Unlock operational advantage with Puzzle's Accounting AI. Eliminate manual drudgery, catch errors earlier, and close books 20-50% faster, keeping full professional control.
Accounting firms today face a difficult balance: increasing client demands, higher complexity, tighter deadlines, and rising expectations for faster reporting. At the same time, most “Accounting AI” products promise full automation or “self-driving bookkeeping” or “Autonomous bookkeeping” often at the expense of accuracy and professional control.
Puzzle built a different type of Accounting AI.
Instead of removing accountants, Puzzle’s AI helps firms eliminate manual drudgery, catch errors earlier, and deliver more accurate books in less time. Firms that adopt Puzzle aren’t chasing the latest technology trend, they’re gaining a concrete advantage in how they work that compounds month after month.
This article outlines the Accounting AI features Puzzle has brought to market, how firms use them, and why Puzzle’s approach to AI is built around accuracy, control, and firm scalability, not automation for automation’s sake.
We test, validate and refine the best models for the job at hand, to bring you the best capabilities the market offers, all in a SOC2 secure, encrypted environment.
TLDR:
Accounting AI that reduces back-and-forth and manual categorization.
Puzzle was an early adopter of LLMs (ChatGPT 3.5 → 4) to categorize transactions from simple natural-language messages between accountants and their clients. Clients could reply in plain English:
“This was our AWS hosting payment.”
The system categorized it instantly, with no rules to maintain and no manual tagging.
How this helps firms:
This was Accounting AI used the right way: to support clear communication and remove friction.
Accounting AI that saves hours on reconciliations.
Accounting AI can read documents better, and it can also draft, assemble, review and identify anomalies better than historical OCR or machine learning.
Puzzle began matching transactions and balances automatically, reducing reconciliation time dramatically.
Outcomes for firms:
Firms consistently cite this as one of the biggest time savings in their month-end close.
Accounting AI that checks every transaction for accuracy, not a sample.
Puzzle launched accuracy reviews that scan the full ledger for:
How firms use this Accounting AI:
Because issues surface during the month, this feature can reduce the time reviewers spend hunting for errors at close.
Accounting AI that learns vendor patterns with greater accuracy.
Puzzle quietly introduced vector-similarity categorization using Vertex AI to identify patterns across thousands of vendors and transaction descriptions.
This improves classification accuracy for edge cases and reduces the need for custom rules.
How firms benefit:
This matters because categorization quality impacts every downstream workflow.
Accounting AI that reviews your financial health and identifies insights
This AI acts like a financial analyst, telling you financial health and insights in multiple ways. The good, bad and ugly, which we call “Steve Jobs” style, a blend of good and bad, which we call “like a friendly investor” style, and just the good stuff, which we call “don’t hurt my feelings” style.
Firm impact:
Accounting AI doesn’t replace judgment. It keeps anything from getting overlooked.
Accounting AI that checks financial statements line-by-line.
This AI reviews books using GAAP-aligned logic to identify misstatements or unusual trends.
Firm impact:
Accounting AI that drafts accruals, prepaids, depreciation, and revenue recognition entries.
Status: accrual drafting now runs inside AI Close, part of the Puzzle AI Suite, which became generally available to firms on July 9, 2026. Close drafts revenue recognition, payroll, depreciation, and amortization schedules, and nothing posts to the ledger until a human approves it.
Puzzle automates accruals natively in software, built through a mix of software and design.
Benefits for firms:
This is Accounting AI that supports better accounting, not shortcuts.
| Feature | Released | Key Outcome |
|---|---|---|
| Natural-Language Categorization | 2023 | Fewer uncategorized transactions; less time chasing client details |
| Automated Bank Reconciliation | 2023 | Reconciliation time drops from ~2 hours to ~5 minutes |
| Transaction & Vendor AI Accuracy Reviews | 2024 | Every transaction scanned for errors; books audit-ready by default |
| Vector Similarity Categorization | 2024 | More accurate edge-case classification; faster client onboarding |
| Financial Health Insights & Variance Analysis | 2025 | AI financial analyst surfacing good, bad, and unusual trends |
| GAAP-Informed Statement-Level Reviews | 2025 | Line-by-line statement checks; faster anomaly identification |
| Accrual Agents (prepaids, depreciation, rev rec) | 2025; now part of AI Close (2026) | Fewer spreadsheets; faster move from cash to accrual |
Puzzle’s automation reduces categorization and reconciliation workloads so firms can grow without proportional hiring.
Review-only workflows replace manual matching, checking, and spreadsheet-heavy processes.
AI Accuracy Reviews and accrual drafting catch issues before they become problems.
Accountants spend less time on repetitive tasks and more time on client-facing work.
Firms using Puzzle report month-end close time cut by 25% (Burkland) and reconciliations cut from 2 hours to 5 minutes (Trivium), so they deliver reports sooner.
Accounting that is drafted by software and AI, and reviewed and improved by accountants. No self driving accounting AI or changes made autonomously behind your back.
Our public pledge:
https://puzzle.io/blog/our-commitment-to-accounting-firms
Puzzle is built on a modern ledger designed for real-time checks and error detection. We constantly test the best models on the market, so you don’t have to.
No hidden automation, no “books changed overnight.”
Firms like Trivium and Burkland use Puzzle and report results such as bank reconciliations cut from 2 hours to 5 minutes (Trivium) and month-end close time reduced by 25% (Burkland).
Puzzle helps firms:
This is Accounting AI with measurable business impact, not marketing claims.
The industry data has caught up to what early adopters already knew. According to Karbon's State of AI in Accounting Report 2026 (nearly 600 accounting professionals surveyed), 98% of firms now use AI, most of them daily or several times a day. Average time savings reached 60 minutes per day per employee, or about 21 hours per month. Firms with AI training, policies, and a documented strategy see the biggest gains, yet only 21% have an AI policy or strategy.
The frontier has also shifted. In 2026, the conversation moved from "should we use AI?" to "which AI model is right for which task?" Agentic AI systems that draft work and wait for a human to approve it are now in production: Puzzle made its AI Suite, including Close, generally available to firms on July 9, 2026. The gap between firms that adopted AI deliberately and those that didn't is becoming measurable in client capacity, close times, and margins.
That's the environment Puzzle was built for. The features in this article are not experiments. They are production capabilities that firms use today to stay on the right side of that gap.
Adoption is no longer the question. Depth is. Intuit's 2026 Accountant Technology Survey (725 US accounting and bookkeeping professionals) found that 88% used AI for at least one client service in the past 12 months, and 77% agree the gap is widening between firms with AI built into daily workflows and firms that use it occasionally. In October, Inside Public Accounting's 2026 IT survey, discussed on its podcast, reported that just 3.2% of firms say AI is fully integrated into how they operate.
Control matters as much as depth. In the Intuit survey, only 6% of respondents want AI to execute client deliverables autonomously, while 34% want AI to draft with a human reviewing and signing off. That is the model Puzzle built: AI drafts, flags, and matches transactions during the month, and your team approves before anything posts. The same survey found that manual data cleanup is the top barrier to advisory work, cited by 30% of respondents. Categorization, reconciliation, and accuracy reviews are built to take that cleanup off your desk.
The most effective firms have replaced rules-based categorization with LLM-driven workflows. Instead of manually tagging transactions or maintaining long rule lists, accountants describe what a transaction is in plain English and the AI categorizes it instantly. Puzzle takes this further with vector-similarity matching across thousands of vendor patterns, so even edge-case transactions get classified correctly the first time. The result: categorization that once consumed hours of staff time each month now runs in the background, and accuracy reviews scan every transaction (the full ledger, not a sample) to catch anything the model missed. Firms using this approach consistently report that their staff has shifted from data entry to review-only workflows, which is where AI bookkeeping automation and accountant judgment actually add value.
At seed stage (single entity, no controller, lean team), you need reconciliation that runs automatically without custom setup. Puzzle connects directly to your bank and fintech accounts (Mercury, Brex, Ramp, Stripe) and matches transactions and balances automatically, reducing reconciliation from roughly two hours to about five minutes per month. It also flags timing differences, missing entries, and duplicates so you're not finding errors after the close. Legacy tools like QuickBooks were built for manual workflows; Puzzle was built AI-native, which means reconciliation is a background process, not a monthly project. For a seed-stage startup, that's the right size for the problem.
Dual-basis accounting means maintaining both cash-basis and accrual-basis books simultaneously: cash so you can monitor daily burn and runway, accrual so your financials are GAAP-compliant for investors and auditors. Most software forces you to pick one, or requires manual journal entries to bridge the two. Puzzle maintains both bases natively in the same ledger, so your real-time cash position and your accrual statements are always in sync without extra work. AI Close, generally available to firms since July 9, 2026, also drafts prepaids, depreciation, deferred revenue, and revenue recognition entries for your review, reducing the spreadsheets and manual entries that dual-basis accounting traditionally requires.
Gusto and Rippling both offer native accounting integrations that push payroll journal entries automatically when payroll runs. The entries typically include gross wages, employer taxes, benefits deductions, and net pay, mapped to the GL accounts you configure in your accounting software. In Puzzle, payroll runs from Gusto or Rippling sync automatically and Puzzle's AI accuracy reviews flag any mismatches between payroll data and your chart of accounts, so errors surface before month-end instead of during review. If you're on the accrual basis, Puzzle's accrual agents can also draft wage accrual entries for pay periods that cross month boundaries, eliminating the manual journal entries that trip up most teams running payroll on a bi-weekly cycle.
For payroll providers without a native Puzzle integration (ADP, Justworks), you upload the payroll report, typically a PDF or CSV export. Puzzle's AI payroll agent reads the uploaded report and drafts the journal entry from it: gross wages, employer taxes, benefits, and net pay mapped to the correct GL accounts. You review the draft before it posts. Puzzle's AI accuracy reviews also flag payroll-period mismatches or missing entries when they're detected against your bank transactions, so gaps don't stay hidden until the close.
The AI Close agent runs your month-end close checklist automatically. It scans for uncategorized transactions, reconciliation gaps, accrual entries that need drafting, and statement-level anomalies, then surfaces everything that needs attention before you sign off. Available agents and templates include bank reconciliation, accrual drafting (prepaids, depreciation, deferred revenue), accuracy reviews, and GAAP-informed statement checks. AI credits are consumed per agent run; the number of credits a monthly close uses depends on entity size and which agents you turn on. Firms on partner plans typically receive a credit allocation sized for their client book. Your Puzzle account will show credit usage per client so you can track it across your portfolio.
Yes. Firm partners get a dedicated onboarding path that direct customers don't. Puzzle's partner program includes a dedicated partner success contact, priority support routing, and access to firm-specific training on multi-client workflows, the client portal, and AI feature rollouts. Direct business customers receive standard onboarding and support. The distinction matters in daily work: when you're managing 30+ clients, you need answers fast, and a ticket queue built for individual founders doesn't serve that.
The right answer depends on how far back the messiness goes and what you actually need from those historical records. The three approaches each have a real cost: importing raw transactions means inheriting the errors and cleaning them inside Puzzle; waiting until the books are clean delays your go-live; starting fresh with a beginning balance is the fastest path but loses transaction-level history. Most firms migrating messy books choose a hybrid: import a clean beginning balance for each account as of a cutoff date (typically the start of a fiscal year or quarter), bring in raw transactions only for the most recent period you'll actively work in, and archive the older QuickBooks file for reference. Puzzle's AI accuracy reviews will flag inconsistencies in whatever you import, so errors surface quickly instead of quietly compounding. If you have a controller or CPA helping with the migration, the cutoff-balance approach is almost always the cleaner call.
When personal and business accounts are connected through the same bank login (common with Mercury, Chase, or Bank of America), Puzzle pulls all accounts visible under that credential. During setup, you choose which accounts to activate, so you can connect the login without syncing every account it touches. If personal transactions are imported accidentally, you can exclude or delete them from the transaction list; they won't affect your books until they're categorized and posted. For ongoing separation, the cleanest approach is to keep your business banking fully separated at the institution level: a dedicated business account under a separate login eliminates the problem permanently. If transactions were already categorized incorrectly, Puzzle's accuracy reviews will flag the anomalies so you can correct them before close.
The constraint is how much manual work each client requires, not client count. Puzzle reduces per-client work by automating categorization, reconciliation, and accrual drafting, so accountants spend less time per client. Firms using Puzzle report month-end close time cut by 25% (Burkland) and reconciliations cut from 2 hours to 5 minutes (Trivium) by shifting from manual-entry workflows to review-only workflows. For a firm at 50+ clients, the compounding effect matters: every hour saved on reconciliation and categorization across the book of business adds up to meaningful capacity. Puzzle also supports multi-client accounting software natively, so firms see their entire client base at once, with every entity in one view.
Firms are replacing the month-end PDF package with a live ledger that clients can check any day. In Puzzle, cash position, P&L, and balance sheet update daily, and burn rate, runway, and ARR/MRR are tracked automatically from connected sources (Stripe, Mercury, Brex, Ramp, Gusto). AI does the categorization and reconciliation during the month, so the numbers a client sees are already reviewed instead of reconstructed at close. Financial Health Insights and variance analysis add plain-English commentary on what changed, and AI Chat answers client questions from the actual books (read-only). When a client still wants a PDF for a board or lender, PDF export is built in on all plans. The shift for your firm: the monthly report becomes a review meeting about decisions, not a delivery of last month's numbers.
Use an AI system that runs inside the accounting software instead of pasting client exports into a general-purpose chatbot. Puzzle's AI Chat is a secure, closed-loop system built for accounting: it queries the client's ledger in plain English, does not send sensitive data to public LLMs, and does not search the web. It is grounded in accounting principles and the client's real data, and it is read-only, so nothing changes in the books without a human approving it. Puzzle is also SOC 2 compliant. Before you adopt any AI tool, check three things: where client data goes, whether it is used to train models, and what retention controls apply. Then write those answers into a firm AI policy. Karbon's 2026 data shows only 21% of firms have one.
Look for accounting software that connects to your bank and fintech stack directly and reports burn, runway, and cash position without a spreadsheet in the middle. Puzzle connects to Mercury, Brex, Ramp, Stripe, and Gusto, updates cash position, P&L, and balance sheet daily, and tracks burn rate, runway, and ARR/MRR automatically. About 90% of transactions are categorized automatically and about 85% are finalized, so a founder typically spends 15 to 20 minutes a week reviewing instead of building reports. Know the limits: Puzzle tracks runway and burn but has no forecasting or scenario planning module, and native daily reconciliation covers Mercury and Ramp, while other banks need manual CSV uploads. If you want a bookkeeper in the loop, Expert Reviewed or Full Service accounting runs on the same ledger.
Cowork turns a one-time request into a repeating job. You describe a task once in plain English (for example, a weekly review of uncategorized transactions or a monthly spend variance report), and Cowork saves it as a prompt-based agent that runs on a schedule and prepares the work for your review. Nothing changes in the books until a person approves it. Cowork is included in all plans. It sits alongside the automation that already runs in the background: about 90% of transactions are categorized automatically and about 85% are finalized, and account-based categorization rules apply as you finalize transactions. Revenue recognition runs from policies you set by product or subscription type.
Migration costs $149 per system moved (QuickBooks, Xero, or another system), charged once per system. A QuickBooks Online migration takes a couple of days through a direct integration. QuickBooks Desktop, Xero, and Zoho Books take about five to seven days because the process is manual. The QuickBooks Connect button imports historical financial data only and does not sync live. If the historical books are messy, import a cleaned trial balance as your opening balance or set a clean go-live date instead of bringing in raw history. On categorization, Puzzle learns from the transactions you finalize inside Puzzle, so expect accuracy below the typical rate (about 90% categorized automatically and about 85% finalized) in month one, improving after the first month-end close. Do not assume imported history will train the AI.
Native integrations include Stripe, Mercury, Brex, Ramp, Gusto, Rippling, Deel, and Carta (equity journal entries only, not cap table management), plus Zapier and Finch. Most connect in about two minutes. For Stripe revenue, you set recognition policies by product or subscription type, and Puzzle defers and recognizes revenue over the correct period at the invoice line-item level, so different line items can follow different policies. You can confirm schedules in bulk (up to 100 items per batch). You can also attach a policy to manually created invoices. The limit to know: revenue recognition is Stripe-centric, so billing that does not run through Stripe or a manual invoice needs manual journal entries.
For digital products billed through Stripe, yes. Revenue flows in from Stripe, recognition follows your policies, and you can view customer-level profit and loss and project-level P&L. Shopify is not on Puzzle's native integration list, and Puzzle has no inventory module, so order-level inventory and item-level COGS tracking are not supported. That is a permanent product boundary. If you carry physical inventory, keep inventory tracking in a dedicated inventory tool. If your business sells software, subscriptions, or other digital products, Puzzle fits.
Puzzle does not offer tax services, so tax filing and tax advice sit with your CPA. What a CPA needs from you is clean, accurate books with every account matched to its bank statement. If you are a few months behind with simple activity, catching up yourself in Puzzle is realistic, since the AI drafts categorization and you review about 15 to 20 minutes a week. If you are many months behind, catch-up bookkeeping starts at $300 per month of books (eleven months behind is $3,300) and requires an active Puzzle subscription. For ongoing work, Full Service starts at $499 per month per entity on top of a software plan. Ask your CPA for the filing deadline first, then work backward. Puzzle cannot promise a specific turnaround date.
The future of accounting isn’t AI that replaces accountants, it’s AI that empowers them to deliver better service at scale.
Puzzle gives firms:
That’s why more firms are adopting Puzzle as their Accounting AI software, and why the most forward-thinking teams see it as their competitive advantage for the next decade.
Want to see it on your own books? Book a demo.
