A few weeks ago I asked Dispatch readers what they wanted me to write about. Many of you replied, and I will work through those suggestions over the coming issues. This one comes from Nagendra Sherman , who works in data strategy, AI readiness, and governance. His topic? Is your finance function ready for AI, or is it simply buying AI tools?

It connects the last three issues. Issue 28 argued that process debt and data debt live largely in GBS. Issue 29 walked through the sequence from paying down that debt to deploying agents. Issue 30 covered what I heard while chairing SSOW in San Diego, where the AI conversation had moved from deployment to proof. Nagendra’s question addresses all three

Buying Is Easy

Buying AI has never been easier. ERP, AR, AP, and treasury vendors all sell some form of AI capability now. Signing the contract and launching a pilot can make it look like progress on the quarterly review, even when the way the work gets done has not changed at all.

Readiness is harder to see because demos don’t typically hit on it. It sits in how consistently a process runs across regions, who owns a data field when it is wrong, and whether anyone can explain why an exception happened. A finance function can own several AI tools and still not technically be ready for any of them.

A Readiness Checklist

Nagendra suggested a readiness checklist, and it is a useful format because it turns a vague question into one you can answer. Here is the version I would use and recommend to clients. Any question you cannot answer with a confident yes is a good place to start.

Process:

• Does the process run the same way in every region and entity, or does each team have its own version? (Exclude necessary regional variances)
• Are the rules documented well enough for a new hire to follow without asking someone?
• Do you know your exception volume and the main reasons behind it?

Data:

• Does every critical data element have a named owner who is accountable when it is wrong?
• Can you rely on your customer, vendor, and GL master data without manual checks?
• Are your source systems connected, or does someone move data between them manually?

Controls:

• Have you decided which actions AI can take on its own and which need human approval? Be thoughtful to not just add human reviews because “that’s how we always do it.”
• Will segregation of duties and access controls still hold once an agent is doing part of the work?
• Will there be a traceable record of what the AI recommended and who approved it?

Measurement:

• Did you capture a baseline before deployment taking variance drivers into consideration?
• Have you named the outcome you expect to change and by how much?

Readiness Tested

Nagendra also pointed to three areas where readiness matters in different ways.

In liquidity management, an AI cash forecast is only as reliable as the cash position data. If balances across banks and entities arrive late or need manual cleanup, the forecast is then inherently flawed. Before an AI forecast informs a funding decision, treasury teams should validate it against actual results over several cycles and understand the range of uncertainty around it.

In GBS, readiness starts with standard processes, clear data ownership, and fewer disconnected systems. It also means deciding before go-live where AI can act independently and where a person still has to approve. That boundary is far easier to set at the start than to redraw after something goes wrong.

In finance operations, reconciliation, collections, and exception handling are strong candidates because volumes are high and many decisions follow rules. They also sit close to the controls that auditors care about. An agent that clears reconciling items or sends collection notices has to work within the same access rules and segregation of duties as the people it supports. Every recommendation and every approval needs to leave a trail someone can follow months or even years later

It Works When:

Nagendra suggested three measures that I would put at the center of any AI business case, because the business should be able to verify each one:

• Forecast accuracy: compare forecast cash to actual cash at the same time each week, before and after deployment.
• Rework: count how often an item gets touched (by people or technology) more than once before it is resolved.
• Exception resolution time: measure the time from when an exception is raised to when it is closed.

Measures such as the number of bots deployed or estimated hours saved are easy to report and hard to verify for ROI. As I wrote after SSOW, the standard has shifted to showing that a result changed. Measures like these are how you show it.

Where You Should Start This Week

Pick one process you are considering for AI and run it through the checklist with your team. The questions you cannot answer become your readiness plan. That work takes longer than signing a purchase order for new technology and it is what makes the purchase worth something.

Nagendra framed this as a series, so I will come back to it with a closer look at readiness in liquidity management, in GBS, and in finance operations controls. Thank you, Nagendra, for the suggestion. Other readers sent topics on unapplied cash, B2B collections, cash forecasting, and whether finance KPIs reflect real business value, and those are on the list as well. If there is something you want covered, reply and tell me.

If you want an outside view on how ready your own function is performiung, Clear Cycle Advisors works with finance and GBS leaders on liquidity management, finance transformation, and fractional CFO services. Reply to this email or reach me through clearcycleadvisors.com.