Start your Finance AI strategy with the work

01

Choose the result before the technology

Start with a Finance decision or service that matters to the organization: a forecast people can use, a close with fewer unresolved items, timely cash visibility or a better investment decision. Describe the problem in the words of the people doing and using the work. Ask what happens today, what goes wrong and what a better result would let someone do.

Write the current measure, its definition, period and source. If there is no reliable baseline, establishing one is part of the first stage. A target without an agreed measure gives the team little basis for judging progress. Keep different outcomes separate: a faster draft, an accurate forecast, lower cash spend and better adoption need different evidence.

02

Find out what the organization can already do

Ask which entities and Finance processes are in scope, which systems and contracts support them, and where data comes from. Have the people who use the process walk through a normal case, a correction and an exception. Record manual handoffs, duplicated work, controls that must remain and the source used when figures disagree.

Compare a process change, configuration of the existing system, ordinary rules or RPA, and AI assistance. Retaining the ERP is compatible with improving a task around it. Check actual licensed features, access, integration and support before proposing another product. Record whether a capability was demonstrated, needs configuration, depends on an extra purchase or remains unproven.

03

Keep the roadmap broad and the first stage deliverable

Keep one roadmap for the work ahead: strategy and priorities; data and process requirements; people and responsibilities; evaluation; design and testing; preparation for use; release; and improvement after release. Give each activity an owner, an output, an effort estimate, prerequisites and a decision point. A shorter first-stage plan should identify what remains on that roadmap and why it waits.

Add effort in the same unit and period. If a two-week window has 24 total team-hours and the proposed work needs 32, there is an eight-hour gap. Reordering tasks does not create those hours. Reduce or defer a named activity, extend the window or obtain confirmed capacity from the person who controls it. Check dependencies before moving work; a later activity cannot finish before its required input exists.

A review can authorize the next investigation without authorizing implementation. Define the evidence needed for an initial AI use, who can accept it and what happens if it fails. A non-AI improvement may proceed on its own merits, but it does not demonstrate that an AI stage is ready. Dates remain conditional until scope, effort, inputs and people are confirmed.

04

Put value, authority and people in the same decision

Ask the accountable Finance leader to confirm the intended result and the people who hold the relevant permissions. Distinguish permission to investigate from permission to access data, spend money, change records or release a service. Bring in technology, data, security, legal, procurement and people leaders for the decisions that affect their responsibilities. Assignments need confirmed skills and time as well as a name.

Show one-off cash costs, recurring cash costs and internal effort separately. State currency, period, included costs and assumptions. Hours released, valued at a labor rate, describe possible capacity value; they are not cash savings unless an actual budget change supports that claim. Test the important assumptions one at a time, including lower hours released or higher recurring charges. Do not turn a capacity estimate into an ROI or funding approval.

05

Test the work people will actually use

Agree on the permitted inputs, representative tasks, awkward cases, expected result and review criteria before testing. Include correction of an answer, a changed fact, saving the result, leaving and returning, and taking the record into the meeting where it will be used. Record failures and incomplete work with the successes; an unreviewed result is still unreviewed.

Prepare the people and operating arrangements: practice time, task-specific guidance, support, access, exception handling and the route back if the new way cannot be used. Choose a next review with a clear decision: continue within the boundary, change the scope, resolve a missing condition or stop. A completed form, successful demonstration or training attendance does not by itself establish routine use or permission to expand.