How to Include AI in Your TMS RFP

Have you heard of this exciting new piece of tech? It’s called AI. You might not have come across it yet, but it could change everything about the way treasury teams work.

Ok, while you obviously have heard of AI (in fact, it’s probably all you're hearing about at the moment), the gap between the marketing vision and the real world application can still be tough to cross. That is, unless you’re very clear about how best to use and adapt it for your specific use case. Nowhere is this more important than when shopping for a new TMS. 

It’s your central treasury hub, these days it is absolutely going to have AI embedded, but the trick is seeing past a dizzying features list and ensuring you end up with technology that fulfills its promise. That is, greater efficiency, faster workflows, reduced errors and the ability to scale output without scaling heading.

Getting this mix right makes your RFP more important than ever. The right questions will separate genuine AI capability from surface-level feature dressing., while the wrong ones will leave you selecting a system that looks impressive on paper and disappoints in practice.

Here's how to approach it.

Use Cases, Not Feature Lists

Before you write a single question, be specific about where AI would actually make your team's life easier. The most common pain points are cash flow forecasting accuracy, manual reconciliation, and the time spent aggregating data from multiple bank portals and ERPs. 

If you don't anchor your RFP to these, you'll end up evaluating AI features you'll never use. Too often, companies set out to find treasury management software without clearly defining their goals. This leads to bloated feature lists, RFPs full of irrelevant requirements, and ultimately a solution that doesn't fit.

Do you need better short-term cash positioning, or longer-range scenario modelling? Are you trying to reduce manual data entry, or surface anomalies in payment flows? Are you running a lean corporate treasury team that needs AI to do the heavy lifting, or a larger treasury function that wants AI as a decision-support layer rather than an autonomous one? The answers shape very different sets of requirements, and vendors who can't map their AI tools to your specific workflows probably can't deliver on them.

The outcomes you’re trying to achieve need to be the beginning, middle and end of your TMS search. 

A Demo is Worth a Thousand Words

Most RFP responses will say things like ‘Our platform uses machine learning to improve forecast accuracy over time.’ Sounds great, but it’s meaningless without context. Push vendors to show you the actual mechanism. What data does the model train on? How does accuracy improve? How are data anomalies accounted for? What's the baseline, and what’s the best-case over the long term? 

The best AI forecasting tools can combine information from bank data with accounts receivable and payable to calculate customer-level DSO and vendor-level DPO, and then use that same information to generate forecasts based on actual payment behaviour, not assumptions. If a vendor can't walk you through that level of specificity in their demo, their AI is probably more marketing than mechanics.

Good questions to include in your RFP:

• Describe the AI/ML models used in cash forecasting. What inputs do they require, and how do they improve with use?

• How does your system detect anomalies in transaction data, and what triggers an alert to treasury staff?

• Can your AI tools generate scenario analysis? Walk us through a specific example with real outputs.

• What percentage of your current clients are actively using AI features, and what outcomes have they reported?

That last question matters more than it might seem. Vendor adoption rates within their own customer base are a reliable signal of whether a feature is genuinely useful or merely shipped.

Want to see a demo of Trovata AI? See what’s new.

Probe the Data Architecture

The success of any AI or ML application in a TMS depends on the quality, availability, and timeliness of the underlying data. A vendor with a robust connectivity architecture is better positioned for long-term success with AI and ML than one running on a legacy architecture.

This is the part many TMS RFP’s overlook. A sophisticated forecasting model is only as good as the data it can access, and if a vendor's connectivity architecture is legacy-based, the AI layer sitting on top of it will underperform. 

Some TMS platforms claim to offer universal multibank access, but if they rely on legacy connection methods they can be unreliable, and getting accounts linked can require costly external consultants and months of configuration. If that's the foundation your AI is running on, no amount of machine learning sophistication will compensate.

Ask specifically about API infrastructure. APIs are foundational to connecting disparate systems such as your TMS, ERP, banking platforms, and fintech solutions, and a vendor's API comprehensiveness, reliability, and security directly determines its ability to unlock AI and ML potential. Any vendor who can't give you a clear, confident answer about real-time bank connectivity and data normalisation should be treated with scepticism, regardless of how impressive their AI demos look.

Build in Governance Requirements

If your TMS is making recommendations on hedging strategy, cash positioning and payment timing, your team, your auditors, and potentially your board need to understand the basis for those recommendations. Robust governance processes are arguably the most important ‘feature’ of any new technology, especially in an area as important as treasury.

There's a meaningful difference between a system that surfaces a recommendation and one that can explain, step by step, how it arrived at it. The former might be fine for low-stakes operational decisions, but for anything touching liquidity management or FX exposure, you need the latter. This is especially true as we move beyond generative AI chat interfaces to automated agentic AI workflows.

Include these questions in your RFP:

• How does the system communicate the reasoning behind AI-generated recommendations or alerts?

• Can treasury staff override AI-driven actions, and is there a full audit trail when they do?

• How is the AI model updated, and what controls exist around model changes that could affect outputs?

• How does the system handle edge cases or scenarios outside the model's training data?

The final question is particularly important for organisations with more complex or unusual treasury structures. AI models trained on typical corporate treasury behaviour may perform poorly when applied to treasury teams with specific regulatory constraints, unusual cash flow patterns, or non-standard hedging strategies.

In short, you don’t want your treasury data going in and out of a black box.

Consider the Vendors Roadmap

A TMS is typically a multi-year commitment, and while the AI capabilities a vendor ships today are obviously important, their roadmap and the pace at which they're developing it is equally so. The vendors best positioned to compound on their AI investment are those with genuinely modern architecture, not those retrofitting AI onto legacy platforms as a defensive move against more agile competitors.

In your RFP scoring, weight the following heavily:

• Is AI development done in-house, or dependent on a third-party model provider?

• How frequently do AI features receive updates, and how are customers notified of changes? 

• Is the product team investing in proprietary data models built on treasury-specific training data, or relying on general-purpose LLMs with a treasury-flavoured prompt layer on top? These are different things, and it’s an important distinction.

You should ask vendors directly what AI capabilities are on their 12–24 month roadmap and how are customer requirements incorporated into that roadmap. The answers will tell you a lot about whether a vendor sees AI as a strategic differentiator or a checkbox.

Recommended: Announcing Trovata AI 2.0: AI Chat, Insights and Agents for Corporate Treasury

Making The Most of Your Next RFP

The goal of your TMS RFP is to find the one whose AI capabilities map cleanly to your workflows, sit on a solid data foundation, meet your governance requirements, and will continue to develop over the contract period.

The vendors who can answer your questions with specificity, using real numbers, real use cases and real client outcomes are the ones worth taking seriously. The ones who respond with polished language but no substance are telling you something important, even if it's not what they intend.

Ask the right questions upfront, and you'll save yourself a much harder conversation twelve months into implementation. To see how Trovata can modernize your treasury function and help them become more efficient and more strategic, book a demo today.