You already have the data to answer the most important question in your business. It sits in your accounting file, your bank feed, and a spreadsheet you stopped trusting somewhere around March. And yet when you need to decide whether you can afford the next hire, survive a late-paying client, or take on a loan, the honest answer is usually a shrug: your books record what happened, but they never tell you what it means. A startup that launched in June 2026 just raised $1.5 million on the bet that this gap — between having numbers and understanding them — is the most valuable problem in small-business finance. Whether or not you ever use its product, the launch tells you where every financial tool you touch is headed, and what to demand from yours.
This post covers what the new platform actually does, why plain-English financial questions are becoming the standard interface, the cost math that makes this shift inevitable, the first questions worth asking any tool like it, and where you still need your own judgment — because an answer that sounds confident is not the same as one that is right.
What launched: CFO-grade answers without the CFO invoice
CentSight, an AI-native financial intelligence platform based in Lakewood, Ohio, announced its public launch on June 24, 2026, backed by a $1.5 million pre-seed investment from Mudita Venture Partners and built with the firm's company-creation studio. The founder, longtime operator Gerald Hetrick, previously built a staffing software company that was acquired in 2023. The pitch is aimed squarely at owners who feel the gap described above: the company says too many good businesses "run blind," with plenty of data and no real direction.
The product connects to the tools a small business already uses — QuickBooks Online for the accounting record and bank accounts through Plaid for the cash record — and lets the owner ask questions about cash flow, expenses, revenue, profitability, runway, and margins in plain English instead of building reports. Its answers arrive with supporting numbers attached, so a claim about next month's cash position comes with the figures behind it rather than standing alone as a pronouncement.
The more interesting half of the launch is a proactive alert system called Signals. Instead of waiting for the owner to ask, it watches the connected data continuously and flags issues early: a cash gap forming in the next few weeks, a customer falling behind on payment, a vendor category suddenly running hot, a margin narrowing, or spending drifting outside its normal pattern. Each alert is framed as a specific business prompt — the next question to ask, while there is still time to act — rather than a generic dashboard badge. Early access is open now through the company's site.
Strip away the launch language and the product is two ideas: questions in, answers out; and attention directed before the owner knows to ask. Both ideas are bigger than one company, which is why the launch matters even if you never become a customer.
The assistant is becoming the front door to your books
CentSight is not an outlier; it is the latest entry in the clearest product trend in small-business accounting. An October 2026 roundup of the AI finance stack put it bluntly: the assistant is becoming the front door. Ledger choice still matters — QuickBooks or Xero for most small businesses — but the interface owners touch first is increasingly a chat box that answers "how much did we spend on software last quarter?" or "how does this month compare to last year?" in a sentence rather than a report grid.
The examples multiplied through 2026. An AI-native bookkeeping platform called Tabby shipped a chat assistant in October that answers plain-language finance questions and completes bookkeeping tasks. Other tools now work inside the general assistants owners already use, answering money questions conversationally against imported bank transactions. QuickBooks itself layered AI assistance over the ledger most American small businesses already keep. The pattern is consistent: the bookkeeping record stays where it is, and a conversational layer translates it into answers.
Adoption data explains why vendors are racing here. Research published by Xero in August 2026 found 62% of small businesses open to using AI, and businesses using AI daily were twice as likely to report revenue growth as those that never touch it. But the same research found a readiness gap — over half of owners said AI disappearing would change nothing for them — and named data privacy the top concern, cited by 38%. Owners want the answers; they do not yet trust the machinery. Every launch in this space is really a bid to close that trust gap first.
The math that makes this inevitable: the CFO you cannot afford
There are more than 36 million small businesses in the United States, and strategic financial guidance has always been priced for companies far larger. Industry estimates cited at the launch put fractional CFO advisory services at $3,000 to more than $18,000 per month. Independent 2026 pricing guides land in the same range: most small businesses in the $500K-to-$10M revenue band pay $4,000 to $8,000 a month on retainer, with hourly rates of $150 to $500, while businesses under $1 million in revenue can sometimes find lighter engagements near $1,500 to $3,000. A full-time CFO, meanwhile, carried a median salary above $430,000 by mid-2026 — more than many small firms' entire payroll.
Set that price against the stakes. A widely cited U.S. Bank study, popularized by SCORE, attributes 82% of business failures in whole or in part to cash flow problems — not to a lack of customers, but to mismanaged timing between money in and money out. The Federal Reserve's 2026 survey work found three in five small firms applied for financing in the prior year, and an October 2026 survey of more than a thousand owners found 78% had been hit by at least one unexpected cash crunch in twelve months. The businesses most exposed to a cash surprise are exactly the ones priced out of the professional who would have seen it coming.
That mismatch is the whole commercial logic of the AI finance assistant: the guidance gap is enormous, the human version is expensive, and the questions owners need answered — can I hire, can I spend, will I make payroll in six weeks — are answerable from data they already own. A tool that reads the books they have and speaks the language they speak does not need to be as good as a CFO to be enormously useful. It needs to be better than the shrug.
The five questions to ask first
If you get access to a tool like this — CentSight's or any competitor's — resist the urge to start with vanity metrics. The questions below are ordered the way a good CFO would order them: survival first, then performance, then growth. Try each one, and judge the tool by whether its answer names specific numbers you can verify.
1. "Will I run out of cash in the next 90 days?"
This is the only question that can kill you, so ask it first. A useful answer shows your current cash position, expected inflows dated by when customers actually pay (not when you invoice), and committed outflows including payroll, rent, loan payments, and tax deposits — then points at the lowest week. If the tool instead gives you a profit figure, it has answered a different question: profitable businesses run out of cash constantly, usually because a big receivable slips two weeks. Press for the dated forecast, and check its assumptions about collection timing against your own experience.
2. "Which customers are paying me late, and how much is stuck?"
A cash gap almost always has a name on it. Ask who owes you money past terms, how the overdue balance changed over the last three months, and which single account represents the largest stuck balance. Then act on the answer the same day: the invoice that is 15 days late is a reminder; the one that is 75 days late is a negotiation. An alert system earns its keep here by noticing the drift at day 20, when a nudge still works.
3. "Where did my margin go?"
Revenue up and bank balance flat is the classic small-business mystery, and the answer is usually a cost category that crept while nobody watched. Ask for gross margin by month for the last six months, then ask which expense categories grew faster than revenue over the same period. Common culprits: a software stack that added seats nobody removed, a supplier that raised prices twice, or labor hours that expanded to fill a fixed-fee project. You want the tool to name the category and the dollar change, not to observe that "expenses increased."
4. "What is my real monthly burn, and how many months of runway do I have?"
Founders ask this constantly and calculate it inconsistently. Monthly burn is not total expenses — it is the net cash your operations consume in a typical month after collections. Runway is current liquid cash divided by that burn. Ask for both, then ask what changes the answer fastest: cutting one subscription rarely moves it, while collecting receivables ten days sooner often moves it a lot. If the tool cannot separate operating cash from loan proceeds or owner draws, its runway number is fiction; those inflows are not earnings and must not flatter the math.
5. "What changed this month that deserves my attention?"
This is the open-ended prompt that tests whether the tool thinks like an advisor or a parrot. A good answer picks two or three genuine movements — a vendor category up 40%, a revenue concentration shifting onto one client, payroll crossing a threshold as a share of revenue — and skips the noise. A bad answer lists every variance, which is just the report you were already ignoring, rendered as sentences. Judge ruthlessly: attention is the scarce resource, and a tool that spends it badly is worse than silence.
Where the AI still needs you: verification, privacy, and judgment
Enthusiasm first, then the brakes — because every one of these tools has failure modes you must manage, and the launch announcements never list them.
Verify every material number against the source record. Generative AI can produce figures, vendor names, and dates that read correctly and exist nowhere in your books — a hallucination dressed as a calculation. One finance-focused study popularized in 2026 put hallucination rates on finance-related AI queries as high as 41%. The rule practitioners recommend is simple: treat every AI-generated number as provisional until a human traces at least a sample back to the ledger or bank statement. When the tool says your runway is five months, open the cash account and confirm the starting balance yourself. When it says a vendor category spiked, look at the actual transactions. The tools that attach supporting numbers to their answers — as CentSight promises to — make this check fast; tools that answer in bare prose make it impossible, which is itself a reason to prefer the former.
Mind what you connect and what it trains on. The 38% of owners worried about AI data privacy have a point: connecting your accounting file and bank feed hands a young company your complete financial life. Before you link anything, read how the vendor stores credentials (read-only connections through an aggregator like Plaid are the standard for a reason), whether your data trains shared models or stays siloed to your account, how to revoke access and delete data, and what happens to your history if you cancel. The U.S. Government Accountability Office's standing cautions about AI in finance — data quality issues, privacy concerns, new cybersecurity exposure — apply to a five-person firm exactly as they do to a bank.
Keep the judgment calls for yourself. An assistant can tell you the margin narrowed; it cannot tell you whether to fire the client whose work causes it, because that decision involves the relationship, the pipeline, and your risk tolerance — none of which are in the ledger. Use the tool to shrink the time between "something moved" and "I understand it," then make the call the way you always have: with context no model holds. Owners who treat AI output as automatically correct, finance teams warn, end up carrying last period's error forward at machine speed.
Keep your books answer-ready: the boring work that makes the magic work
Here is the part no launch announcement emphasizes: every one of these assistants is only as good as the record it reads. Ask a sharp question against a messy ledger and you get a confident wrong answer faster than before. The owners who benefit most from the AI finance wave will be the ones whose books are clean, current, and reconciled — which has always been the job, and still is.
Three habits matter most. First, reconcile every account monthly so the assistant's "cash position" matches the bank's, not last quarter's guess. Second, keep a consistent chart of accounts — one software category, spelled one way — so "spending on software" is a real total rather than five fragments the model must guess at merging. Third, separate operating flows from financing flows: loan proceeds are not revenue, and owner draws are not expenses, and any tool that mixes them will misstate your burn.
If you keep a plain-text ledger, this discipline is mechanical. Recording October's activity with distinct accounts for each cost category is what lets any downstream tool — including the dashboards in Fava and the reports described in the Beancount documentation — split burn from noise:
2026-10-01 open Assets:Checking:Operating USD
2026-10-01 open Income:Services:Retainers USD
2026-10-01 open Expenses:Payroll USD
2026-10-01 open Expenses:Software USD
2026-10-01 open Expenses:Rent USD
2026-10-01 open Equity:Opening-Balances USD
2026-10-01 * "Opening cash position"
Assets:Checking:Operating 48000.00 USD
Equity:Opening-Balances
2026-10-15 * "October client retainers collected"
Assets:Checking:Operating 22000.00 USD
Income:Services:Retainers
2026-10-31 * "October payroll run"
Expenses:Payroll 14500.00 USD
Assets:Checking:Operating
2026-10-31 * "October software and rent"
Expenses:Software 640.00 USD
Expenses:Rent 2100.00 USD
Assets:Checking:Operating -2740.00 USDOctober ends with $52,760 in operating cash — $48,000 plus $22,000 collected, minus $14,500 of payroll and $2,740 of software and rent. Monthly operating burn, excluding the timing of any single collection, is roughly $17,240, so that balance represents about three months of runway before new revenue. An assistant reading this ledger can answer the runway question in one sentence; reading a ledger where software, payroll, and the owner's draw share one account, it can only guess. Structure in, answers out.
Your next step: audition one assistant this month
The CentSight launch is a signal, not a verdict. Plain-English questions, proactive alerts, and CFO-style summaries are becoming standard equipment, and within a year you will likely have several credible options — including, quite possibly, one built into the accounting software you already pay for. The winning move is not to adopt today's newcomer sight unseen; it is to audition the category with clear eyes.
Pick one tool this month — your current vendor's AI feature if it has one, or an early-access newcomer — and run it through the five questions above against your real books. Score it on three things: does every answer carry checkable numbers, does it surface at least one issue you had not noticed, and does its data handling survive your reading of the privacy policy? Keep the tool only if it passes all three. And whatever you decide, do the unglamorous part this week: reconcile October, tidy the account names, and separate the operating flows from the financing ones. The assistants are coming either way. The owners with answer-ready books will get answers; everyone else will get sentences.
