OneStream

Full profile

This profile covers OneStream’s offerings for consolidation, account reconciliation, transaction matching, journal entries and close task management, and the AI features OneStream documents for them. It does not cover OneStream’s planning products except where an AI feature spans both.

Which AI this covers: AI built into OneStream’s finance platform, as OneStream documents it: generative features, agents and established machine learning or matching automation. It does not cover general-purpose assistants a team might use alongside OneStream. See What We Mean by AI.

At a glance

OneStream’s April 1, 2026 press release describes its Digital Finance Cloud as a platform that unifies core financial and broader operational data and processes and embeds AI for planning and forecasting, and states that it has over 1,800 customers, including 18% of the Fortune 500. Its close packaging bundles OneStream Financial Close, Account Reconciliations, Journal Entry Manager, Task Manager and Transaction Matching, and its consolidation packaging includes currency translation, eliminations and equity pickup. OneStream states that its acquisition by Hg was completed on April 1, 2026, with General Atlantic and Tidemark as minority investors, and that it is now privately held.

Vendor-stated capabilities

Close and consolidation

• Consolidation. OneStream’s Solution Description Guide states that Consolidation – Standard includes financial model consolidation, with currency translation, direct and indirect eliminations and equity pickup, and entity aggregation.
• Close applications. The same guide lists the applications included with Close – Standard: OneStream Financial Close, Account Reconciliations, Journal Entry Manager, Task Manager, Transaction Matching, Reporting Compliance, Contract Compliance, an SAP connector and a BAI parser.
• Approver access. A separate Close – Approver package is read-only except for approving tasks, reconciliations and journal entries.
• Reconciliation anomaly detection. OneStream’s documentation lists anomaly types that include unusual reconciliation balances, sparse supporting documents, shifts between automatic and manual reconciliation, repeated rejections, sign changes and changes in the number of approvers required. A release dated August 3, 2026 added four detectors that work at the level of individual reconciliation detail items, such as aged items without commentary and manual items without supporting documents. One of them flags items aged more than 60 days without a comment or reference, and items past 120 days without an updated comment; another flags item aging that stopped, moved backwards or increased by more than 31 days period over period, or that cannot be traced to prior-period history. A later release dated September 1, 2026 added a reconciliation Frequency field for rules in that second detector; existing anomaly arenas are not affected; to use the new field, the customer exports the rule configuration and resets the arena.
• Certification gate. When an administrator enables the Critical Anomaly Resolution Required setting and certification is active, preparers and approvers must address critical unresolved anomalies before a reconciliation can be certified; when the setting is not enabled and certification is active, they can certify, including in mass actions, and those anomalies are marked Acknowledged.
• Anomaly record. The documentation states that each anomaly carries a status (Unresolved, Reviewed, Auto-Resolved or Acknowledged), that the user and date of each run, comment or status change are recorded, and that a reconciliation’s Has Anomalies flag stays set for the period even after the anomalies are resolved.

AI features

The kind of AI and the suggests-or-acts reading in this table are CoE classifications based on OneStream’s descriptions. The "who controls it" column reports only what OneStream documents.

OneStream’s marketing page lists four agents as now available, including a Forecast Agent offered only to SensibleAI Forecast customers. The Solution Description Guide states that the SensibleAI portfolio uses several large language models available through Microsoft Foundry, including models from OpenAI and Anthropic.

On control and data use, OneStream states that the agent gateway applies the user’s role-based permissions so an agent cannot reach data beyond the user’s access, that activity is checked before and after execution and logged, and that these controls apply to third-party assistants connected through the MCP server. Its marketing page states that existing OneStream administrators control what agents can access, that agents execute deterministic calculations on OneStream’s financial engines, and that customer financial data is not used to train external AI models. The same page states that agent queries, workflow executions and outputs are logged with an audit trail, for OneStream’s agents and connected third-party agents alike. The guide also records OneStream’s own classification of its SensibleAI solutions as Limited Risk under the EU AI Act; that is a self-classification, not a regulator’s determination.

What the company’s SEC filings disclose

The filings described below are OneStream’s own statements, made under securities law; they are not third-party assessments.

Listing and reporting status. OneStream is no longer an SEC-reporting company. It filed a Form 15 dated April 13, 2026, to terminate the registration of its Class A common stock and suspend its duty to file reports, stating that on April 1, 2026 it became a subsidiary of Onward AcquireCo, Inc. under a merger agreement dated January 6, 2026. Its last annual report, the Form 10-K for fiscal 2025, speaks as of February 2026 and predates the acquisition.

What the last annual report says. The 10-K describes the platform as enhanced by native machine learning and generative and agentic AI for finance, names financial close and consolidation as a core solution, and states that SensibleAI Studio routines and SensibleAI Agents became commercially available in 2025. It names Oracle, SAP, Anaplan, BlackLine, Wolters Kluwer and Workday among its primary competitors. Three risk factors bear directly on customers:

• AI output. OneStream disclosed that its AI use could expose it to claims if output its AI helps produce is, or is alleged to be, deficient, inaccurate or biased, and that evolving AI law, including the EU AI Act, could require it to change or restrict its use of AI.
• Errors in close functions. It noted that customers use the platform for critical functions such as the financial close and account reconciliation, that errors could damage them, and that its customer agreements typically contain provisions designed to limit its exposure, though it added that existing or future laws or unfavorable judicial decisions could negate those limits.
• Availability at peak. It reported that most customers run on Microsoft-operated Azure infrastructure, that it has experienced disruptions and outages, and that the seasonal pattern of financial close makes peak capacity hard to predict.

Independent market evidence

Analyst placements and regulators. This profile includes analyst placements only where the analyst’s own public page can be read, and none could be read. No regulator or standards-body action concerning OneStream’s AI features was found in the sources reviewed.

Independently verified outcomes

No independently verified outcomes were found. CoE | Vendor Intelligence searched OneStream’s securities filings, its documentation, release notes and product pages, and the web for independent studies or audits of its reconciliation anomaly detection and agents, and found no outcome measured or attested by a party independent of OneStream.

The outcome figures OneStream publishes are vendor-stated, and CoE | Vendor Intelligence has not verified them. Its April 1, 2026 press release quotes its founder and chief executive as saying that OneStream more than doubled its AI customers year over year in 2025, without defining AI customers or stating a baseline.

Pricing and packaging

• List prices. OneStream publishes no prices: its agentic finance page directs readers to request a demo, and its sitemap lists no pricing page. On this evidence, OneStream’s pricing is available by quotation, so there is no list price to compare with negotiated terms.
• Packaging. OneStream’s Solution Description Guide treats each solution as a separate SKU that can be bundled in the customer’s order. Its 2025 annual report said it had begun letting customers buy products individually or in bundles, having historically sold mainly on a platform and seat-based model, and that it had limited experience pricing its AI technologies. The same report gave the typical initial subscription term as three years, within a range of less than one year to ten years.
• AI consumption units. SensibleAI Agents are metered in AI Credits, consumed per invocation (including model calls, evaluation runs and document ingestion), and consumption varies with question complexity, the data processed and the third-party AI provider used. SensibleAI Studio Essentials and Standard each carry quarterly limits of 3,000 language-model chat invocations (surfaced through Studio and Xperiflow business rules) and 300,000 embedding generations, with no rollover.
• Matching volume. Transaction Matching included with the Close packages is limited to 100,000 transactions per month across the service, with additional entitlements for additional fees, under the Solution Description Guide of August 19, 2026.

Known limitations

• Licensing prerequisites. Reconciliation anomaly detection requires SensibleAI Studio, which OneStream prices separately, and version 9.1 or later. SensibleAI Agents require Platform 9.3 or later, run in one production environment only, and may be used only with the prerequisite Digital Finance Cloud solutions by users who hold permissions for them.
• Statistical sensitivity. OneStream notes that statistics-based detectors with fewer lookback periods have less statistical power, so smaller anomalies may not trigger a flag. It recommends limiting anomaly detection to 15,000 to 20,000 reconciliations per run.
• Defaults change in releases. The August 3, 2026 release disabled four reconciliation-level detectors by default because new detail-item detectors supersede them; they can be re-enabled.
• Rule changes take effect immediately. Detector and rule changes apply at the next run without a save step, rules cannot be edited (only added, copied, deleted or toggled), and re-running detection after a rule is removed or switched off sets its anomalies to auto-resolved.
• Data retention. SensibleAI solutions do not provide long-term retention; production storage supports point-in-time recovery for a trailing 30 days, and test-environment backups are kept for at least seven calendar days. By contrast, the guide says production Platform databases have weekly snapshots kept for a trailing 52 weeks, with longer periods available for an additional fee up to 520 weeks.
• Model providers and sub-processors. OneStream’s personal-data sub-processor list, dated February 2026, names no AI model provider; it lists Microsoft for hosting in the United States, Ireland, the Netherlands, the United Kingdom, Australia and Singapore, and six other entities for email, performance monitoring, access management, support and incident management. The Solution Description Guide names OpenAI and Anthropic as providers of models available through Microsoft Foundry. Apart from the Singapore note below, the documents read do not say where those models process customer prompts or data.
• Sub-processor changes and data terms. OneStream’s Data Processing Terms of June 17, 2025 say it will inform customers of intended sub-processor additions or replacements, and that a new sub-processor becomes authorized if the customer does not object within 30 days of notice. They commit to breach notice without undue delay and within 48 hours, set the duration of processing as the service term plus up to 180 days, and offer deletion or return of personal data at the customer’s option after the service ends, unless the law requires OneStream to keep a copy. These terms cover personal data that OneStream processes for the customer and do not refer to AI features or model training, so they should not be read as AI-specific terms.
• Third-party models. Large language models connected through OneStream’s MCP server are outside OneStream’s service, governed by the third party’s terms, and data shared with them is at the customer’s discretion.
• Hosting region. Customers are hosted by default in the United States, the European Economic Area and United Kingdom, or Australia; for the Singapore region, OneStream states that AI inputs and outputs, among other data, may be processed outside Singapore.
• A non-human account. SensibleAI Agents add a service account, named wernicke.server, that looks like a user and runs scheduled jobs.
• Beta feature. Agent Memory is in beta.

CoE analysis

The useful distinction in OneStream’s close AI is between features that flag and features that write. Reconciliation anomaly detection suggests: it flags patterns for a preparer or approver to judge. The Finance Analyst agent can act, by writing commentary into the financial model after a person confirms each proposed change, and it can run on a schedule. A finance team should establish, before relying on either, the following.

• What is switched on, and what changed. Default detectors and rules arrive with the product, and a vendor release can change which are enabled or add new rule fields. Keep a dated record of the detector and rule configuration that applied each period, so a reviewer can tell which checks ran. See What We Mean by AI and the close and consolidation playbook.
• Whether a flag can stop a certification. Unless the critical-anomaly setting is enabled, a reconciliation under active certification can be certified, even in bulk, with critical anomalies simply marked Acknowledged. An approval in that state is not evidence that anyone examined the anomaly. See Review Is a Control Only When It Can Catch the Error.
• What the detectors cannot see. The detectors compare a reconciliation with its own history and attributes. They do not establish that every account, entry or supporting document that should exist does exist. See Treat AI Output Like an Unsupported Journal Entry.
• Who can change the checks. Administrators can disable detectors and rules, changes take effect without a save step, and switching a rule off can auto-resolve its earlier anomalies on the next run. Treat detector configuration as a change-managed control setting, with approval and logging, rather than an administrator preference. See Internal Control Over Generative AI, Applied to the Close.
• What the agents are authorized to do. Commentary write-back, scheduled analysis plans and a service account that looks like a user each belong in the authority register: who approved the agent, what it may write, under whose permissions, and how the limits are evidenced. See Your AI Agent Is an Unlisted Signer.
• Which outputs are generated. OneStream’s statement that agents execute deterministic calculations describes the arithmetic on its engines; summaries, explanations and commentary are produced by language models and can vary from one run to the next. Review generated commentary as a statement made to readers of the numbers, not as part of the calculation.
• Where data goes. Connecting an outside assistant through the MCP server brings that assistant’s provider and terms into the data flow, outside OneStream’s contract. Map those flows before enabling them, and note that the 30-day sub-processor objection window in the data processing terms covers OneStream’s own sub-processors, not providers a customer connects itself.
• What you can still learn about the vendor. With OneStream now privately held, quarterly SEC reporting is no longer available to customers assessing the vendor’s financial position and risk disclosures. Request that information directly during diligence and renewal.

Community Observations

Not yet published. Community Observations are added after a licensing review.

Member input under review

None yet.

Relevant diligence questions

1. Which SensibleAI features are licensed in our contract, which are enabled in our production environment, and which detectors and rules are active by default after each release?
2. Can the Critical Anomaly Resolution Required setting be enforced for our in-scope reconciliations, and what evidence shows who changed detector, rule or certification settings and when?
3. For Finance Analyst write-back and scheduled analysis plans, which identity and permissions does the agent use, what can it write, who approves a write proposed by a scheduled plan, and how is each write a user approves evidenced for audit?
4. Do OneStream’s assurance reports, such as SOC 1 and SOC 2, cover SensibleAI Agents, the Agentic Gateway and the AI anomaly detection, and for which periods?
5. Which language models serve our environment, where do they process our prompts and data, how are model changes notified, and how do AI Credit consumption and quarterly limits translate into cost and overage terms?
6. What contractual commitments govern training on our data, retention of prompts and outputs, and data shared with third-party assistants connected through the MCP server?

Evidence gaps and unanswered questions

• AI contract terms. OneStream’s commercial terms archive lists an AI Terms document, but the copy available is an image-based PDF that could not be read as text, so its contents are not described here. The position that customer data is not used to train external models comes from a marketing page, not from contract text read for this profile.
• Current terms. The Solution Description Guide, Data Processing Terms and sub-processor list used here were found on a page titled Archive; no current commercial terms page was located, and the version that governs a given contract depends on the customer’s order.
• Assurance coverage. OneStream’s trust center is rendered by script and could not be read as text, so whether its assurance reports cover the AI features is unknown.
• Packaging of the reconciliation AI. The Solution Description Guide does not list reconciliation anomaly detection as a separate SKU; the documentation ties the feature to SensibleAI Studio, so which order line licenses it should be confirmed with OneStream.
• Release dates. OneStream’s documentation release notes for the Financial Close solution and SensibleAI Agents do not show release dates.
• Detection performance. No detection rates, false-positive rates or independent tests of the anomaly detectors or agents were found.
• Analyst placements. OneStream’s website cites analyst recognition; the analysts’ own reports could not be read publicly, so no placement appears here.

Right of reply

Organizations referenced on this page may respond under our Editorial Independence & Right of Reply policy.

Review status

This profile was last reviewed on October 12, 2026. Vendor-stated information is rechecked monthly. It is not legal, accounting or procurement advice.