Why 90 Days Is the Right Timeline
Most finance transformation projects fail for one of two reasons: they move too fast (big-bang deployments with no validation baseline) or too slowly (endless pilots that never convert to production). The 90-day framework is designed to avoid both failure modes.
The logic is simple: 30 days to establish a baseline and prove accuracy in shadow mode. 30 days to expand scope while the team builds confidence. 30 days to hand over the keys and measure real impact. By day 90, you have a result you can defend to the CFO — or a clear signal that something needs adjusting before you go further.
Phase 1: Discovery (Days 1–14)
The goal of discovery isn't to plan everything — it's to understand enough to pick the right pilot scope. Map just enough to identify your highest-volume, highest-friction bottleneck, and start there.
- Time-track the close for one cycle. Where does the team spend time? What percentage goes to reconciliation vs. exception handling vs. review vs. rework?
- Identify the three largest time drains. Typically: daily reconciliation, payment exception resolution, and inter-company matching. Pick the one with the highest volume and clearest rules.
- Define what the agent can decide vs. what needs a human. This is the most important conversation of the 90 days. Write it down. "Agent auto-applies cash if match confidence is above threshold. Routes to AR if below." That's a rule. Get ten of these and you have a policy.
- Establish baseline metrics. Cycle time, exception rate, hours per close, first-pass match rate. You cannot prove ROI without a baseline measured before the pilot starts.
Don't over-scope the pilot. If you try to automate everything in the first 90 days, you'll spend 60 days on integration work and have nothing to show. One process, one team, one geography. Success there funds everything else.
Phase 2: Shadow Pilot (Days 15–42)
The shadow pilot is where most of the real learning happens. The agent runs alongside the team — it makes decisions, the team reviews them, both compare notes. Disagreements are the most valuable data you'll generate in the 90 days.
- Run the agent on live data from day one. Synthetic data tells you nothing. Real remittances, real exceptions, real edge cases.
- Review every disagreement. When the agent and a human reach different conclusions, find out why. Usually one of three causes: missing context (agent didn't have a data source), wrong rule (the written policy doesn't match what the team actually does), or a genuine exception (edge case that needs a new rule).
- Track accuracy daily. By day 14 of the pilot, you should be above 90% match accuracy on your defined scope. By day 28, above 95%. If you're not, don't expand — diagnose.
- Don't change the human workflow yet. The team should keep doing what they always did. The agent is a shadow, not a replacement.
By end of Phase 2: the agent handles 85–90% of the defined scope with accuracy that exceeds the team's manual rate. The team trusts the output. You have evidence to expand.
Phase 3: Scale (Days 43–90)
Phase 3 is where the close actually changes. The agent moves from shadow to primary. The team shifts from doing the work to reviewing exceptions. The close gets faster.
Weeks 7–9: Add the next process to the agent's responsibility. If Phase 2 covered daily reconciliation, Phase 3 adds accruals and inter-company matching. 95% accuracy and a clear exception-handling path is enough to expand.
Weeks 10–12: The agent runs the defined workflows autonomously. The team reviews the exception queue — not the full output. For most finance teams, this cuts close-related manual work by 60–70% and compresses the actual close timeline from 10+ days to 4–5 days within the first production month.
What Day 90+ Looks Like
The finance team doesn't stop working. They shift what they work on. The reconciliation queue that used to take four people two days now runs overnight. The exception queue that used to have hundreds of open items has fewer than twenty — and each one has full context attached so resolution takes minutes, not hours. The close compresses. Ledger imbalances surface and resolve within the day instead of at month-end. The CFO sees numbers that are days fresher. And the finance analysts who were buried in matching work start doing the analysis they were hired for.
Three Things That Derail the 90-Day Path
1. Skipping the baseline. Teams that don't measure before the pilot can't prove what changed. Without a baseline, the CFO hears "we think it's better" instead of concrete numbers. Measure everything before day one.
2. Over-scoping the pilot. The instinct to automate everything at once is understandable. It's also reliably fatal. One process, proven, funds the next four. Pick the bottleneck that matters most and make it undeniable before touching anything else.
3. Measuring hours saved instead of cycle time. "We saved 200 hours" is a hard sell in a board meeting. Close time reduction and working capital impact are the numbers that fund the next phase. Track those from day one.
