The AI-Assisted Close: How CAS Firms Are Using Keeper and Numeric in 2026

Jul 28

Part 4 of the Lumiere AI Stack Series. Per-Client Close and Review. Start with the pillar post if you are new to the series.

The monthly close is where outsourced accounting firms earn or lose client trust.

A clean close, delivered on time, with a brief summary of anything that looked unusual, is the baseline clients are paying for. A close that runs late, surfaces errors at review, or requires three rounds of clarifying questions signals operational fragility regardless of how good your advisory work is.

At low client volume, experienced staff carry the close in their heads. They know which clients have recurring miscodes, which bank feeds need manual review, which clients will not upload their receipts until the last day of the month. That institutional knowledge works until it does not, which is typically somewhere between 15 and 25 clients per staff member, or when someone leaves.

In 2026, the tools that matter for the close layer are not trying to replace that judgment. They are trying to make it transferable: surfacing the exceptions, documenting the decisions, and letting a staff member run a consistent close across a larger book of clients without rebuilding the process from scratch each month.

This post covers what those tools are, how the close workflow runs inside them, and what the real tradeoffs are before you commit to any of them.

Why the Close Looks Different Inside an Outsourced Firm

The month-end close for an internal finance team is a single-entity problem. One chart of accounts, one bank, one set of adjusting entries, one review sign-off. The tooling built for that workflow, FloQast, BlackLine, Trintech, reflects that architecture.

An outsourced firm is a different operational structure entirely. You are running 20, 30, or 50 simultaneous closes, each on a different client's chart of accounts, each with its own quirks, its own client communication requirements, and its own deadline. The complexity is not depth on a single file. It is breadth across many files with a finite team.

That distinction drives three specific requirements that enterprise close tools do not address:

Cross-client visibility. Where is every client's close right now? Which files are at review stage, which have open questions, which have not started? This view needs to exist across all clients simultaneously without logging into each individual accounting file.
Standardized review, applied per client. The same quality checks need to run on every client file, even though each file is different. Miscoded transactions, unreconciled accounts, missing support, uncategorized items: these checks should be automated and consistent, not dependent on which staff member happens to be reviewing the file that month.
Client communication tied to the file. When a question arises during close, it needs to go to the client with context, be tracked, and be resolved before the close is signed off. Managing that communication through email means losing the thread. Managing it through a general-purpose tool like Slack means losing the connection to the specific transaction.

The tools worth evaluating in 2026 are the ones built around these three requirements, not the ones adapted from single-entity software.

The Tools Worth Knowing in 2026

Double (formerly Keeper)

Double is the most CAS-native close management tool in the market. It rebranded from Keeper in October 2025 after closing a $6.5 million Series A, a signal of meaningful investment in the roadmap. The rebrand reflects a deliberate broadening: from a transaction review tool to a unified close and practice management platform for bookkeeping firms.

The core workflow in Double runs directly against the connected ledger. It pulls transactions from QBO or Xero, runs automated checks for miscategorizations and reconciliation gaps, and surfaces exceptions in a review queue. Staff work through the queue, resolving clean items and sending questions to clients through Double's built-in client portal. That portal keeps client communication attached to the specific transaction, not floating in an email thread.

The AI layer handles the classification and anomaly detection: flagging transactions that do not match historical patterns, suggesting categories for uncoded items, and auto-generating journal entries from complex source documents like payroll reports or settlement statements. Human judgment stays in the loop at the review and sign-off stage.

Pricing starts at $200 per month for the firm, plus per-client costs, with unlimited users. Double is a Platinum partner in the Intuit App Partner Program, which reflects the depth of its QBO integration. Xero integration is also supported. NetSuite and Sage Intacct are available on the Enterprise tier only.

The meaningful limitation is the ledger dependency. If your client base is primarily QBO or Xero, Double is purpose-built for your workflow. If you have meaningful volume on Sage Intacct, NetSuite, or other ERPs, the core review workflow does not apply at the standard tier.

Best fit: CAS and bookkeeping firms closing 10 to 175 client files per month, primarily on QBO or Xero, that want the close review and client communication workflow in a single platform.

Numeric

Numeric is a close management platform with a strong focus on flux analysis, variance review, and reconciliation oversight. Its AI layer is well-regarded for surfacing the "why" behind account movements, not just flagging that something changed. For firms that spend meaningful time on close review, explaining variance to clients, or managing accrual schedules, the analytical depth is genuine.

The honest tradeoff is that Numeric was built with corporate finance teams in mind. Its architecture is not inherently multi-tenant in the way Double's is. Firms managing a large, diverse client portfolio may find the workflow clunkier than purpose-built firm tools, particularly around client communication and cross-client visibility. The platform excels when a staff member is running a close for a single complex client and needs analytical depth. It is less optimized for the staff member who needs to move efficiently across 20 client files in close week.

Pricing is per user at approximately $30 per month, which can be cost-effective for smaller teams but scales differently than per-client pricing as the book grows.

Best fit: Firms with larger, more complex client engagements where close review involves significant flux analysis and variance explanation, and where the team is working more like an embedded finance team than a multi-client bookkeeping practice.

Financial Cents (close module)

Financial Cents is primarily a practice management platform, covered in more depth in Post 6, but its month-end close module is worth noting here because it bundles close task management with workflow automation, client portal, and document collection in one platform. For firms that do not want to run a separate close tool alongside a practice management platform, Financial Cents offers close functionality within a broader operational system.

The depth of the close-specific features is less than Double's dedicated review workflow or Numeric's analytical layer. The tradeoff is integration: the close checklist lives in the same system as client records, billing, and staff assignments, which reduces the coordination overhead of managing multiple tools.

Best fit: Smaller firms, or firms in an earlier operational stage, that want close management and practice operations in a single platform without the complexity of integrating purpose-built tools.

FloQast

FloQast is a well-established close management platform that handles close checklists, reconciliation sign-off, and flux analysis at scale. It is worth knowing because some outsourced firms serve clients sophisticated enough to have existing FloQast deployments, and running the close inside the client's own FloQast instance is occasionally the right answer.

For a firm's own close operations across multiple clients, FloQast is not the right fit. It was designed for internal accounting teams and its multi-client architecture reflects that: it works well for one entity's close, not for a firm running parallel closes across dozens of clients from a single firm dashboard.

Best fit: Larger clients with internal finance teams who already use or want FloQast. Not a primary platform recommendation for the outsourced firm's own close operations.

The Close Workflow, Step by Step

Here is what a well-structured AI-assisted close looks like inside a CAS practice in 2026. This assumes Double as the primary platform, but the underlying sequence applies broadly.

Step 1: Pre-close setup (done once per client, updated as needed). Connect the client's accounting file to the platform. Configure the standard close checklist for that client: bank reconciliations, recurring journal entries, accruals, accounts to review. Flag any client-specific rules, for example, a vendor that always miscodes to the wrong expense account, or a category that should always be questioned before sign-off. This configuration work is what makes the automated review meaningful. Without it, the tool flags everything and nothing.
Step 2: Automated transaction review. As transactions come in throughout the month, the platform runs continuous checks against the configured rules. At close time, the exception queue contains only the items that need human attention: uncoded transactions, pattern anomalies, items flagged by the AI as likely miscodes. Staff work through the queue rather than reviewing every transaction. On a well-configured file, this reduces review time significantly.
Step 3: Client communication. Questions that cannot be resolved from the file go to the client through the platform's portal. The client receives a specific question tied to a specific transaction, responds, and the response is recorded in the file. The staff member can see all outstanding client questions across all clients in one queue. Nothing gets lost in an email inbox.
Step 4: Reconciliation and journal entries. Bank reconciliations run against the connected feed. Recurring journal entries post automatically where configured. AI-assisted journal entry generation handles complex source documents: payroll summaries, depreciation schedules, prepaid amortization. The staff member reviews and approves; the AI prepares the first draft.
Step 5: Review and sign-off. When all exceptions are resolved and reconciliations are complete, the staff member conducts a final review. The platform surfaces a summary of what was flagged, what was resolved, and what was approved. Sign-off is documented in the platform, creating an audit trail for the close.
Step 6: Delivery. The close summary, which may include a brief narrative of anything material that came up during the month, goes to the client. In most practices this is a lightweight PDF or email generated from the platform, not a formal report. The formal reporting layer is Layer 04 in the stack, covered in Post 8.

The Uncategorized Transaction Problem

Every outsourced firm has it. Clients who transact through platforms that do not map cleanly to the chart of accounts. Vendors with ambiguous descriptions. One-off purchases that do not fit any existing category. These pile up in the "Uncategorized" bucket and get cleaned up manually at close, which is one of the most time-consuming parts of the monthly cycle.

In 2026, the better close tools handle this in two ways. First, rule-based auto-categorization: if a vendor has been coded the same way for six consecutive months, new transactions from that vendor get auto-coded with a confidence flag. Staff confirm rather than code. Second, AI-assisted categorization: for new or ambiguous vendors, the AI suggests a category based on the transaction description, amount, and context. Staff accept, modify, or reject.

Neither approach eliminates uncategorized transactions entirely. Clients will always generate genuinely ambiguous transactions. But reducing the review burden by 60 to 70 percent on a well-configured file is realistic with current tooling, and that reduction compounds across a large client portfolio.

One tool worth knowing here is Uncat, which is specifically focused on the uncategorized transaction problem. It integrates with QBO and Xero, routes uncategorized items directly to clients for clarification, and tracks responses. For firms whose primary close bottleneck is uncategorized volume rather than reconciliation complexity, Uncat as a standalone addition to the existing stack can be more cost-effective than moving to a full close management platform.

Building the Per-Client Close Checklist

The most important thing you can do before configuring any close tool is build the per-client close checklist in writing. Not in the tool's configuration interface. In a document, with a colleague reviewing it, before you touch the software.

The checklist should answer three questions for each item: What needs to happen? Who is responsible? How do we know it is done? A close checklist that cannot answer all three for every item is not a checklist. It is a reminder list, and reminder lists do not scale.

A standard outsourced firm close checklist covers: bank and credit card reconciliations, accounts receivable aging review, accounts payable aging review, recurring journal entries, accrual entries, fixed asset and depreciation schedules, balance sheet tie-out, and final review sign-off. Each client will have variations on this standard depending on their business model, accounting method, and the specific services in scope.

The close tool makes the checklist operational, trackable, and consistent. It does not generate the checklist. That thinking is yours.

What AI Is Actually Doing Here

The AI in close management tools is doing specific, narrow, high-value things. It is worth being clear about what those are so you can evaluate vendor claims accurately.

Pattern recognition for miscodes. When a transaction does not fit the expected pattern for a vendor, account, or amount range, the AI flags it. This is the most reliable AI function in the close layer and the one that delivers the most consistent time savings. Firms that have configured their files well report catching material errors that would have passed manual review.
Journal entry preparation from source documents. AI that can read a payroll summary or settlement statement and produce a draft journal entry is genuinely useful and meaningfully faster than manual entry preparation. The staff member reviews and approves; errors still get caught. But the first draft is no longer a manual task.
Flux analysis and variance explanation. For firms doing substantive close review, AI-assisted variance analysis can surface the accounts with material changes and propose explanations based on transaction detail. This compresses the analytical step of the review without replacing the accountant's judgment on whether the explanation is correct.

What AI is not doing reliably: making accounting judgments on ambiguous transactions, catching errors that require business context the system does not have, or replacing the final review. The automation handles the volume. The accountant handles the judgment.

The Rollout Sequence

If you are adding a close management platform to your practice, here is a realistic sequencing for a four to six week pilot with one client before expanding.

Weeks 1 and 2: Select a pilot client with moderate complexity, not your most straightforward file and not your most complicated. Build the per-client close checklist for that client in a document before touching the platform. Connect the platform to the client's accounting file. Configure the standard checklist and the client-specific rules. Do not configure every possible rule on day one. Start with the five to ten items that consume the most manual time on that file.
Weeks 3 and 4: Run the first full close inside the platform. Run your prior process in parallel on paper so you can compare what the platform catches versus what you would have caught manually. Document every exception that required a judgment call. These become your rule refinements for month two.
Weeks 5 and 6: Measure the time. How long did the review step take compared to prior months? How many client questions were resolved through the portal versus email? How many items in the exception queue were genuine issues versus noise from over-configured rules? Use these numbers to calibrate before expanding to additional clients.
The most common failure mode in close tool rollouts is over-configuring on day one. Every rule that is too aggressive generates false positives that erode staff trust in the queue. Start narrow, validate, expand. The configuration layer improves over three to four close cycles and then stabilizes.

The Owner Takeaway

The close is your quality control checkpoint. It is the moment in the month when errors get caught, client questions get answered, and the books get signed off. Automating the mechanics of the close does not reduce its importance. It increases the time your team has available for the review, which is where errors actually get caught and where the firm's judgment is irreplaceable.

A close management platform is also a scalability investment. If your firm's capacity to take on additional clients is constrained by the manual overhead of the close, the right tool reduces that constraint. Firms that have deployed Double across a mature client portfolio report closing the same number of client files in meaningfully less calendar time, which compresses close week and frees staff for client-facing work in the weeks that follow.

The Operator Takeaway

Build one standard close checklist before you configure anything. Not one per client: one firm-wide standard that becomes the template, with client-specific variations documented on top of it. The tool makes the checklist operational. It does not replace the thinking required to build it.

The second thing: resist the urge to configure every possible rule in week one. The exception queue only works as a productivity tool if staff trust that items in it are genuine issues. Over-configured queues become queues that staff learn to dismiss, and that defeats the purpose. Start with the highest-volume, highest-confidence rules. Add complexity after the first two or three close cycles confirm the baseline is working.

What Comes Next in the Series

Post 4 has covered the close tool layer: the platforms, the workflow, and the AI functions worth trusting. Post 5 goes one level up to the process design question: how do you build a repeatable close standard across every client file when each client is different, and how do you sequence closes across a full client portfolio without creating bottlenecks at the end of every month?

Back to Post 1: The AI-Enabled Close | Post 2: AP on Autopilot | Post 3: AR Automation

Resources

The Lumiere AI Stack Map for Outsourced Accounting Firms (PDF download) -- the four-layer visual referenced throughout this series
Two-week time study template -- track where your firm's close hours are actually going before you automate anything