I spent the past few weeks in back-to-back conversations with treasury leaders in the US. First at Treasury Table, our own intimate dinner in San Francisco with teams from some of the world's most sophisticated companies. Then at the Treasury and Cash Management Summit on the West Coast in Santa Clara, where over two days, treasury teams dug into how AI and automation are reshaping cash management, liquidity, and working capital. Real-world use cases and roundtables where peers worked through shared challenges together.
The conversations were different in format but identical in theme: these teams have already done the hard work. The TMS is configured. The ERP integrations are live. Bank connectivity is running. They've spent years and millions getting the infrastructure right. And yet, despite all of that investment, AI still isn't a part of how these teams actually operate. The data is flowing, but no one is using AI to make sense of it.
It's not a talent problem. These are world-class operators at world-class companies. It's that the platforms they rely on were never built with AI at their core. Sure, every vendor talks about AI now. It's on every website, every pitch deck, every conference slide. But bolting a chatbot or a copilot onto a system that was architected 15 years ago isn't AI. It's marketing. If the underlying data layer is fragmented, if the platform still depends on rules-based logic and manual inputs, then the 'AI' is just a wrapper around the same broken workflow.
Treasury teams don't need more dashboards or bolt-on features. They need AI that's native to how the platform works. Trained on treasury data, embedded in treasury workflows, and designed to handle the complexity that makes this function unique. No amount of AI branding on top of a legacy system changes that. And, with the exponential growth of AI's capabilities (primarily in the last 3 months), teams that are still in "wait and see" mode are being left behind.
Where AI actually makes a difference
AI in treasury isn't about replacing your team with a chatbot. It's about removing the manual weight so your team can focus on what actually matters: the decisions that drive liquidity, manage risk, and optimize capital allocation.
We've all heard the use cases: cash flow forecasting, scenario planning, fraud detection. But the real question isn't what AI can do. It's what's required underneath for AI to drive meaningful outcomes. Here's where it gets tangible.
It retains what your team can't
What makes AI different from automation is that it's dynamic. It learns. When a senior analyst leaves, their mental model of how your cash flows behave walks out the door with them. AI doesn't forget. It builds institutional knowledge that compounds over time, detecting business patterns as they emerge across all accounts, not just the ones someone remembered to check.
Think about cash forecasting. Today, most teams rebuild their forecast from scratch every cycle, pulling historical transactions, open AP/AR items from the ERP, layering in manual inputs. AI can synthesize all of those sources continuously, learning from actual payment patterns to predict when cash will really move. It handles multi-entity consolidation without chasing subsidiaries for Excel submissions. And critically, it understands which scenarios are relevant for your corporate context, not the generalised assumptions you'd get from a consumer AI tool.
The same principle applies to transaction categorization. Hard-coded rules break every time your business changes. A new entity, a new currency, a new bank relationship. AI that retains institutional knowledge analyzes transaction descriptions alongside your business context to assign the right categories for every inflow and outflow. Each classification comes with an explainable reason, not a black box. When you correct something, the system learns automatically and improves over time. It handles intercompany detection, bulk reclassification, and historical corrections without filing a support ticket. The result is consistently high accuracy without your team maintaining a single rule. Your team stops writing rules and starts managing exceptions.
It learns without being taught
There's been a lot of analogies about treating AI like a smart entry-level analyst. Teach it what you know and it'll do the work. The problem is, who actually has time to teach an entry-level analyst everything they should know? That's where the real bar for AI sits: tools that make learning feel seamless and controlled, not another onboarding project.
This shows up most clearly in forecasting accuracy. Every forecast line should show its source: what came from ERP data, what the ML model predicted, what was manually adjusted. That transparency builds trust. And when the model gets something wrong, the correction becomes the training data. The system doesn't just accept the fix; it learns the pattern behind it. Over time, teams using this approach see meaningful reductions in forecast variance. Not because someone spent weeks tuning parameters, but because the AI watched how the business actually behaves.
The same feedback loop applies to variance analysis. AI-powered variance analysis compares forecasts against actuals and traces misses back to the specific transactions that caused them. Which payment types are drifting? Is accuracy improving over time? Without this closed loop, teams can't improve their forecasts and stakeholders stay skeptical of the numbers.
It has to be precise, not probabilistic
In a world where pennies matter, "close enough" isn't good enough. Large language models are incredible at generating text and reasoning through ambiguity. But treasury demands determinism. When your forecast says cash will land on Thursday, the CFO needs to know whether that's a data-driven prediction or a best guess.
This is why scenario planning can't be an afterthought. When plans change, you shouldn't need to duplicate spreadsheets to answer "what if" questions. AI-powered scenario modeling applies adjustments on top of your base forecast while preserving the underlying daily and weekly patterns. What if collections drop 5%? What if that payment doesn't land? You compare scenarios side by side and see the liquidity impact in real time. Precision, not approximation.
It's only as good as the data underneath
None of this works if your data foundation is broken. And in treasury, the data problem is uniquely fragmented. Cash sits across dozens of bank portals, investments live in separate systems, and balances get reconciled manually every morning.
AI changes this by aggregating balance data across all connected banks into a single view, sliceable by entity, currency, or bank, drillable from consolidated position to individual account. But the real shift goes beyond operational cash. Most teams track term deposits in one place, money market funds in another, bank balances in a third. AI brings all of this into a single liquidity view: maturity dates, interest rates, and counterparty exposure visible alongside operating balances. When you can see total liquidity across your entire portfolio, you stop making overly conservative investment decisions driven by uncertainty.
What we're building at Palm
Before building Palm, I spent a decade in treasury at Uber, Levi's, and Remote, trying to make the existing platforms do what I needed. I configured the TMS. I built the integrations. I wrote the rules. And every time the business changed, the whole thing broke. The problem wasn't execution. It was that these systems were never built to learn.
So we built something different. Palm is AI-native treasury infrastructure. Not a legacy system with AI bolted on, but a platform where AI is foundational to every workflow. It connects to your existing banks, TMSs and ERPs, ingests your data, and starts delivering insight from day one.
Pulse is coming
Next week, we're launching Pulse, and it changes how treasury teams interact with their data entirely. I can't say too much yet, but if you've ever wished your treasury platform could surface the insight before you had to go looking for it, Pulse is built for that moment.
The companies that start building their treasury AI infrastructure today will have a compounding advantage over those that don't. That gap only widens with time.
Photos from Palm's Treasury Table in San Francisco, February 2026
Want to join the next one in your city? Register your interest here
Photography by Lana Dubkova | sfcorporateeventphotographer.com
