Two to four years before AI pays you back
Your AI business case runs on a clock that is three to four times slower than the IT you are used to buying. Deloitte asked more than 1,800 executives about their AI investments and found payback periods of two to four years. Around 6 percent saw their money back inside twelve months. For a board that reports every ninety days, that is the whole problem in one number.
The temptation is to read that as a reason to wait. It is a reason to change how you fund and sequence the work, because the same research shows one area paying back fast and one method cutting the time to market by a factor of four. This piece is about where the money comes back, when, and what you can decide in your next portfolio review.
The trough and the peak
Gartner has placed generative AI in the trough of disillusionment while AI agents are still sitting on the peak of inflated expectations. Two technologies, one portfolio, opposite ends of the expectation curve.
That explains the strange atmosphere in a lot of boardrooms right now. The generative AI programme from eighteen months ago is being quietly defended in one agenda item, and in the next item someone presents an agent roadmap with numbers that look exactly like the ones that disappointed. Same room, same quarter, same budget.
The useful part of the curve is the timing it gives you. Generative AI in the trough means the hype premium is gone and the vendors are cheaper, the integrators are better, and the internal expectations are finally realistic. Agents on the peak mean your business case for them will be written on assumptions rather than results, so keep the commitment small and the learning loop short.
- Treat generative AI spend as a maturing capability with known failure patterns.
- Treat agent spend as an experiment with a named owner and a hard review date.
- Put both on the same portfolio sheet so the board sees one picture instead of two stories.
The learning gap behind the 95 percent
MIT looked at enterprise AI spending of roughly thirty to forty billion dollars and found that about 95 percent of pilots produced no measurable effect on profit. That number travelled fast, mostly as an argument against AI. Read the cause and it becomes an argument about how organisations work.
MIT points at the learning gap. The technology performed. What was missing was the loop: systems that keep no memory of the work they did, teams that get no feedback on what the model got right, processes that stay exactly as they were around a tool that could have changed them. A pilot that cannot learn from Monday will do the same thing on Tuesday, forever, and then get switched off in the second budget round.
A pilot that never learns from the work it does is a demo with a licence fee.
The fix is unglamorous and it sits in your operating model rather than your tech stack. Who reviews the model output every week. Where the corrections go. Which process step disappears once the accuracy holds. Boards rarely ask those three questions, and they are the ones that separate the 5 percent from the rest.
Where the money actually comes back first
Customer service is the exception in the Deloitte data: around 63 percent of those investments pay back within the first year. Set that against roughly 6 percent across all use cases and you have your sequencing decision made for you.
The reason is structural. Service is the one place where volume, cost per contact, handling time and customer satisfaction are already measured daily, by people who know the numbers by heart. You are not building a measurement system before you can prove anything. You are plugging into one that has been running for years.
The rest of the portfolio needs patience the average board does not have. Two to four years of payback, three to four times longer than the IT investments your CFO benchmarks against, and only about 28 percent of use cases fully meeting their ROI expectation. Those three numbers together tell you to fund fewer things, deeper.
The numbers worth putting on one slide
- Payback of two to four years across AI investments, based on more than 1,800 executives.
- About 6 percent of investments returning inside twelve months.
- About 63 percent of customer service investments returning inside twelve months.
- About 28 percent of use cases fully meeting the ROI expectation that was written for them.
Take that slide into your next quarterly review and the conversation changes from enthusiasm to sequencing. Start where the meter already runs. Use the returns from service to fund the slower work behind it.
Four levels of exposure, horizons up to seven years
Every part of your business sits at a different level of exposure to AI, on a horizon that runs from right now to beyond seven years. Four levels, four different conversations, one portfolio.
Boards get into trouble by treating the whole company as one horizon. Everything becomes urgent, or everything becomes a 2030 topic. Neither survives contact with a quarterly cycle. The work is to place each part of the business honestly: which revenue streams are exposed this year, which ones have three to five years of protection from regulation, physical assets or contract structures, and which ones sit beyond seven years where any plan you write today is a guess.
The near horizon earns the investment. The far horizon earns attention and option value, nothing more. And the middle is where most companies quietly lose, because the exposure is visible, the urgency is not, and the budget goes to the fire that is already burning.
Do the placement as a board, on one page, with names attached. It takes an afternoon. It is the cheapest strategic exercise available to you this year.
What method buys you
The CX+ Method delivers roughly four times faster time to market and about 68 percent more cost efficiency. On a two to four year payback, a factor of four on delivery speed is the difference between a programme that survives two budget cycles and one that gets cancelled in the second.
Speed here comes from the sequence, not from working harder. Start at the customer question that already has a number attached. Build the learning loop into the first release instead of the third. Kill use cases at week six when the data says the value is not there, and move the budget to the one that is working.
Four times faster to market is not a technology result. It comes from deciding earlier what you will stop doing.
That 68 percent cost efficiency lands in the same place the MIT research pointed: the work around the model. Fewer handovers, fewer parallel pilots, fewer teams solving the same integration twice.
What you do on Monday
- Put the four numbers on one page for your next portfolio review: two to four years payback, 6 percent inside a year, 63 percent inside a year for customer service, 28 percent meeting full ROI.
- Move your next AI investment to the service domain where the measurement system already exists, and use the return to fund the slower cases behind it.
- Give every running pilot a learning loop with a named owner: who reviews the output weekly, where corrections go, which process step disappears at what accuracy.
- Spend one board afternoon placing each revenue stream on the exposure map, from now to beyond seven years, with a name against each one.
- Set a review date for every agent experiment before it starts, and keep the commitment small enough that a cancellation costs a quarter rather than a year.