Level 2 taught one person to do one accounting task well with AI, under supervision, with a record. Level 3 is about the operation: the same work run by a team across many clients, entities or returns, where the person accountable for the result cannot personally read every item. It is written for engagement and practice leaders, tax and audit managers, client accounting services managers, controllers running close teams, and firm operations and innovation staff. Every module builds a control an operation needs, with the arithmetic that shows whether it will hold at volume.
The first four modules build the operating frame. Module one explains why a task that is safe once becomes risky as a system: volume, handoffs, shared inputs, complacency under workload and confident wrong output travelling into ledgers, reports and filings, and why an operation needs a named owner and a measured baseline before any redesign. Module two maps a close, tax or engagement workflow as it actually runs, places each proposed AI use by the capability types in COSO's 2026 GenAI publication, marks handoffs, review points and the IT general controls beneath them, and recognises when a public company process change needs evaluation. Module three tiers AI-assisted steps by consequence and reversibility, raises tiers for reliance factors, attaches review depth, reviewer role, records and escalation to each tier, checks reviewer capacity and starts agentic uses low. Module four writes standard operating procedures a reviewing professional can enforce and tests them on someone who did not write them.
Modules five to seven govern the knowledge an AI tool draws on. Module five selects firm memos, policies and templates for a retrieval corpus, keeps authoritative sources distinct from firm material, tags every document by tax year, jurisdiction, entity type, audit framework and status, and records provenance, ownership and retirement. Module six covers how retrieval fails with confidence, through proposed or superseded rules, the wrong year, state, entity or framework, and poisoned or instruction-laden documents, and maps controls to the 2026 OWASP list. Module seven keeps one client's information out of another's work, limits access to duties for staff and AI service accounts, explains why deidentified compilations of tax return information remain tax return information, and classifies outputs, logs and transcripts at the level of their inputs.
Modules eight to eleven apply the frame to the work itself. Module eight sets out what the amended PCAOB standards say about technology-assisted analysis on issuer audits, keeps AICPA SAS No. 142 separate for non-issuer audits, and covers flagged items, the untested remainder, the reliability of company-provided electronic information and validating what the tool did. Module nine runs journal entry review as a queue of flags with reasons and evidence-backed dispositions, designed against omission errors, commission errors and algorithm aversion. Module ten runs return preparation across a season: intake gates, review layers matched to complexity, pre-filing tie-out to the tax engine, the signing preparer's unchanged responsibility, and tax data security under the FTC Safeguards Rule and section 7216. Module eleven runs close and reconciliation under management's responsibility, with statuses that cannot overstate, IT general control evidence each period, verification sampling and findings reported to certifying officers with confidence and limits.
Modules twelve to fourteen keep the operation honest over time. Module twelve builds a watchlist of the official sources that change accounting and tax work, tells a proposal from a final rule and a repealed law from a current one, applies a trigger test and logs what was checked, where and when. Module thirteen samples AI-assisted output independently, codes every error by type, severity and detection point, sets thresholds, reads trends and feeds findings into procedures and the firm's quality management evaluation. Module fourteen designs escalation paths that hold in deadline week, stop conditions agreed before the work starts, measures that would reveal failure, the ways measures get gamed, and a return-on-AI record in net hours.
Statements of authority are labelled throughout as law or rule, professional standard, professional guidance, best practice, emerging practice, or an AI Coalition Network recommendation, with whom each binds. PCAOB and AICPA requirements are never mixed, proposals are labelled as proposals with the date checked, and no state's rule, professional body's code or voluntary framework is presented as binding on everyone. The level ships with a printable workbook and templates: an accounting workflow map, a risk-tier matrix, a standard operating procedure template, a knowledge corpus governance checklist, a technology-assisted testing protocol, a quality assurance sampling plan, a regulatory and standards watchlist and the workflow redesign template. The final examination draws forty scenario questions at random from a reviewed bank, and the capstone is a Workflow Redesign of one operation, handed over with everything its owner would need to run it.
Everything here is professional education. It is not tax, accounting, audit or legal advice, it does not replace the professional standards, SEC and IRS rules or state board requirements that govern accounting, tax and attest work, and learners must check the rules that apply to their own licence and practice. Completing the level earns an independent educational certificate issued by AI Coalition Network with a public verification page. It is not a CPA licence, carries no professional education hours, and does not satisfy any state board of accountancy requirement.