Software advice
Best AI Treasury Management Software in 2026
AI treasury management software is a platform that connects to a company's bank accounts and ERP, shows cash positions across entities, forecasts liquidity, runs payments and reconciliation, and uses AI agents to propose actions that a treasurer approves. It differs from accounting software, which records the past, and from a bank portal, which shows only one bank.
Quick answer
Atlar ranks first because it states SOC 2 Type II, ISO 27001 and AWS hosting in Europe and requires approval for its agents. Nomentia fits mid-sized European groups that want audited standards. Agicap suits smaller companies that accept a 12-month term.
A treasurer who looks at new software usually starts from a spreadsheet of bank balances and a forecast that is stale by Wednesday. Vendors now offer agents that sweep cash, match payments and brief the team each morning. The risk is not whether the model is clever but whether it acts on money without a person. Most pages hide prices behind a demo, so this ranking weighs what can be checked: named certifications, hosting regions, approval controls, bank coverage and contract terms. We read vendor pages in November 2026 and did not test any forecast, so accuracy figures belong to the vendors.
The 10 best AI treasury management software in 2026, ranked
There is no score. Vendors can pay for more visibility on this website, for example a higher place in a ranking or more room on a page. Commercial relationships can also affect which products we feature and how prominently. How we rank.
| # | Tool | Best for | Free plan | Entry price |
|---|---|---|---|---|
| 1 | Atlar | Growing groups with several banks that want agents with human approval | No | Quoted |
| 2 | Nomentia | Mid-sized European companies that want audited standards and a modular treasury suite | No | Quoted |
| 3 | Embat | Spanish and European finance teams that want modules chosen one by one | No | Quoted |
| 4 | Agicap | Small and mid-sized companies that want cash tracking tied to many banks and accounting tools | Trial | Quoted |
| 5 | Salmon Software | Corporate treasuries that want a configurable system from a European vendor | No | Quoted |
| 6 | Trovata | US-leaning teams that want cash visibility, forecasting and AI answers on normalized bank data | No | Quoted |
| 7 | Kyriba | Large global treasuries that need risk, payments, connectivity and working capital in one vendor | No | Quoted |
| 8 | HighRadius | Companies already using HighRadius for receivables that want treasury on the same platform | No | Quoted |
| 9 | Fides | Companies that want multibanking and Swift connectivity without replacing their systems | No | Quoted |
| 10 | Serrala | SAP-centered finance teams that want cash visibility and payments inside one receivables and payables suite | No | Quoted |
- 1
Atlar
Atlar ranks first because its security page states SOC 2 Type II with annual audits, ISO 27001:2022, AWS hosting in Europe in a dedicated account and alignment with DORA. The product page says agents check actions against uploaded policies and that users approve or reject recommendations. Prices are not shown, and the page does not say how AI uses customer data.
Strong on
- SOC 2 Type II, ISO 27001:2022 and European AWS hosting are all stated
- Agents check actions against uploaded treasury policies before proposing them
- Connects to banks and ERPs in more than 100 countries
Weak on
- No price is shown on the pages read
- Training and retention rules for AI features are not stated
Skip it if: Treasurers who need a published price or a long-established system with deep risk modules. No price on the home page. A separate pricing page is linked and was not read.
- 2
Nomentia
Nomentia is second because its security page lists the most audit detail in the group: ISO 27001:2022 with a downloadable certificate, ISAE 3402 Type II, SWIFT-compatible status and FSQS-NL registration. The suite splits into five modules, from Connect and Pay to Manage and Optimise. AI assistants and agents are named but barely described, and hosting is not stated.
Strong on
- ISO 27001:2022, ISAE 3402 Type II and SWIFT status are listed
- Five modules let a company start with connectivity and add more later
- Offices across Finland, Sweden, Austria, Germany, the Netherlands, Poland and the UK
Weak on
- AI assistants and agents are named without any description
- Hosting regions and prices are not stated on the pages read
Skip it if: Teams that want a self-serve start or an AI product explained in detail. Demo and contact only. A treasury ROI calculator is offered instead of a price.
- 3
Embat
Embat ranks third because its pricing page shows the module picker, payments in over 50 currencies and a no-commitment demo, and its footer carries an AENOR ISO 27001 seal. The AI agent TellMe matches transactions and flags exceptions, and Slack, Teams, Claude and ChatGPT connect. Hosting is not stated and the privacy policy is silent on AI training.
Strong on
- Bank connectivity covers API, host-to-host, EBICS and SWIFT for 15,000-plus institutions
- TellMe matches transactions, flags exceptions and suggests journal entries
- Modules for risk, payments and reconciliation can be added one at a time as needs grow
Weak on
- No prices or trial terms are shown on the pages read
- Hosting and AI training are not described in the privacy policy
Skip it if: Buyers outside Europe or those who need a published tariff before a call. Modules are chosen on the pricing page, then priced in a personalized proposal.
- 4
Agicap ranks fourth because its terms and hosting are the most concrete among cash tools: annual billing with a 12-month minimum, cancellation before renewal, Belgian Google Cloud hosting and a registered Lyon address. The privacy policy also says personal data may train AI models, and customer data follows separate agreements. Prices did not load, and the AI pages were bare.
Strong on
- Compatible with 300-plus UK and Irish banks and 150-plus accounting packages
- Policy states hosting in Belgium on Google Cloud
- A trial with an expert is offered before the annual term
Weak on
- Privacy policy allows personal data to train AI models
- Agicap AI and MCP are named in navigation but not described
Skip it if: Large groups needing risk, debt, in-house banking or intercompany modules. Offers are annual with a 12-month minimum. Prices were not loaded on the page read.
- 5
Salmon Software
Salmon Software ranks fifth because the home page names an NQA ISO 27001 accreditation and a Dublin address, and organizes the product around visibility, control and intelligence. The AI claims are short: AI-driven insights and automation with no named agent, model or data rule. It is a credible system, but a buyer must ask for detail the page leaves out, from hosting to agents.
Strong on
- NQA ISO 27001 accreditation is shown in the footer
- Registered address in Dublin is given on the page
- Product is organized into visibility, control and intelligence
Weak on
- AI is described in one phrase, with no named feature or data rule
- Hosting, prices and agent behavior are not stated
Skip it if: Buyers who want specifics about AI behavior or a printed price. No pricing is shown on the home page.
- 6
Trovata
Trovata ranks sixth because its policy gives a concrete hosting statement (mainly US data centers) and its footer shows an AICPA SOC logo, while the report type is not named. The product splits into Cash, TMS and a data layer, with Trovata AI 2.0 promising governed automation. Prices and a training rule are missing, and aggregated data may serve benchmarking.
Strong on
- Cash, TMS and a bank connectivity layer can be bought separately
- Policy states that data centers are primarily in the United States
- Trovata AI is framed around governed automation of repeat tasks
Weak on
- SOC report type and date are not named
- Policy allows aggregate data for benchmarking and improvement
Skip it if: European groups that need EU-only hosting stated in writing before a pilot. A pricing page is linked but no figures appear on the home page.
- 7
Kyriba
Kyriba sits seventh because its home page lists broad product lines and 10,000 bank connections, yet gives no certifications, hosting or price, only links to a Trust Center and an AI fact sheet. The policy lists San Diego and a regional headquarters in France. Its agentic AI, branded TAI, is announced as available without detail.
Strong on
- Treasury, risk, payments, connectivity and working capital product lines
- Reaches around 10,000 banks according to the home page
- Trusted AI fact sheet and Trust Center are linked
Weak on
- No certifications or hosting details on the home page
- TAI agents are announced without a description of actions
Skip it if: Small teams that want a quick start, a printed price or a light implementation. No price on the pages read. Product lines are sold separately by quote.
- 8
HighRadius
HighRadius takes eighth place because the treasury page describes forecasting, reconciliation and payment automation in detail but leans on unproven claims of 95 percent accuracy and bank fee savings of up to 30 percent. An IDC MarketScape leader mention is the only third-party signal, and a Trust Center is linked without named certifications or hosting.
Strong on
- Forecasts at daily, weekly, monthly or yearly intervals up to 12 months
- AI reconciliation and fraud detection sit beside payments automation
- Multi-currency and multi-entity cash positions are described on the page
Weak on
- Accuracy and savings figures are vendor claims without a method
- No certifications, hosting or prices appear on the treasury page
Skip it if: Treasurers who want a standalone system rather than a wider finance suite. Demo or discovery call only. The treasury page prints no price.
- 9
Fides
Fides closes the ranking because its page mentions no AI feature at all, only six solutions for multibanking, reporting, payments, cash pooling and format conversion. It earns a place for transparency on what it is: a certified Swift service bureau in Zurich that can host a company's own BIC. Buyers who want agents should look higher.
Strong on
- Certified Swift service bureau, with a fully hosted Corporate BIC option
- Access by SFTP or the ONEHub web platform
- Cash pooling and format conversion are separate solutions
Weak on
- No AI features are described on the page read
- No price, certification list or data policy appears on the home page
Skip it if: Treasurers who want AI forecasting, agents or an assistant in the product. No price. A demo or contact request is offered.
- 10
Serrala takes the last place because treasury is one part of a wider receivables, payables and payments suite, not the product's whole purpose. Cash visibility across banks and currencies and an SAP-embedded treasury module are real, and a fraud detection agent is named. Hedging and forecasting are not named on the pages read, so dedicated treasury tools rank above it.
Strong on
- Treasury module that runs embedded inside SAP, next to cash management and payment processing
- Payments Cloud with built-in fraud screening that the vendor says reaches 150+ countries
- Fraud detection agent described with explainable AI and human review
Weak on
- No price, free plan or trial is published, so every deployment is a quote
- Hedging and cash forecasting are not named on the pages read
Skip it if: Treasury teams that need a standalone system with forecasting, hedging and bank connectivity as the core product. No price is published; the pricing URL returned a 404 and every deployment option is a custom quote.
How these AI treasury management software were judged
How reliable are the AI agents in treasury software?
Reliability depends on who can act on the result. Atlar says its agents check actions against uploaded treasury policies, and users approve or reject recommendations. Not every vendor draws the same line, so ask whether an agent can release a payment. Embat's TellMe matches transactions and flags exceptions, Kyriba promotes its TAI agents and a Trusted AI fact sheet, and HighRadius claims 95 percent forecast accuracy up to twelve months out without showing a method. Ask each vendor to back-test a forecast on your own history and show every action in a log.
How much does treasury management software cost?
Treasury software rarely prints a price. As of November 2026, Embat shows a module picker that leads to a personalized proposal, Agicap states an annual offer with a 12-month minimum paid by SEPA debit or card, and Nomentia, Kyriba, Salmon, Trovata, HighRadius, Fides and Atlar point to a demo or quote. Cost usually follows entities, bank connections and modules. Ask for the price of each bank connection, of every extra legal entity and of the AI features, and whether implementation is included or billed apart.
Where does treasury software host bank data?
Bank data is among the most sensitive a company holds. Atlar states AWS hosting in Europe in a dedicated account, SOC 2 Type II and ISO 27001. Nomentia lists ISO 27001:2022, ISAE 3402 Type II and SWIFT status, and Embat and Salmon show ISO 27001 seals. Agicap's privacy policy says personal data is hosted in Belgium on Google Cloud and may be used to train AI models, with customer data governed by separate agreements. Trovata states it mainly uses data centers in the United States. For the rest, the vendor's policy was not checked for AI training.
How hard is it to switch treasury systems?
Switching is slow because bank connections, payment formats, approval chains and accounting mappings all have to be rebuilt and tested. Kyriba claims 10,000 banks and Embat 15,000 institutions, but a connection to your own bank still needs a live test. Agicap requires 12 months and allows cancellation before renewal. Fides can fully host a company's own Swift BIC, which ties payment identity to the provider. Ask which historical data, rules and payment templates can be exported, and how long both systems must run in parallel.
Which treasury platform fits your company and banks?
Start with the number of entities and banks, then the ERP. Atlar and Embat suit growing groups with several banks and a modern ERP. Nomentia and Salmon fit European corporates that want established standards, and Kyriba and HighRadius fit large global groups. Trovata suits teams that mostly need visibility and forecasting, Agicap serves smaller firms, and Fides suits companies that want multibanking without changing systems. Companies with SAP should confirm connector depth before the pilot, not after, and should ask which team maintains each connection once the project closes.
How to choose an AI treasury management platform in three steps
- Map your banks, entities and ERP before any demo. List every bank account by country and entity, the connection method each bank supports and your ERP with its version. Then ask each vendor to confirm in writing which accounts connect by API, which by host-to-host or SWIFT, and which need manual files. A claim such as 100-plus countries or 15,000 institutions says little about your five banks. Treat any gap in your own list as an implementation cost, not a footnote, and ask for the price of each missing connection before you compare vendors on features.
- Set the rule for what an agent may do alone. Agree in advance which tasks an AI agent may do without a person, such as reconciliation proposals or daily briefings, and which need approval, such as payments or sweeps. Atlar describes approval of recommendations and existing permissions, and other vendors describe manual, semi-autonomous and autonomous modes. Ask each vendor to show the permission screen and the audit log of one action, and decide who owns the policy file that agents read, since a stale policy produces confident but wrong proposals.
- Run a parallel forecast on real history. Give each shortlisted vendor twelve months of bank data and ask for a forecast of the following quarter, produced before the actuals are revealed. Compare variance by week with your current spreadsheet. HighRadius, Trovata and Agicap all describe AI forecasting, but none of their pages give a method, so a blind test on your own data is the only fair comparison. Include a quarter with unusual payments, since models fail there first, and score the miss by size as well as by count.
Frequently asked questions
What is the best AI treasury management software in 2026?
Atlar ranks first here because it states SOC 2 Type II, ISO 27001 and AWS hosting in Europe, and requires user approval for agent recommendations. Nomentia suits mid-sized European groups that want ISO 27001:2022 and ISAE 3402 Type II, and Agicap suits smaller firms. The best choice depends on your banks, entities and ERP.
What can an AI agent do in a treasury system?
Vendors describe agents that prepare cash positions, brief the team, match bank transactions, propose sweeps, monitor FX exposure and flag covenant risk. Atlar lists these tasks and says users approve or reject them. Embat's TellMe matches transactions and suggests journal entries. Ask which actions are executed by the agent and which are only proposed.
How accurate is AI cash forecasting?
Vendors publish claims, not tests. HighRadius states 95 percent accuracy up to twelve months ahead, while other pages describe forecasting without a figure. Accuracy depends on clean bank history, seasonality and how well payables and receivables data flow in. Run your own blind test on a past quarter before you accept any number.
Do I need a treasury system or is an ERP enough?
An ERP records transactions, but it seldom connects to every bank, forecasts cash daily or manages payments across entities. A treasury system becomes worth considering when you have several banks, several currencies or entities, and a manual consolidation that takes days. A single-entity company with one bank usually does not need one.
Which treasury tools are built in Europe?
Embat is registered in Madrid, Salmon in Dublin, Agicap in Lyon, Fides in Zurich and Nomentia lists Espoo, Finland, first among its offices. Atlar hosts on AWS in Europe and says it aligns with DORA. Kyriba and Trovata are US-headquartered, though Kyriba lists a regional headquarters near Paris. Check hosting and sub-processors rather than headquarters alone.
How long does a treasury implementation take?
It ranges from weeks to many months. Agicap lets you test its tool for a few days with an expert, and bank-connection work is the usual delay. No vendor page we read gave a binding go-live date for a multi-entity rollout. Request a written plan per bank and per entity.
Does treasury software connect to SAP?
Many vendors claim to, with different depth. Embat lists bidirectional ERP integration, and Nomentia lists ERP integration across its modules. Neither page named SAP versions. SAP landscapes differ widely, so confirm in writing which SAP version and modules are supported, whether the connection is certified and who maintains it after go-live.
Is treasury software with AI safe to use with bank data?
It can be if the vendor states certifications, hosting and data use. Atlar names SOC 2 Type II, ISO 27001 and AWS in Europe, while Agicap's policy says personal data may train AI models and customer data follows separate agreements. Ask for the data processing agreement, the sub-processor list and a written answer on model training.