Puzzle brings AI work into the accounting ledger

Puzzle combines accounting software with AI tools that work on the same ledger and source transactions. Chat answers financial questions, Cowork prepares tasks and Close organises a reviewed month-end process. The current offer also includes paid accounting services through Puzzle’s service arrangements and partner network. The central choice is how much work your team will review itself and how much it will assign to a bookkeeper.

01 / ProductThe AI tools work inside the accounting system

The Puzzle overview presents two routes: software for teams operating their own books, or software with accounting services layered on top. Its ledger supports cash and accrual views, with connected banking, spending, payroll and payment data. The AI functions are part of that accounting environment rather than a separate document-chat application.

The AI Suite page distinguishes Chat, which answers questions from the books, from Cowork, which prepares accounting tasks for review. The useful promise is traceability from an explanation to the transactions behind it. That does not establish that every categorisation or interpretation is correct; the quality still depends on complete records and appropriate accounting judgement.

A financial answer has several layers. The imported transaction must be accurate, the accounting treatment must be appropriate and the explanation must preserve the meaning of the resulting report. A fluent summary can be wrong at any of those layers. An evaluation should let the reviewer move back through them rather than only judging the prose.

02 / AudienceChoose who owns the books before choosing the AI interface

Puzzle can suit a founder who wants clearer day-to-day visibility while an accountant handles judgement and close review. It can also suit a firm standardising repeatable work across clients. Those are different operating models: a founder may need explanations, while a firm needs task ownership, review evidence and consistent treatment across periods.

Compare Numeric when the main requirement is organising accounting close and review around an existing finance stack, and Rillet when evaluating a broader AI-oriented accounting system. The key question is whether the ledger itself is changing or whether the team mainly needs a better process around an established system.

Before migration, establish which entity, accounting basis and historical periods the pilot covers. Agree who signs off opening balances and how discrepancies will be resolved. An AI-assisted workflow can accelerate preparation, but a new interface does not make a mismatched opening balance disappear. Keep that responsibility explicit.

03 / WorkflowProposed workflow: review one month with known exceptions

Start with an accountant-approved demonstration dataset or a carefully scoped test entity. Include ordinary expenses, a recurring subscription, a duplicate-looking payment and an unmatched receipt. Record the expected treatment before asking AI to prepare work. This makes it possible to distinguish genuine help from a plausible answer that merely agrees with the prompt.

The integration catalogue lists connections across banking, payroll, payments and spend management. Confirm the exact accounts and data types the pilot needs. A bank feed and a payment-processor integration may describe different sides of the same transaction; review how the ledger avoids duplication and how missing detail is surfaced.

Ask Chat to explain a change in spending and require the reviewer to inspect the referenced transactions. Then use Cowork for a bounded task such as preparing a list of unmatched payments. Separate the proposed work from approved entries. A useful result should leave uncertainty visible where more than one invoice could match the same payment.

The AI Close page describes defining workflows, reviewing proposed entries and keeping an audit trail before posting to the general ledger. Use one close task with a clear owner and due date. Review the source, reasoning and proposed entry together, then inspect what the record shows after approval.

Change one input and repeat the task. A robust workflow should expose the changed evidence and avoid silently duplicating an earlier entry. Also test a rejected proposal: the reviewer should be able to explain what was rejected and what needs correction. Automation is valuable when it reduces preparation work while preserving a clear decision record.

Finish by reconciling the resulting report with the agreed fixture and documenting unresolved exceptions. Do not use a polished dashboard as evidence that every account has been reviewed. This is a proposed product evaluation, not a report that Sequenced closed a set of books or verified compliance with a particular accounting standard.

04 / PricingSoftware, AI usage and accounting services are separate purchases

The pricing page separates software subscriptions, accounting services and additional projects. It has monthly/yearly controls, so confirm the selected commitment in the live checkout rather than treating every displayed monthly figure as cancellable monthly billing. The table below focuses on the commercial structure and the service terms that were explicit in the current help material.

The accounting-services guide lists Expert Reviewed at USD $171 per month including Core software for eligible businesses with up to $10,000 in monthly expenses. Eligible businesses must have one entity, one accounting basis, no payroll processing, no accounts payable or receivable and no inventory, with books already in good shape.

The AI credit guide lists purchased packs of 40 credits for $25, 100 for $50 and 250 for $100. It says purchased credits do not expire. Estimate usage from the actual tasks the team will run; an included allowance does not mean every bulk close or repeated analysis is unlimited.

05 / DistinctionsReview is part of the product’s operating model

The useful distinction in Puzzle’s Close positioning is that the accountant remains responsible for approving entries. A team can encode a repeatable preparation process while preserving a place to inspect exceptions. That arrangement is more informative than a general claim that AI does accounting, because it identifies where a person should exercise judgement.

The current service offer also makes the software-versus-service boundary explicit. A support agent explaining how a feature works is different from a bookkeeper closing the books. The service guide says a vetted partner performs the monthly work. Confirm the responsible firm, scope and review arrangement before assuming the subscription includes an accounting engagement.

Puzzle publishes claims about faster closing and greater capacity. Those are vendor statements and customer examples, not a forecast for your books. Compare the time spent preparing, correcting and reviewing the same kind of period, and account for one-off migration work separately from steady-state use.

06 / QuestionsMCP is an additional access route with alpha limits

The API and MCP page describes connecting compatible assistants to selected companies. MCP starts read-only; read-write access is opt-in and proposed transaction changes require confirmation. The same page labels high-volume limits as part of the current alpha and says connections re-authenticate every five days. A team should plan around those actual limits rather than assuming uninterrupted unattended operation.

Keep external-assistant access separate from the built-in close workflow. Decide which companies the assistant can see and whether it needs write access at all. For a first evaluation, read-only financial explanation may be sufficient. Review the resulting answer against the ledger rather than treating the external assistant as a new accounting authority.

Confirm the service’s fit for your geography, entity structure and accounting requirements. The public offer does not establish that every jurisdiction, inventory process or complex consolidation is supported. Ask the accountant to identify the consequential requirements before importing historical records or promising a migration date.

07 / DecisionChoose an accountable review process first

The decision should begin with who prepares the work, who reviews it and what evidence accompanies an approved entry. Puzzle can be assessed against that process with one period and a small exception set. Expanding AI access should follow a reliable accounting workflow.