Each AI draft is checked against your books, and mismatches wait for you.
Already on Compass? 2.0 is in your account. Log in →
AI reasons well when the facts are laid out right. It slips when it has to pick the basis, the source or the formula, and the sentence around a wrong figure still reads perfectly. So as AI drafts more of the work, the hours move to checking it. Compass 2.0 is built for that check: one definition for every number, each client's rules held as settings, and every figure in a draft compared with the books before you see it.
The film follows a sample firm, Kestrel & Vale CPAs, and its clients.
Turn on skills and Compass keeps an eye on cash, margins, collections, spend and the books. Margin drift flags Harbor Street Kitchen: food cost climbed to 33%, up from 28% ($9,176 in August). The supplier's price notice is already on file in Knowledge. Each finding arrives with the rows behind it. Start with a pack like Get paid on time or Keep the books clean.
Every time new data lands, Compass compares Stripe, Square payout files and bank files that state a balance with the books. Square and the books are $3,358 apart, so the month-end summary waits. The books get fixed, the check passes, and the summary goes out. Every data check reads passed, failed or could not run, with the reason, and the Checks page keeps the history.
How Checks work →
Workflows draft letters and reports on your schedule, from a template or from a sentence you write. Before a draft reaches you, every figure in it is compared with the number Compass worked out. Northgate Dental's owner letter says revenue grew 0.7%, and the books say 10.3%. The AI had picked up the growth in operating costs. The letter is held with the figure flagged, and one click sends it with 10.3%. A figure Compass can't trace says Not traced, and a workflow set to send on its own falls back to approval.
Why we check AI's numbers without AI →
Connect your AI once. It reads every client's numbers, definitions and House rules, each answer carries the company's data checks for the period, and each read lands in Activity as Claude via MCP. Roles decide what your AI can reach, and anything destructive asks first.
34 standard numbers like runway, gross margin and days to get paid, each defined in plain words with its inputs and sources. Each one appears when your data supports it and says what's missing when it doesn't. Change a definition for one client with a reason on the record, or set a firm standard every client inherits. Computed with plain arithmetic, with no AI involved.
Each client's basis, year end, closed-through month, reporting currency, what's worth mentioning, which source wins, business model, who approves and reporting rhythm. Prefilled from the data and confirmed by you. Open months carry a Preliminary mark everywhere, including in Claude, and anything the AI had to assume is labeled Assumed.
Each client sits on one board with its owner, stage, closed-through month and how its data stands. One firm Inbox holds what needs a person across every client. Ask a question across all of them and get one answer that cites the rows it read. Write a condition once and apply it to the clients you pick, switch skills on for a whole segment, and compare clients side by side. Filing deadlines for VAT, payroll, annual accounts and corporate tax are tracked for clients in the US, UK, Croatia, Slovenia, Austria and Germany, and reach the Inbox before they're due.
$99 a month for the hub, plus $19 per client company.
Bringing a firm? Book a walkthrough →
We watched advisors paste spreadsheets into chat windows, then trace every figure that came back by hand. 2.0 takes the legwork and the arithmetic, and the judgment stays with you. Tell me what you'd want checked next. Nikola Jakić, founder
Every feature free for 14 days, then plans from $29 per company per month.
Bringing a firm? Book a walkthrough →
