A curriculum for bookkeeping firms, e-residency service providers and fintech ops teams: documents classify themselves, client emails draft themselves, and your accountants review instead of retype.
The foundation day happens on-site in Tallinn, before any industry-specific work: we audit how each team works, map the right Claude tool to each person, then set it up — so everyone leaves with a setup matched to their role.
The team splits by function — bookkeeping/ops first, then onboarding/KYC, then reporting. Each track configures a Claude Project with their firm's templates, chart of accounts, and standard email language. Labs run on your real (anonymised where needed) client files so the output can be judged against actual quality standards.
Invoices, receipts and statements parsed into structured fields and written to the client ledger sheet. AI extracts and flags exceptions — the accountant reviews and posts. AI never posts entries directly.
Claude Project loaded with the firm's tone and client context: chase emails, clarification requests and onboarding replies drafted in Estonian, English or Russian — reviewed before sending.
Client call notes or recordings pasted in → structured summary, open items list, and draft follow-up email. The accountant sends; Claude assembles.
For e-residency and international client firms: intake forms summarised, document completeness checked, compliance file assembled for human sign-off — AML rules stay with the accountant.
AI assists; the accountant signs off. Claude can summarise ledger exceptions, flag reconciliation gaps, and draft the monthly report narrative — but it never posts entries, never certifies figures, and never produces a statutory output without a qualified accountant reviewing the numbers against source records. This rule is installed as firm policy in Week 2 and documented in the AI use policy.
Why this matters: Claude does not have access to your accounting software in real time, cannot guarantee numeric accuracy, and can state incorrect figures with confidence. Every workflow includes a human verification step for any number that will appear in a client-facing or regulatory document.
In the labs, each function builds reusable Claude Skills and Cowork workflows on its own client files. No coding required — a Skill is a plain-language file that captures how your firm wants the job done, and Cowork runs the multi-step version on your files:
Each accountant runs one real client month-end (or one full onboarding) through the new setup during the normal work week, logging minutes per file before and after. The review session tunes the extraction prompts and exception rules to your chart of accounts.
Each function demos what it built. Named owner per workflow, one metric each — minutes per client file, documents processed per day, chase emails sent without human drafting. Handover pack: prompts, Claude Projects, Skills, Cowork workflows and templates, documented in-house.
One published customer result, and the two Claude capabilities the labs are built on:
"It helps us get more done with fewer headaches, and that's everything in today's 'do-more-with-less' world."
Campfire runs its accounting on Claude and reports a 90% cut in bank-reconciliation time, 50% less time on reporting, and monthly close three days faster — the same document-and-narrative work this curriculum builds.
Read the case study →Claude ships prebuilt Skills for Excel, Word and PDF work — and your team writes its own, a plain-language file capturing your chart of accounts, templates and house email language. Claude then does the job your firm's way, every time, with no coding.
Point Claude Cowork at a folder of client invoices, receipts and statements and it works through them end to end — reading, extracting and drafting the ledger update — while the accountant reviews and posts. Delegated work, not another chat window.
The Campfire figures are Anthropic's published customer results, not Soch results. Tallinn cohort case studies will follow once firms complete the programme.
One on-site day in Tallinn — the Week-1 foundation on its own. Firms of 1–50 people.
All four weeks of this curriculum — foundation day plus the accounting & fintech build labs, brief and demo day. Firms up to 100.
Flat, per-company pricing — no day rates, no scope creep. You get the exact figure on the scoping call, before you commit to anything.
Book a scoping call