Uncat vs Keeper vs AI Agents for Client Query Lists

By Jude Lee · · Comparison

Two bookkeepers reviewing a list of uncategorized transactions on a monitor in an accounting firm office

The workflow nobody prices correctly

Every month, a bookkeeper hits transactions they cannot code with confidence: a $1,842 charge to a vendor name that means nothing, a transfer that might be an owner draw, a Home Depot run that could be repairs or a capitalizable improvement. Those become the open-items list — the query list, the “ask client” list, whatever your firm calls it.

Four distinct jobs hide inside that list:

  1. Detect the item that needs a human answer (vs. one a rule could have coded).
  2. Draft a question the client can actually answer — with date, amount, payee, and a plain-English ask.
  3. Chase the answer until it arrives.
  4. Apply the answer: code it, memo it, and ideally create a rule so it never comes back.

Most firms only measure step 3, because that’s the part that feels like nagging. Steps 2 and 4 plausibly consume more staff time, and they’re the steps the tooling market has covered least well.

What Uncat and Keeper actually do

Uncat is narrow by design: per its public product materials, it connects to your ledger, surfaces transactions sitting in the uncategorized (or a designated) account, presents them to the client in a simple list, collects the explanation, and pushes the answer back toward the ledger, with automated reminders. It’s cheap to adopt and easy to explain to a client.

Keeper is positioned more broadly: client queries sit inside a wider month-end close workflow alongside close checklists, ledger review flags, and reporting. If your close is already scattered across spreadsheets and email, the query list is one of several things it consolidates.

Treat both descriptions above as directional, not as a verified feature matrix. Supported ledgers, AI-assisted suggestion features, and pricing all move fast — check each vendor’s current documentation before you buy, rather than trusting any article, including this one, as of 2026.

A third option gets skipped too often: your ledger’s own bank rules. If a large share of your query list is the same handful of recurring payees per client, no AI is needed. Write the rule. We’ve argued this at length in bank rules vs offshore staff vs AI agents — deterministic beats probabilistic whenever the logic is stable.

Where a firm-built AI agent fits

An AI agent here is not a chatbot. It’s an assistant (Claude or a comparable model) given governed, read-mostly access to your ledger through MCP — the Model Context Protocol, an open standard for connecting an AI assistant to real systems and tools. Wired to QuickBooks Online or Xero, the agent can pull the transaction, look at prior coding for that payee, check the class/location pattern, and draft a question that already contains the firm’s best guess.

Instead of “What was this $1,842 charge on 3/14?”, the client gets: “On 3/14 you paid $1,842 to Ferguson Enterprises. The last four payments to this vendor were coded to Repairs & Maintenance for the Elm St. property. Same thing, or is this for the new build?”

That difference — a yes/no question instead of an open-ended one — is where most of the cycle-time savings live, in our view.

The second half is a skill: a packaged, reusable instruction set that teaches the assistant to handle this job the same way every time. Your query-list skill would encode escalation rules (anything over $X or touching fixed assets goes to the senior, not the client), client tone, memo format, and the requirement that every proposed coding cite the evidence behind it. Skills are how you stop getting different-quality output depending on who prompted it. The same pattern applies to workpaper prep from a trial balance.

Uncat / Keeper
Fast to deploy, predictable, client-friendly UI, vendor absorbs sync maintenance and reminders. Limited ability to reason about why a transaction is ambiguous or to propose coding with supporting history. You adapt to their workflow, and you pay per seat or client indefinitely.
Firm-built AI agent
Drafts specific questions with ledger context, proposes coding and memos, can suggest a bank rule when a pattern repeats. Requires build effort, access governance, a review gate — and ongoing ownership: ledger API or schema changes can break the MCP connection, skills drift as staff edit them, and someone has to re-test after every QBO or Xero update.

These aren’t mutually exclusive. A realistic hybrid: keep Uncat or Keeper as the client-facing intake and reminder layer, and run the agent behind it to draft the questions and pre-code the returned answers for staff review.

Building the agent version, step by step

  1. Connect with least privilege

    Stand up an MCP connection scoped to read transactions, accounts, and vendor history — plus, at most, draft or staged-queue rights. Do not give an agent unreviewed posting rights to a live client ledger. Our walkthrough on connecting an AI assistant to QuickBooks or Xero via MCP covers the scoping decisions.

  2. Define detection in code, not in the model

    Query the uncategorized account, unreconciled items, and any transaction above a threshold with no memo. This is deterministic work. Let the agent start from a clean list rather than deciding what belongs on it.

  3. Write the query-drafting skill

    Specify: pull payee history, state the firm’s best guess, ask a closed question, and flag fixed assets, owner transactions, payroll, or sales tax for internal review instead of client-facing questions.

  4. Route answers through a review gate

    The agent proposes the coding and memo; a human approves before anything posts — see building AI review gates rather than autopilot.

  5. Name an owner and a re-test cadence

    Assign one person to own skill definitions, connection health, and a regression check after ledger platform updates. Budget that time; it is the real recurring cost of the build path.

  6. Close the loop with rules

    When the same payee gets the same answer repeatedly, the agent should propose a bank rule. The goal is a query list that shrinks month over month.

What “automation in accounting” really means at this layer

Automation in accounting means replacing repeated manual steps — data entry, matching, categorization, reminders, report assembly — with software that performs them without a person. In the query-list workflow the examples are concrete: bank rules that auto-code recurring payees, a sync that pulls uncategorized items into a client-facing list, automated reminders, receipt capture that attaches a source document before anyone has to ask (see Dext vs Hubdoc vs AI agents), and now agents that draft the question and propose the answer.

The distinction worth holding onto: rules automate decisions you’ve already made. Agents help with decisions that require reading context. Use the cheapest tool that fits the decision type.

What this does and doesn’t replace

AI-native bookkeeping startups and practice-management vendors are both shipping agent features aimed at exactly this busywork, and some of that marketing is framed around replacing accountants outright.

Our read, stated as opinion: the query list is exactly the kind of work that gets compressed, and firms whose CAS margin depends on billing hours for chasing clients should expect pressure. What doesn’t compress is the part where someone signs their name to the financials, decides whether a $1,842 charge is repairs or a capital improvement, and answers for it later. Licensure carries responsibilities a model cannot hold. The realistic near-term outcome is fewer hours per client file and more client files per person — not an empty chair.

The bottleneck was never sending the question. It was writing a question the client could answer in one line — and knowing what to do with the answer.

A cost model you can fill in

Don’t trust anyone’s headline savings number, including ours. Build your own:

Q × T ÷ 60
WORKSHEET INPUT — hours per client per month; plug in your own counts
4 steps
WORKSHEET INPUT — detect, draft, chase, apply; price each separately
100% review
POLICY CHOICE — share of agent-proposed postings we'd send to a human gate at launch

These are inputs and a policy stance, not findings. The upside isn’t only cost: recovered hours reallocated to advisory or additional client capacity is the larger line for most firms, and a query list that closes earlier pulls the whole close forward.

How to choose

Pick Uncat if the query list is your only pain point, you want it solved this week, and your clients need the simplest possible interface. Pick Keeper if the query list is one symptom of an unstructured close and you want checklists, review flags, and reporting in one place. Build an agent layer if you’ve already tightened the process, your remaining cost is in writing good questions and coding returned answers across many clients, and you have someone who will own the skill definitions, the review gate, and the maintenance. And if a large share of your list is a handful of recurring vendors per client — write the bank rules first and re-measure before buying anything.

Not sure where to start?

Get a free automation audit: we map your bookkeeping, month-end close, client onboarding, document collection, and AP/AR — and show you what's worth automating before you spend a dollar.

Get a free automation audit

Related guides