Basis prepares accounting work for professional review

Basis develops AI agents for accounting work that normally involves collecting source material, preparing calculations and assembling an inspectable result for review. Its current offer spans client accounting services, tax, audit and corporate accounting. The intended shift is from starting with an empty workpaper to starting with prepared work and exceptions. Professional judgment remains central: an agent-generated result still needs an accountable person to decide whether its evidence and treatment are appropriate.

01 / ProductA preparation layer across several accounting practices

The Basis overview describes separate workflows for client accounting and advisory, tax, audit and corporate accounting. The common thread is preparation that can be inspected: source material, calculations, explanations and exceptions are brought together for the professional reviewer. The breadth of the offer makes it important to identify the exact practice and procedure being evaluated rather than asking whether an undifferentiated accounting agent works.

For client accounting services, Basis describes transaction coding, reconciliation and reporting using client-specific policies and firm standards. Its material says high-confidence transactions can be pushed automatically while lower-confidence items are flagged. That is a consequential operating choice. A buyer should establish which actions are merely prepared and which can affect the ledger under the selected configuration.

The tax workflow covers intake, workpaper preparation and review support, while the audit page presents evidence-backed execution of audit work. These are distinct professional processes. A successful demonstration in transaction coding does not establish suitability for a tax return or an audit procedure with different evidence and sign-off requirements.

02 / AudienceUseful where preparation standards are already understandable

A good starting point is recurring work with a defined input, a known output and an experienced reviewer. For a small firm, that might be a reconciled account package or a draft journal entry tied to an invoice. For a larger practice, it might be a repeatable procedure whose documentation standards vary between offices. The product can be evaluated against those concrete expectations.

Compare DataSnipper when the principal challenge is working with evidence inside document and spreadsheet-based processes. Compare BlackLine when the central requirement is structured financial close and control across an enterprise. Basis’s agent-led preparation should be assessed in relation to the actual work that remains with people, rather than assuming these products have interchangeable scopes.

A firm with inconsistent client files and undocumented review preferences may still benefit, but it should budget for making those procedures explicit. Otherwise the pilot risks judging the agent against standards that differ between reviewers. Establish a small shared reference set, including an example that should be escalated, before using the product across a broad client portfolio.

03 / WorkflowProposed workflow: prepare one client’s reconciliation package

Choose a fictional client or a carefully authorised test dataset with a bank statement, a ledger extract and a known unresolved item. Define the intended output as a draft reconciliation package with supporting references, proposed adjustments and an exception list. Preserve the approved reference result so the reviewer can distinguish a genuinely useful inference from a plausible but unsupported explanation.

The small-firm product page describes procedures, client organisation and proposed journal entries with linked source documents. Use one named procedure and one client workspace in the pilot. Give the agent the relevant account mapping and review preferences, then inspect whether it applies them consistently. Similar client names or document filenames should not cause evidence to cross between engagements.

Include a payment that appears in the statement after the ledger cut-off and an invoice with incomplete coding information. The expected result should preserve those different explanations rather than force both into an automatic adjustment. A useful reconciliation distinguishes a timing item from an error requiring a new entry. Ask the reviewer to explain why each proposed treatment is acceptable.

Inspect every proposed journal entry before posting. Confirm the account, period, amount, narrative and source reference, and establish what the configuration allows to happen without a fresh approval. The CAS page’s automatic handling of high-confidence transactions and the small-firm page’s staging language should be resolved for the exact procedure in use. Confidence labels should not replace the firm’s own criteria for materiality and approval.

Have the reviewer reject one deliberate misclassification and correct the procedure or client context. Repeat the task with a related but different transaction. The second run should demonstrate whether the correction is applied in the intended scope, without creating a rule that incorrectly affects another account. Retaining feedback is helpful only when the team understands what the system has learned to do next.

Prepare the final review package and trace a reported balance back through the proposed adjustments to the original documents. Record any missing source, unsupported assumption or manual workaround. The pilot’s output should remain understandable to another accountant who did not watch the agent work. An inspectable result has more durable value than a successful demonstration that depends on the presenter’s explanation.

Evaluate total engagement effort, including the time needed to resolve exceptions and organise source records. Do not measure only the speed of the first draft. The relevant benefit is whether a reviewer reaches a defensible conclusion with less repetitive preparation, while retaining control over decisions that affect the client. No part of this proposed evaluation was performed for this article.

04 / PricingAccess and pricing are determined by the current order

The public site offers demo and access requests and also exposes waitlist wording. The product terms, revised June 2026, describe business-use subscriptions whose plan, price, billing cadence and term are specified in an order or checkout. No current numerical public tariff was verified. Confirm that the requested procedure is available to the intended firm before planning a live engagement around it.

The terms place responsibility for reviewing output with the customer and distinguish software output from professional advice. This is consistent with evaluating Basis as a preparation tool within an accountable practice. Obtain the actual agreement rather than inferring engagement suitability from a marketing description or a customer quote about hours saved.

For budgeting, identify the work that remains outside the subscription: preparing client permissions, connecting systems, defining procedures, training reviewers and resolving unusual evidence. If a trial is offered, agree a bounded dataset and success criteria before access begins. That makes it possible to decide whether the product changes the economics of a recurring procedure instead of merely producing an impressive first example.

05 / DistinctionsClient context and procedures can make the agent more specific

Basis’s small-firm material describes reusable procedures and separate client context. This is a meaningful direction because accounting work is rarely determined by a source document alone. The same invoice description can have different treatment depending on a client’s chart of accounts, policy and prior decisions. The useful evaluation question is whether that context is visible, correctable and applied only where appropriate.

The tax and audit offerings also highlight the importance of evidence-linked preparation. An agent that supplies a number without its basis transfers work to the reviewer, who must reconstruct the reasoning. A draft that clearly distinguishes sourced values, calculations and unresolved assumptions can be more useful even when it completes less automatically. Judge the product by the quality of the professional handoff.

06 / QuestionsResolve data location and automatic action boundaries

The security page states that client data is hosted in US regions and describes isolation, auditability and a no-model-training policy. It also labels GDPR and CCPA as coming soon. That wording should not be turned into a blanket statement of global readiness. Organisations with specific residency or contractual requirements need to resolve the applicable arrangement directly.

The product pages describe both staged review and automatic handling in particular workflows. Ask the vendor to demonstrate the selected procedure’s actual approval configuration, including what happens after an exception is corrected. Keep a record of which actions can reach the ledger, who can change that configuration and how the resulting work is reviewed. These are product-specific questions about authority, not reasons to dismiss all agent-assisted preparation.

07 / DecisionSelect a procedure whose review outcome can be judged