are already running.

Note: If there are any changes you need to make that you’re not sure how to do, ask Claude or QBO support. If you’re not comfortable doing this work, stop and reach out for support. It’s cheaper to pay someone to fix the problem at the start rather than wait and add another year’s worth of data on top of it.

Week 3, implement. Now you make QuickBooks do the work.

On the bank feed, sort by description to group charges from the same place, then pick out the high-activity vendors (the ones like advertising that bill you daily). Where you can, set rules to code and auto-post so the entry happens without you.

Turn on your accounts receivable and accounts payable reports to email to you weekly, so you always know what’s outstanding.

Set up your accounts receivable workflow end to end: invoice entry, approval, emailing the customer, and payment instructions.

If your transaction or sales volume is high, this is where add-on software earns its keep: A2X, Dext, or Synder for ecommerce entries, and Dext, Bill, or Hubdoc for bills. Worth investigating, but you don’t need them to get value out of this week.

Week 4, review and analyze. Close the loop and actually read your numbers.

Pull your P&L and balance sheet.

Look at QuickBooks’ built-in AI reporting and judge for yourself how useful it is.

Give your statements to Claude and ask it to analyze them.

Then ask Claude to build you a simple financial analysis tool aimed at whatever you actually worry about (margin, cash, a specific cost line), so next month you’re reviewing, not rebuilding.

Note: Not sure what you should be worrying about? This is when you should call in professional support to help guide you.

What good looks like at day 30

You won’t have overhauled your accounting in a month, and you shouldn’t expect to. For the one module you worked, here’s what good looks like:

The process is documented and lives somewhere you can actually find it, not just in your head.

One part of that module runs with less manual touch than it did (rules auto-posting, invoices going out on a workflow, reports landing in your inbox).

You know whether your coding in that area is clean, or you know exactly where the problems still are.

You can point to one specific thing that got faster, or a chunk of time you got back.

You trust it enough to leave it running and move to the next module next month.

Once you’ve rolled out all of the modules, you should be able to tick these off:

Your month-end close is faster than it was.

Your data makes more sense to you.

Whoever does the accounting is less stressed.

The numbers are useful for decisions, not just a chore you do after the fact.

4. A Real SMB Use Case

A 16-person specialty food brand in Ontario, around $3.5M in revenue, selling wholesale to retailers and direct through a Shopify store.

What was broken: Month-end close took close to three weeks, and the owner didn’t trust the margins at the end of it. Shopify payouts landed in the bank as lump sums with no split between sales, discounts, refunds, and fees, so the revenue picture was mush. Receipts lived in glove boxes, inboxes, and one very brave shoebox.

What AI and the tools did: Claude turned the process brain dump into real documentation and helped clean up the chart of accounts. A2X read the Shopify data and pushed clean journal entries into QuickBooks by period. Dext handled receipt and AP capture. Bank rules took over the recurring transactions. Then twelve months of clean data went back to Claude for trends and a first-draft budget, interview-style.

What the owner did: Approved bills, made the actual payment decisions, and brought in a fractional controller to review the setup so the AI wasn’t quietly hallucinating into the financials.

Tools and cost: QuickBooks, A2X, Dext, Claude. Roughly $250/mo in software, plus about $4,000 for the controller engagement to set it up and sanity-check it. About a month of the owner’s attention, not a frantic weekend.

Result: Close dropped from about three weeks to about five days. The owner got back the better part of a week every month and stopped paying for the cleanup-and-catch-up that used to eat the bookkeeper’s hours, so the whole setup paid for itself inside a quarter. Margins by channel became legible for the first time, and the owner moved from data entry to review, which is exactly where an owner’s time should go.

The part I want you to take from this isn’t the specific stack. It’s the pattern: clean the data, let the tools talk to each other, keep a human on the decisions, and bring in a pro when the implementation feels outside of your wheelhouse.

5. Guardrails — What an SMB Should NOT Do

The mistakes I see most

Three show up on repeat:

Trusting a confident output without verifying it. It will be wrong sometimes, and it won’t warn you first. The issue is that you won’t know when that will happen so you do need to build controls into all of your workflows.

Expecting it to know context about your business that nobody ever told it. It can’t read your mind, only your data. If you want to build a plan or a budget, ask AI to interview you to get the best results.

Cutting out the professionals entirely because the tool felt good enough. It isn’t, not yet. You should bring in a professional when:

You’re implementing a new system that you don’t feel comfortable setting up on your own.

When you’re sacrificing spending time on other more productive areas of your business to handle the books.

If you’re looking at your results, but you don’t know what you should be looking for.

The deepest one

Your books are only as good as the foundation under them. And usually that foundation was built by the same person now trying to build on top of it, which makes the mistakes invisible. If you knew what you’d done wrong, you wouldn’t have done it. This is where a professional earns their fee fast.

Just know which professional you need. A year-end tax accountant and an operational one are different jobs. The tax accountant handles filing and compliance once a year. An operational accountant (a Bookkeeper, Fractional Controller or Fractional CFO) keeps the day-to-day clean and useful all year and provides insight into your operations.

Keep humans on these, always

Year-end tax and compliance. It needs a real accountant’s review and sign-off. AI has hallucinated information into public audits and big firms have been fined hard for it. It’s not allowed, for good reason.

Moving money. AI can review your payables and help you decide what to pay this week against terms and cash on hand. It should not make the payment. Even if you can connect it to your bank, don’t. It could fat-finger a hundred thousand dollars out the door, and that’s still your fault.

Complex AP coding where the why changes. Same vendor, different departments, for reasons that shift. Genuinely hard to automate without very good intake points, and even then a human is still doing real work.

Privacy and data

Be careful what you feed AI. There have been cases of people loading in specific company data and competitors later querying their way to pieces of it. Turn training off (in Claude you have to do it yourself on a personal account; it’s off by default on enterprise) and strip out identifying information anyway. Think about data residency and what every connected tool actually touches. And mind the audit trail: AI won’t tell you step by step what it changed unless you instruct it up front to log its work.

6. Lessons Learned + One Thing You’d Tell Every SMB Owner

Let AI check AI. QuickBooks codes and reports, then I give Claude access to ask whether it makes sense and tracks with the industry. Two passes catch what one misses.

Confidence is not accuracy. It will hallucinate if you’re not careful. Good partner, terrible final word on your plans, your business, or your numbers.

The unsexy wins compound. Bank rules. An accounts receivable-to-approval-to-invoice-to-payment loop where the only thing you touched was who to invoice and the amount. Automated tools like A2X reading your Shopify data and turning it into clean QBO entries. None of it is particularly glamorous, but it adds up fast to tangible time savings.

Don’t custom-build a finance app for your business. You can, technically. But without knowing what a good end result looks like, you’ll build something that feels right and quietly isn’t, and you’ll find out at the worst possible time.

The one thing

AI is a tool, not a solution. It makes you more efficient. It is the best junior hire you’ll ever have: fast, tireless, and never the one who signs off. It is not a department you can replace. You still need your year-end tax accountant, and an operational accountant as you grow. Either a bookkeeper, a Fractional Controller or Fractional CFO depending on your size and needs. The only thing that will remove you entirely from your accounting is hiring someone to replace you in it.

Part of the FrUn SMB AI Guide. Read on canonical guide →