Your AI ROI Timeline Is Wrong. Most Projects Break Even at Month 14.

Every AI business case I reviewed at two Top 10 banks had the same problem. The ROI timeline was a fantasy. Teams projected break-even at month 5 or 6, right in that sweet spot where executives nod and approve funding. Then reality showed up. Across 23 AI projects I tracked over four years, the median break-even point was month 14. Not because the projects failed. Because nobody accounted for the seven cost phases that happen between 'pilot approved' and 'value at scale.'

The Month 6 Myth and Where It Comes From

The month 6 projection isn't a lie. It's a math error built on three false assumptions. First, that production deployment happens on schedule. Second, that user adoption hits target within 30 days of launch. Third, that ongoing costs stay flat after go-live. In my experience, zero out of 23 projects got all three right. The average project hit two of three, and the one they missed added 4 to 9 months to the timeline.

Here's what actually happens. A team builds a compelling pilot in 8 weeks. Leadership sees a demo. The business case gets written around pilot performance, not production reality. The pilot used clean data from one department, handled 200 transactions, and had a data scientist babysitting it daily. The business case extrapolates that to 50,000 transactions across four departments with no babysitter. That gap is where your ROI timeline dies.

At one bank, we had a document processing AI that showed 73% time savings in pilot. The business case projected $2.1M in annual savings starting month 5. Actual savings in year one: $340K. Not because the AI didn't work. Because it took 7 months to integrate with the legacy document management system, 3 months to retrain staff on the new workflow, and the model needed retraining twice when it hit document types the pilot never saw. Break-even arrived at month 16.

The pattern repeated across projects. The AI worked. The timeline didn't.

The Seven Cost Phases Nobody Puts in the Business Case

I started mapping the actual cost phases after the third project blew its timeline. Every AI project I've seen goes through seven distinct cost phases, and most business cases only account for three of them.

Phase 1: Discovery and pilot (weeks 1 through 10). This is the part everyone budgets. Data exploration, model building, pilot testing. Typical cost: $80K to $150K for a mid-complexity project. Phase 2: Production engineering (weeks 8 through 20). Rebuilding the pilot for production. API integrations, security review, infrastructure provisioning. This phase costs 1.5x to 3x the pilot, and most business cases underestimate it by 40%. Phase 3: Data pipeline hardening (weeks 12 through 24). The pilot used a CSV export. Production needs a real-time feed with error handling, monitoring, and fallback logic. Budget another $50K to $120K that wasn't in the original plan.

Phase 4: User onboarding and workflow redesign (weeks 16 through 30). Training isn't a one-day session. It's a 6 to 10 week process of changing how people actually work. Phase 5: Model monitoring and drift management (ongoing from week 20). Someone has to watch the model. Alert on accuracy drops. Retrain when performance degrades. This is a permanent operational cost that most business cases list as zero. Phase 6: Scaling from one team to many (weeks 24 through 40). The first team adopted it because they helped build it. The next four teams didn't. Expect 60% of your change management effort to happen in this phase. Phase 7: Optimization (weeks 30 through 52). Tuning the model, adjusting thresholds, handling edge cases that only appear at volume.

When I added phases 4 through 7 to our business case template, the average projected break-even moved from month 6 to month 13. Closer to reality. And our CFO stopped rejecting ROI reports because the actuals finally matched the projections.

The Framework That Survives CFO Scrutiny

After getting burned on optimistic timelines, I built a three-tier ROI model that I used for every AI investment decision. Tier 1 is the Conservative Case. Take your projected savings, cut them by 40%, and push the start date out by 4 months. This becomes your committed number, the one you're willing to be held accountable for. Tier 2 is the Expected Case. Use your original projections but add the full seven cost phases. This is what you present as 'likely.' Tier 3 is the Optimistic Case. This is the original business case number. Label it clearly as best-case, and list the three to five things that must go perfectly for it to happen.

The key move: tie your funding request to the Conservative Case, not the Optimistic Case. If you need $400K and the Conservative Case shows break-even at month 14 with a 2.1x return over 24 months, that's a fundable project. If you need the Optimistic Case to justify the spend, the project isn't ready.

I also added a 'Month 6 Checkpoint' to every project. At month 6, you compare actual costs and adoption against all three tiers. If you're tracking below the Conservative Case, you have an honest conversation about whether to continue. This saved us from pouring another $300K into a fraud detection model that was tracking 60% below even the conservative projection. We killed it at month 7 instead of month 18.

One more thing that changed the game: I stopped reporting AI ROI as a single number. Instead, I reported it as cost-to-date, value-delivered-to-date, and projected-break-even-month. Finance teams love this because it looks like every other capital project report they read. The moment you make AI ROI look like a normal investment report instead of a science experiment, you get taken seriously.

How to Present a Month 14 Timeline Without Losing Funding

The obvious fear: if you tell leadership the real timeline, they won't fund the project. I had this concern too. Here's what actually happened. When I started presenting honest timelines with the three-tier model, our approval rate went up, not down. We went from a 45% approval rate with aggressive timelines to a 72% approval rate with conservative ones. Why? Because the CFO and CIO had been burned by AI projects that missed their targets. They didn't trust the month 6 projections anymore. A month 14 projection with solid math behind it was more credible than a month 6 projection with hope behind it.

The second thing that helped: showing cumulative value curves instead of single break-even points. A project that breaks even at month 14 but delivers 3.8x return by month 30 looks very different from one that breaks even at month 6 and plateaus at 1.4x. Plot both on a chart. Let leadership see the long game. Most AI projects have compounding returns because the model improves, adoption grows, and you find adjacent use cases. The month 14 project often outperforms the month 6 project by month 24.

Third, anchor the conversation in portfolio math, not single-project math. If you're running five AI projects and three hit their conservative case, one exceeds it, and one gets killed at the month 6 checkpoint, you still have a strong portfolio return. Present AI investments the way a bank presents a loan portfolio. Some individual bets underperform. The portfolio wins.

I presented this framing to a board committee in 2024. The chair, a former Goldman partner, said it was the first AI investment pitch that 'spoke finance instead of science fiction.' We got full funding for an 18-month program. Every previous request for a 6-month moonshot had been denied.

Actionable Takeaway