Accounting software was built to record work, not prepare it. We are changing that.
AI Close is Puzzle's AI-native close workflow for accounting firms. AI Close prepares the work for you, with the accountant still in control. Governed automation at its core.
This is not for firms chasing hype. It is for firms that want more capacity, better control, and more room for the work that actually requires a professional brain.
You already know what needs to happen in the close. That is not the problem. The problem is the time it takes to execute.
AI Close changes that.
You start with the same month-end checklist and pick a step. Then you describe it the way you would hand it off to someone on your team. Puzzle turns that instruction into a repeatable agent workflow that drafts the work exactly how you want it done, ready for your review.
AI Close is built for the parts of the close that consume the most time. Categorization. Reconciliation preparation. Exception handling. Review.
These steps repeat every month but rarely fit into simple automation rules.
Each instruction becomes a repeatable workflow called an agent. For a deeper look at how these work, see the guide to AI agent workflows for month-end close. The agent drafts the work, flags exceptions, prepares journal entries, and lines everything up for your review.
Nothing posts without your approval. There is no black box. No autonomous system making accounting decisions on your behalf.
This is governed automation in practice. Your process. Your standards. Your judgment. Running faster, under your control.
TLDR:
There are already tools that claim to automate parts of the close.
Most of them sit outside the accounting system.
They are third-party close tools layered on top of the ledger. Data has to sync between systems. When something breaks, teams often have to search across multiple tools to identify the problem.
AI Close works differently.
Puzzle combines the general ledger, the data platform, and the agent workflows in one system.
Because the automation runs directly inside the ledger, agents operate on the underlying accounting data instead of relying on external integrations.
There are no syncing issues to manage. There is no separate system that needs to be kept in sync.
Everything runs in one place.
Because AI Close lives inside the general ledger, it works across both cash and accrual workflows without requiring firms to move data between systems.
You also do not need to know how to code.
You simply describe the task the way you would assign it to someone on your team. Puzzle converts that instruction into an agent that prepares the work for review.
The result is automation that reflects how firms actually run the month-end close instead of forcing them to adapt to a third-party workflow system.
| AI Close (Puzzle) | Third-Party Close Tools | |
|---|---|---|
| Lives inside the general ledger | Yes: automation runs directly on accounting data | No: layered on top of the ledger via sync |
| Data sync required | No: one system, no external integrations needed | Yes: data must sync between systems |
| Cash & accrual support | Yes: works across both without moving data | Often requires separate workflows or exports |
| Setup requires coding | No: describe tasks in plain language | Varies; many require technical configuration |
| Human approval before posting | Always: nothing posts without accountant sign-off | Varies by tool |
| Close time (early adopter results) | 3 to 5 days (down from 15 to 20); clients closed by 4th business day | No equivalent published benchmarks |
That shift changes the economics of a firm.
Firms can expand capacity without hiring at the same pace. The traditional growth model for an accounting firm is linear: more clients means more bookkeepers. AI Close breaks that equation. A firm that previously needed one full-time bookkeeper per 10 clients can serve up to three times as many clients with the same team, because the agents handle preparation and the accountant handles review and judgment. For a broader look at how Puzzle approaches the month-end close, see how AI Close stacks up against the field.
Senior accountants spend less time buried in preparation and review loops and more time advising clients, a shift that reflects the broader debate around AI bookkeeping automation vs. judgment. That's better for firm margins, and it's a better service for the client.
When demand increases, firms can take on more work without immediately adding headcount. Early adopters are already seeing this in practice: Debit & Co. is on track to triple its client capacity with the same team. Accountalent cut its close from 15 to 20 days down to 3 to 5 days.
The close that once stretched across weeks begins moving in the background.
Connect your integrations and an agent can begin preparing reconciliation work before you log in. Upload a statement and the agent drafts the reconciliation while you focus elsewhere.
Client-ready insights begin taking shape before anyone opens the laptop the next morning.
The accountant remains at the center of the process.
AI Close does not replace professional judgment. It amplifies it.
The system prepares the work. The accountant reviews, approves, and signs off.
AI Close was built with firms, not dropped on them after the fact. The workflows reflect how real firms close the books.
Because no two firms operate exactly the same way, you are not forced into someone else's rules.
You define the logic. You define the exceptions.
Over time, your firm's way of working becomes embedded in the system itself.
That is what AI-native, human-led accounting looks like when it is done right.
The most time-consuming parts of the monthly close aren't the hard judgment calls. They're the low-value, high-volume tasks that repeat every month: sorting transactions, coding vendor entries, matching receipts, and preparing categorization for review. Accounting firms adopting AI bookkeeping software for CPA firms are now eliminating most of that manual work.
AI Close handles transaction categorization automatically by reading the underlying accounting data in the general ledger and applying firm-defined rules to code each transaction. Routine entries (payroll, SaaS subscriptions, vendor payments) are categorized without anyone touching them. Exceptions are flagged for accountant review. The result: accountants spend their time on the 5% of transactions that require judgment, not the 95% that don't.
The shift is measurable. As noted above, early adopters like Accountalent have cut close time dramatically: time that previously went to manual data entry now goes toward client advisory, review, and business development. Transaction categorization that once required a staff accountant working through a spreadsheet now runs in the background before anyone logs in.
Because AI Close operates inside the general ledger, not as a third-party overlay, categorization rules stay connected to the live accounting data. There's no export-import cycle, no sync lag, and no duplicate cleanup. The agent categorizes, flags exceptions, and prepares the work for your sign-off, all inside the same system where the books live.
The case for AI in the close is no longer theoretical. A large empirical study of generative AI in accounting, conducted by researchers from Stanford and MIT and published in 2026, analyzed data from 79 small- and mid-sized firms and surveyed 277 accountants. The findings were concrete: greater AI adoption was associated with a 7.5-day reduction in monthly close time, an 8.5% shift in accountant time away from routine data entry toward higher-value work, and an 18% increase in weekly client support capacity.
The study also found that experienced accountants remained irreplaceable: professionals were more likely to intervene when AI systems flagged lower confidence, confirming that AI works as a complement to accounting expertise, not a replacement for it. That is exactly the model AI Close is built on.
The profession is moving. The firms that build governed automation now, with their own logic, their own exceptions, and their own review standards, will be the ones that are hardest to catch.
The timing matters. Thomson Reuters' 2026 AI in Professional Services Report found that organization-wide AI adoption in professional services nearly doubled year-over-year, rising from 22% in 2025 to 40% in 2026. Karbon's State of AI in Accounting 2026 puts it even more sharply: 98% of accounting professionals now report using AI in some capacity. The early-adopter phase is over.
What's coming next is agentic AI: systems that don't just answer questions but execute recurring workflows autonomously. Thomson Reuters found that 15% of firms have already deployed agentic AI tools, and another 53% are either actively planning for them or seriously considering it. The firms that move first on accounting agent AI automation will have a structural advantage that compounds over time: more clients served, lower cost per close, and a process foundation their competitors will spend years trying to replicate.
The firms that win won't be the ones with the most AI. They'll be the ones who deploy it inside a system with clear accountability, human approval, and audit trails: exactly the model AI Close is built on.
AI Close launched as part of the Puzzle AI Suite on July 9, 2026: generally available to accounting firms. Early adopters are already seeing the results: Debit & Co. is closing every client's books by the 4th business day. The firms that move first will close faster, take on more clients, and build a structural advantage their competitors won't be able to close.
AI Close is Puzzle's automated close workflow: it runs agent-based workflows that prepare categorization, reconciliations, and journal entries on a schedule before you log in. Accounting AI chat is a conversational interface for asking questions about your books, running ad-hoc lookups, and getting explanations in plain language. They're complementary, not interchangeable: AI Close handles the recurring preparation work that repeats every month; Accounting AI chat handles on-demand questions and exploration. Pricing for AI Close is available on request and scales with the number of clients and workflows your firm runs. Contact Puzzle to get a quote sized for your practice.
Yes. With AI Close, agents run on a schedule and prepare work before you log in. Connect your integrations (bank feeds, payroll, expense platforms) and an agent can begin categorizing transactions, preparing reconciliations, and drafting journal entries in the background. You don't need to log in to trigger the work. When you do log in, the drafts are waiting for your review and approval. Nothing posts to the books without your sign-off, but the preparation happens automatically, so your close moves forward even when you're focused elsewhere.
Yes. Puzzle is built to scale with a firm, which means it works at day one as well as at scale. If you're building your first client roster, you don't need a large team or a complex tech stack to get value from AI Close. You describe your close logic the same way whether you have two clients or two hundred. For early-stage firms, this matters because you're defining your processes now, and building them inside a governed automation system means your workflows are repeatable and transferable from the start, not locked in your head or in a spreadsheet. Puzzle also supports a partner-only model, so firms can work directly with Puzzle instead of through a third-party reseller.
Puzzle is one of the few accounting platforms where AI agents operate directly inside the general ledger, not as a bolt-on layer that syncs with a separate system. AI Close agents can draft journal entries, prepare reconciliations, and handle categorization as part of an automated close workflow. The key distinction: AI Close prepares and drafts entries for accountant review and approval. Posting requires human sign-off. This keeps the firm in control while eliminating the manual preparation work that consumes most of the close timeline.
A financial dashboard shows you what already happened: account balances, spend by category, burn rate. It's a reporting layer. An AI-native close workflow like AI Close does the work that produces those numbers: categorizing transactions, preparing reconciliations, drafting journal entries, and flagging exceptions, all before you review. For a founder without a full-time bookkeeper, that distinction matters. A dashboard requires your month-end close to already be complete and accurate; AI Close is what gets them there. With AI Close, a founder or a part-time accountant can maintain accurate monthly books without a full-time bookkeeper dedicated to preparation work, because the agents handle the preparation and the human handles review and sign-off.
Puzzle is built for multi-client firm workflows. Each client account has its own permissions layer, so staff accountants can access the clients assigned to them without visibility into others. Firm admins manage access centrally across the full client portfolio. Every action in Puzzle (categorization approvals, journal entry sign-offs, workflow completions) is logged in an audit trail, giving firms a clear record of what was done, by whom, and when. This is especially important for firms operating under review standards where documentation of the close process is required.
Puzzle integrates natively with Gusto payroll automation and Rippling. When payroll runs, AI Close can automatically prepare the corresponding journal entries, mapping payroll components to the correct accounts in your chart of accounts based on rules you define. You describe the logic once (which accounts to debit and credit for wages, employer taxes, benefits, etc.), and the agent applies it every payroll cycle. The entries are drafted and staged for your review before anything posts. This removes the manual step of translating payroll reports into journal entries each period.
AI agents inside Puzzle operate directly on your accounting data; they don't call out to external AI tools or require a separate integration to function. Inside the general ledger, an agent reads your live transaction data, applies the categorization logic you define, matches line items to accounts, drafts journal entries, and flags exceptions, all without leaving the system where your books live. Tasks AI Close handles natively: transaction categorization, bank and credit card reconciliation prep, payroll journal entry drafts (from Gusto, Rippling), vendor payment matching, and exception flagging before accountant review. For vendor invoice PDFs, agents can extract line-item data and map it to the correct accounts based on rules you configure, entirely inside Puzzle, with no export step to an external AI tool. The distinction matters: when automation runs inside the ledger, there's no sync lag, no duplicate cleanup, and no question about which system has the authoritative version of your data.
Early adopters are closing in 3 to 5 business days, down from a typical 15 to 20. What AI Close handles automatically: transaction categorization, reconciliation drafts, payroll journal entries, and exception flagging. What still requires human input: reviewing and approving drafted entries, resolving flagged exceptions, signing off on journal entries before they post, and any judgment calls that fall outside the rules you've defined. Nothing posts without accountant approval; that's by design. The close accelerates because agents eliminate the preparation work that fills most of those 15 to 20 days; the human time that remains is the review, judgment, and sign-off that genuinely requires a professional. The Stanford/MIT study of 79 firms found AI adoption shifted 8.5% of accountant time away from routine data entry toward higher-value work: the human role doesn't disappear, it upgrades.
The traditional model is linear: more clients means more bookkeepers. AI Close breaks that equation. Because agents handle the preparation work (categorization, reconciliation drafts, journal entries), accountants shift from execution to review and judgment. A firm that previously needed one full-time bookkeeper per 10 clients can serve far more with the same team. Early adopters are already seeing this in practice: Debit & Co. is on track to triple its client capacity with the same team. Accountalent cut its month-end close from 15 to 20 days down to 3 to 5. The lever isn't headcount: it's how much of the close your team spends on preparation versus review. AI Close moves that ratio. When new clients come in, the agents absorb the volume increase before you need to think about your next hire.
Yes. You can start with a single client (new or existing) before rolling Puzzle out across your firm. Most firms begin with one client to get familiar with AI Close workflows and the general ledger before committing to a broader migration. For clients migrating from QuickBooks, Puzzle's onboarding team handles the data migration, including chart of accounts mapping, historical transaction import, and opening balances, as seen in the Debit & Co. migration. The migration is structured to minimize disruption: your existing client data comes with you, and the close logic you define in Puzzle is yours to refine over time. Firms that have migrated report the transition taking days, not weeks, for a typical client account (see Debit & Co.).
