30/60/90-day plan for AI in finance

A 30/60/90-day plan for AI in finance is a stepwise roadmap for the first three months — days 1-30 diagnosis and quick wins, days 31-60 first pilots on concrete use cases, days 61-90 first scaling and governance foundation. For finance teams it's long enough to produce change but short enough to stay concrete — and it lives or dies on getting the first 30 days right.

A grand two-year strategy makes little sense for a technology doubling in capability every 2-3 months. A 30/60/90-day plan does: short enough to stay concrete, long enough to produce change in a finance team. The structure below isn't a template, it's a sequence — it lives or dies on getting the first 30 days right.

Underlying principles

Three principles carry the entire plan:

• Start small, prove value, scale fast. A use case that works in week 4 for one controller is more valuable than a plan for the whole finance function still in draft at week 12.
• Augment workflows, don't replace them. AI sits inside the close and AR cycles, not alongside. Rolling out AI as a stand-alone tool without workflow context yields unused licenses.
• Speed of learning = speed of transformation. The goal of the first 90 days isn't "AI implemented in finance," it's "the finance team has learned how it implements AI."

Day 0 — the diagnosis

Before day 1, run an AI readiness assessment per pillar (see AI readiness for finance functions). The outcome decides where the biggest lever sits. Without diagnosis you plan in the dark.

Also ask yourself: which finance priority would benefit most from AI? Faster close? Lower DSO? Better variance analyses? Easier VAT reconciliation? The answer steers the use-case choice on days 31-60.

Days 1-30 — foundation and first quick win

The goal of the first month is not close-cycle results, but the foundation: sponsorship, champion, tools, policy, and one visible quick win.

Week 1 — lock in commitment

• Name the CFO or finance manager as sponsor with calendar time (at least 2 hours per month).
• Name a finance AI champion (typically a senior controller) with protected time (at least half a day per week).
• Capture the why-message in one paragraph anyone in finance can repeat.

Week 2 — tools and policy

• Pick one primary AI tool (Copilot for Business, Claude Enterprise, or Saldus for the accounting connection). No consumer accounts for customer data.
• Write a first finance AI policy of at most two pages: tier classification of finance data, which tools for which tier, basic rules for MCP and accounting integrations. See AI governance for finance.
• Roll out the tool to a pilot group of 3-6 finance staff — not to the whole team.

Week 3 — inspire and practice

• Run one hands-on kickoff of 2 hours with the pilot group: concrete finance briefs (variance commentary, AR mail, summarize a close-manual section), no abstract demo.
• Start the "AI win of the week" ritual in a Teams or Slack channel for finance.
• The champion sets up 1-on-1s with pilot participants to surface blockers.

Week 4 — deliver the first quick win

• Pick one simple finance workflow where AI demonstrably saves time within a week. Examples: first draft of variance commentary, summary of management meetings, draft mails for AR reminders.
• Measure before and after: how much time did it take, how much now, how often does this happen per week? Even rough estimates are enough — it's about visible evidence.
• Share the result in the finance standup, including the name of the staff member who did it.

By day 30 you have: sponsor, champion, tool, policy, pilot group, one measured quick win, and a ritual that's running.

Days 31-60 — first pilots with real finance impact

Month two moves from "everyone in finance can use AI" to "one or two finance workflows work structurally differently." This is where you switch from tool adoption to process change.

Pilot selection (start of month 2)

Pick 2-3 pilots with clear scope:

• High finance impact — touches a workflow that recurs often (close, AR, VAT) or costs a lot of time per occurrence.
• Achievable in 4 weeks — visible result within one month.
• Mandate present — the process owner (controller, AP clerk, finance manager) is involved and wants it.

Typical finance pilots: variance commentary for the month-end close, bank-reconciliation proposals, AR mails with customer segmentation, VAT-variance analysis, automatic extraction of purchase invoices.

Pilot execution (weeks 5-8)

• Named owner per pilot, with deadline.
• Set the success criterion upfront: "we want variance-commentary time from 4 hours to 1 hour, measurable over 3 consecutive close cycles."
• Weekly 15-minute progress check.
• Explicit permission to stop if it doesn't work — stopping is a valid outcome.

Fill in governance-by-design

• Walk the policy against the concrete workflow: which data goes where? Does that include customer IBANs or salary data? Is that contractually covered?
• Where data from Exact, MS365, or a banking portal is needed: involve IT from week 5, not at scaling.
• Audit log per AI action — from the pilot, not added later.

By day 60 you have: 2-3 finance pilots with measured outcomes (success or deliberately stopped), a policy refreshed on the basis of what's actually happening, and a growing group of finance staff using AI regularly.

Days 61-90 — first scaling and learning loop

Month three is the move from experimentation to embedding. Not everything scales; precisely for that reason you have to choose deliberately.

Decide per pilot: stop, iterate, or scale

• Success criterion met and process stable → scale to the whole finance team or the whole close cycle.
• Success criterion met but still shaky → iterate four more weeks before scaling.
• Success criterion not met → stop, document the learning, pick a new use case.

This decision is itself a culture moment in finance. Publicly stopping a pilot and framing it as a win ("we've learned this route doesn't work for our data") normalizes stopping for the rest of the cycle. That speeds up every following pilot.

Buy vs build for finance

For pilots that scale, the question arises: do we do this with standard tooling (Copilot, Claude, a platform like Saldus) or do we build something custom? In SME finance the answer is almost always: pick standard first, build only what is truly finance-unique. A custom invoice OCR is pointless if a good SaaS solution exists; a custom variance-commentary template is useful if it's intertwined with your KPI definitions.

Scale training

• From pilot group to the whole finance team affected by the scaled workflow.
• Plan monthly hands-on sessions as a fixed ritual.
• Let pilot participants train their colleagues themselves — that builds ownership and makes the message credible.

Measure and repeat

• Repeat the readiness assessment for finance. Which pillars have shifted? Where is the bottleneck now?
• Draft the next 90-day plan, based on the new diagnosis. It won't be a copy — after 90 days you're at a different stage.

By day 90 you have: 1-2 scaled finance workflows, 1 deliberately stopped pilot, a policy that's lived, an audit trail that works, and a finance team that knows how it runs pilots.

Pitfalls specific to finance

• Too much at once in month 1. Three pilots started before the basis stands = half-baked month 3.
• CFO at distance. A sponsor only showing up at the kickoff doesn't provide cover when pilots run into IT or compliance obstacles.
• Broad rollout on day 14. Handing Copilot to the entire finance team without a pilot phase = unused licenses.
• No stop criterion. Pilots without explicit success criteria hang around forever; nobody dares conclude something doesn't work for your data.
• Tech before governance. Starting a pilot before it's clear whether customer data can be used leads to a pilot called off in week 8 and lost trust with IT/compliance.

Where this lands in practice

• A 45-FTE accounting firm starts day 1 with sponsor (managing partner finance) and champion (senior controller, 4 hours/week), and Copilot licenses for the 6 keenest consultants. Quick win in week 4: dictation-to-memo saves 20 minutes per client call. In month 2 this scales to all consultants. Pilot 2 (AI variance commentary) stops after 6 weeks — data quality not good enough.
• A 90-FTE scale-up picks three finance pilots in month 2: variance commentary for MT, automatic invoice extraction, and draft AR mails. After 60 days: variance scales, invoice extraction iterates, AR mails are parked (customer-segment data not in order). Clear outcome, no loss of face.
• A 140-FTE family business notices on day 60 that the CFO isn't really engaged. They restart the 30-day foundation with a different sponsor — a month lost, but the pilots are structurally better supported afterwards.