
Concourse is an AI agent platform that reads a finance team's ERP, warehouse, billing and CRM data and drafts variance commentary, forecasts, close reviews and board decks, sold to startups and enterprise finance teams with a forward-deployed build team.
Independently scored, one dimension at a time. Every dimension is sourced, and we say whether a claim is published, reported, or estimated.
Written by the Audit Friendly research team. No vendor edits, no sponsored placement.
Concourse sells AI agents that sit on top of the systems a finance team already runs. Read-only connectors pull from NetSuite, QuickBooks Online, Sage Intacct, SAP, Snowflake, Stripe, Ramp and about two dozen more. Agents compute in code, show the SQL and Python behind each number, and deliver variance commentary, forecasts and PowerPoint or Word decks on a schedule or a trigger. Enterprise deals come with ex-CFOs and forward-deployed engineers who build the first agents onsite.
The catch is that it is an analysis layer, so it does not replace a planning tool. Its pages document no driver-based model, no budget versions or approval workflow, no headcount module and no Excel add-in. Every integration page says access is read-only, there is no public API or data MCP, and the Starter and Professional self-serve plans described in early 2026 no longer appear on concourse.ai, which now quotes each deal. The company was founded in 2023, and most of its headline outcomes, such as a 75 percent cut in manual work, are vendor-claimed.
Before signing, get in writing the price unit (seats, usage or credits), whether onsite forward-deployed work is inside the subscription, and the renewal terms, since the online agreement leaves fees and term to the order form. Ask for the SOC 1 Type II report the security page shows as a badge, alongside the SOC 2 Type II report. Confirm how you export agents, methodology rules and outputs at exit, and ask for two named references on your ERP.
Jobs, mapped to how finance teams actually buy.
Forget the 40-row feature grid. Pick a finance workflow and see how this tool and its closest peers actually do it, with our verdict on who wins for whom.
Here's what it really costs: published rates where a vendor lists them, aggregated quote data where they don't, plus a path to your own number.
Scored on what an outside agent can actually reach: the documented API, how it authenticates, and whether the vendor ships an MCP server.
Exists: Concourse documents no API and no MCP server for workspace data. The Concourse 101 help site at docs.concourse.co exposes a hosted docs MCP endpoint, and the concourseco GitHub organization holds a fork of the QuickBooks Online MCP server rather than a Concourse server (checked 1 November 2026).
Scope: The docs MCP has three tools: two read-only tools that search and read the help pages, and one that submits feedback on the docs. It cannot reach agents, workbooks, connected data or reports. The official MCP registry lists no Concourse server (checked 1 November 2026), and GitHub results for Concourse MCP belong to the unrelated Concourse CI project.
Turnkey: There is no data surface for an outside agent to connect to, so nothing to configure. Work leaves Concourse as email summaries, Slack and Teams messages, SharePoint files, PowerPoint and Word decks, and CSV exports.
In-product AI routes each agent step to a frontier model, runs calculations in deterministic code, and shows the SQL and Python behind every answer, per the technology page and the January 2026 launch post. Agents run on schedules, triggers or events and are tested against custom evals. The vendor claims 99 percent or better accuracy and up to a 75 percent cut in manual work; no independent source confirms either figure.
Implementation
The integrations FAQ says most connectors link in under 10 minutes by OAuth or API keys with no IT ticket, and the deployment FAQ says most teams are live in weeks. Case studies describe CurbWaste on QuickBooks Online live within hours and Lightmatter connecting NetSuite through the SuiteApp in under ten minutes. There is no model to build, but useful agents still depend on the team defining metrics, sources and review steps.
Support
Every deployment pairs the customer with ex-CFOs and forward-deployed AI engineers who come onsite, run change management and maintain agents as data and models change (deployment page and FAQ, checked 1 November 2026). Support otherwise runs by email from the docs site. No SLA, response times or paid support tiers are published, and no independent review-site rating is cited here.
ERP & data connectivity
The sitemap lists 30 integration pages: ERPs (NetSuite, Sage Intacct, QuickBooks Online, Xero, SAP, Oracle, Workday, Rillet), warehouses, Stripe, Maxio, Ramp, Brex, Rippling, Gusto, HubSpot and Salesforce, plus CSV (checked 1 November 2026). Every integration FAQ says access is read-only and nothing is written back. There is no Business Central page, and bank connectivity is described in marketing copy without a documented bank connector list.
Compliance depth
The security FAQ says the SOC 2 Type II examination by Schellman covered more than twelve months across all five Trust Services Criteria, and the security page also shows a SOC 1 Type II badge, SAML SSO, MFA and audit trails (checked 1 November 2026). Roles are scoped at the agent and dataset level, and every agent action and sign-off is logged. Because there are no budgets or plan versions, there is no approval workflow or plan snapshot to audit, and the full SOC reports sit behind an NDA in a Vanta trust center.
Scalability
No limits on users, rows or entities are published. Connectors cover Snowflake, BigQuery, Databricks, Fabric, Redshift and Synapse, and the integrations FAQ says multiple QuickBooks companies, NetSuite subsidiaries or Salesforce orgs can be connected and analyzed together. The vendor names Palo Alto Networks as a customer and claims 40 billion dollars or more in treasury operations managed, which no independent source confirms. YC lists a team of 15, so capacity for many large deployments at once is an open question.
AI / agent readiness
In-product AI is the whole product: model routing per step, deterministic math in code, triggers and schedules, custom evals and versioned company methodology, per the technology page and an August 2026 engineering post. The external surface is thin. No public API or data MCP is documented, and the only MCP endpoint, on docs.concourse.co, offers three tools that search and read the Concourse 101 help pages or send docs feedback (tool list read 1 November 2026).
Modeling depth
Agents build rolling forecasts, run scenarios and compare actuals to plan from live data, and the FP&A page says they learn the team's forecast logic and metric definitions. There is no driver-based planning engine, no budget or forecast versions with approvals, no workforce planning module and no Excel or Sheets add-in. Teams keep their plan of record elsewhere and point Concourse at it.
Pricing transparency
There is no pricing page; the FAQ says price depends on the systems connected, the workflows automated and team size (checked 1 November 2026). Earlier in 2026 the launch post promised usage-based pricing and MakerStack reported $99 and $499 monthly plans on 31 August 2026, but neither appears on concourse.ai now. No Vendr marketplace page covers Concourse, and the online agreement leaves fees and renewal terms to each order form.
A Flux and Variance agent pulls actuals, quantifies what moved and drafts commentary before the review meeting. CurbWaste, on QuickBooks Online, Ramp, Stripe and HubSpot, reports 15 or more hours saved a month in the vendor's case study. The numbers are the vendor's and the customer's, not independently audited.
Agents query several connected systems in one request, for example reconciling ARR between Stripe and Salesforce, per the integrations FAQ. Each answer shows the SQL and Python behind it. Data stays read-only in the source systems.
Agents generate charts and summaries from live data and export PowerPoint and Word files built on the team's templates. The launch post claims presentations that took five days now take under an hour. Review every slide before it goes to a board.
Agents read transaction memos and descriptions, flag miscategorized entries and suggest new categories. Lightmatter's case study says the NetSuite SuiteApp pulled over 300 fields, including custom records. Recategorizations still have to be posted in the ERP, since the connector is read-only.
The FP&A page says agents build rolling forecasts, run scenarios and update the numbers as actuals arrive, reusing the team's forecast logic each cycle. The 50 percent accuracy figure has no named customer or method behind it. The forecast lives in Concourse outputs, with no versioned planning model behind it.
AI Isn't an Upgrade: It's an Order of Magnitude Shift #saas #podcast #shorts #ai #concourse
A Short from The SaaS CFO podcast with the Concourse CEO arguing that AI agents change finance work by an order of magnitude. It shows how the vendor frames its pitch to finance leaders. Posted 26 March 2026, about 70 views. It is opinion from the founder, not a demo. Transcript not read.
19x Revenue Growth, and Fortune 500 Inbound #saas #podcast #ai #concourse #fortune500
A Short cut from The SaaS CFO podcast in which the Concourse CEO talks about growth and inbound interest from Fortune 500 companies. It gives context on who is buying. Posted 24 March 2026, about 100 views. It is a founder talking about traction, so treat the figures as vendor-claimed and expect no product footage. Transcript not read.
Concourse Raises $17M to Bring Agentic AI to Finance Teams | The SaaS CFO | Concourse
A 35-minute episode of The SaaS CFO podcast in which host Ben Murray interviews Concourse co-founder and CEO Matthieu Hafemeister about the company and its agents. It helps a finance buyer understand the product thinking and target customer. Premiered 24 March 2026, about 130 views. It is a founder interview with no product walkthrough, and the funding figure in the title is the channel's. Transcript not read.
Concourse - Proactively understand trends
A one-minute vendor clip showing Concourse surfacing key insights and financial trends without being asked. It illustrates the proactive alerting that separates agents from a BI dashboard. Posted 16 October 2024, under 100 views. It is a short 2024 marketing clip with no detail on alert rules or thresholds. Transcript not read.
Concourse - Financial Report Automation
A short vendor demo of Concourse generating month-end reports automatically. It is relevant for teams that rebuild the same monthly pack by hand. Posted 16 October 2024, about 150 views, 1 minute 25 seconds. It is a vendor clip from 2024 and does not show review or approval steps. Transcript not read.
Concourse - Better Financial Forecasting
A 52-second clip of Concourse creating financial models and forecasts from connected data. It shows the forecasting job an FP&A lead would test first. Posted 16 October 2024 by the vendor, about 180 views. It is too short to show forecast logic, versioning or how assumptions are stored. Transcript not read.
Concourse Launch Video
The vendor's two-minute launch video describing agents that take on manual finance work, plan multi-step tasks and work alongside the team. It gives a buyer the original pitch in brief. Posted 16 October 2024 by Concourse, about 320 views. It is marketing from the beta period, before general availability in January 2026. Transcript not read.
Concourse - Chat with your financial data
A one-minute product clip showing Concourse retrieving data, calculating metrics and generating charts from a plain-language question. It is useful for seeing the chat interface a finance analyst would use day to day. Posted 16 October 2024 by the vendor, about 170 views. It predates the 2026 agent builder, so the current interface will differ. Transcript not read.
Compliance depth
SOC 2 Type II by Schellman across all five criteria, a SOC 1 Type II badge, SAML SSO, MFA, agent and dataset-level roles, and logged agent actions.
Scalability
No published caps; connects Snowflake, BigQuery and Databricks, plus several NetSuite subsidiaries or QuickBooks companies in one analysis.
AI / agent readiness
Agents, triggers, evals and model routing are the product, but no public API or data MCP exists; its only MCP server searches the Concourse 101 docs.
Support
Ex-CFOs and forward-deployed engineers build the first agents onsite, train the team and maintain agents; no SLA or support tiers published.
Implementation
Connectors link by OAuth in under 10 minutes and the vendor team runs setup; the vendor says most teams are live in weeks, with no model to build.
Pricing transparency
Quote only: price depends on systems, workflows and team size per its FAQ. MakerStack reported $99 and $499 monthly plans in August 2026.
Modeling depth
No planning engine: agents build forecasts, scenarios and actuals-to-plan variance in code. No driver model, budget versions or Excel add-in.
ERP & data connectivity
30 read-only connectors incl. NetSuite (Built for NetSuite SuiteApp), Intacct, QBO, Xero, SAP, Oracle, Workday, Snowflake; no Business Central.
AF Score
Audit Friendly score for Concourse across eight dimensions, 5 November 2026.
Price signal
No price list on concourse.ai (checked 1 November 2026); MakerStack reported $99 and $499 monthly plans on 31 August 2026.
Connectors
Integration pages on concourse.ai, all read-only, including NetSuite, Sage Intacct, SAP and Snowflake (checked 1 November 2026).
Tailored plan
Scoped to connected systems, automated workflows and team size, per the vendor FAQ (checked 1 November 2026).
Starter (reported)
One seat, per MakerStack on 31 August 2026. Not listed on concourse.ai as of 1 November 2026.
Professional (reported)
Up to five seats, per MakerStack on 31 August 2026. Not listed on concourse.ai as of 1 November 2026.
Forward-deployed team
Ex-CFOs and engineers build the first agents onsite, train staff and maintain agents. The FAQ calls this part of every deployment, but no page says whether it is priced in the subscription or billed as services (checked 1 November 2026).
Custom connectors
The integrations FAQ says new connectors can be requested and CSV uploads cover any source without one.
It sits beside your planning tool
The FP&A page covers forecasting, variance analysis, board reporting and reusable metric definitions (checked 1 November 2026). It does not describe driver-based planning, budget submission and approval, workforce planning or a spreadsheet add-in. Teams that need a budget of record still need Excel or an FP&A suite, and Concourse reads from it. Price the two together.
Every connector is read-only
All 30 integration pages on concourse.ai, from NetSuite and QuickBooks to Snowflake, say Concourse uses read-only access and never modifies the source system (checked 1 November 2026). The vendor's 11 September 2026 post on journal entry automation says Concourse agents draft entries, attach support and route them for approval before posting, and the treasury page says agents draft transfers for approval. The same post says Concourse connects to more than 100 systems, while the site lists 30 integration pages. Ask which actions can write to a system today, which systems they reach and under what approval control.
Self-serve pricing has gone quiet
The January 2026 launch post promised a free start and usage-based pricing, and MakerStack reported $99 and $499 monthly plans on 31 August 2026. On 1 November 2026 the vendor FAQ says pricing is tailored and there is no pricing page. Airframe describes a credit-based model. Ask for the unit, the overage rule and a cap in the order form.
Exit terms need an order-form clause
The Online Master Services Agreement, last updated 12 August 2024, says the customer owns its Customer Data and leaves term, fees and renewals to each order form. Termination does not entitle the customer to any refund, and a general confidentiality clause has each party return or destroy the other's confidential information. The agreement sets no export window or format for data, agents or methodology rules, and grants Concourse a perpetual license to Customer Data inside its aggregated statistics. The home page says agents are owned by the customer, so put export and deletion terms in the order form.
Outcome numbers are the vendor's
The site cites a 75 to 86 percent cut in manual work, 99 percent or better accuracy, a 50 percent forecast accuracy gain and $260M or more in revenue and savings, none independently confirmed (checked 1 November 2026). The same 'two years of AI roadmap' quote is credited to Carrot's CIO on the home page and to Front's CFO on the deployment page. Ask for references you can call.
SOC 1 is a badge, SOC 2 is the report
The security page and home page show SOC 1 Type II and SOC 2 Type II badges, but the security FAQ describes only the SOC 2 Type II examination by Schellman (checked 1 November 2026). The reports sit behind an NDA in a Vanta trust center. If agents touch close or reconciliation evidence your auditors rely on, request the SOC 1 report and its bridge letter.
Concourse pitches itself as a treasury system
A comparison post positions Concourse as a replacement for Kyriba for most finance teams, while conceding that SWIFT payment factories and in-house banking are not its focus.
Context layer with versioned company methodology
An engineering post describes how agents find data across systems, apply versioned and permissioned company rules, and reject any report figure that does not match its query result. Reviewed user corrections become versioned company knowledge.
Cambrean acquisition and hybrid self-serve model reported
AlleyWatch reported that Concourse acquired Cambrean and moved to self-serve Starter and Professional plans for small teams alongside forward-deployed support for enterprises. Deal terms were not disclosed.
$12M Series A and general availability
Concourse announced a $12M Series A led by Standard Capital, with Andreessen Horowitz, CRV and Y Combinator participating, and opened the platform to any finance team.
Concourse SuiteApp achieves Built for NetSuite status
The SuiteApp, built on the NetSuite SuiteCloud Platform, lets NetSuite users deploy agents that query NetSuite data for reports, variance analyses and board decks.
Concourse publishes no price list. Its FAQ says pricing depends on the systems you connect, the workflows you automate and team size, and asks you to book a walkthrough (checked 1 November 2026).
Third-party reports give a floor. MakerStack reported Starter at $99 a month for one seat and Professional at $499 a month for up to five seats on 31 August 2026, but those plans are not listed on concourse.ai now. No Vendr marketplace page covers Concourse. Ask for the unit, whether usage is metered in credits, and whether forward-deployed work is included.
The online agreement, last updated 12 August 2024, leaves term, fees and payment dates to each order form, charges 2 percent a month on late payments and allows suspension after five days of non-payment. Termination does not entitle the customer to any refund, and fees already due stay payable.
You own your Customer Data under the agreement, and the integrations FAQ says disconnecting a source stops access and removes cached data per your retention settings. Outputs export as CSV, PowerPoint and Word files. The agreement sets no export window for agents, prompts or methodology rules, and it grants Concourse a perpetual license to Customer Data inside aggregated statistics, so write export and deletion terms into the order form.
Connectors link by OAuth or API keys in under 10 minutes, per the integrations FAQ, and the vendor says most teams are live in weeks. There is no data migration and no model to build.
On enterprise deals, ex-CFOs and forward-deployed engineers come onsite, build the first agents with your team, train staff and maintain the agents afterwards. Your side needs a finance lead to define metrics and review outputs, and an ERP admin for credentials. Audit Friendly estimates days for a single-entity startup.
Not today. Concourse documents no public API and no MCP server for workspace data. The only MCP endpoint, on docs.concourse.co, searches and reads the Concourse 101 help pages (checked 1 November 2026).
Concourse is itself the agent layer: it routes work across frontier models internally and delivers results to email, Slack, Teams, SharePoint, PowerPoint and Word. Claude can work with those exported files, but it cannot run or edit Concourse agents.
Answered only from our own published research on this tool, never from general internet noise. If we cannot answer it well, our research agents will dig in and publish a sourced answer.