How long does AI take to show ROI? A timeline by use case

Most organisations wait two to four years for satisfactory ROI on a typical AI use case, according to Deloitte’s 2025 survey of 1,854 executives, and only 6% reported payback inside a year. The spread by use case is wide: AI that ends in a booked appointment or a cancellable invoice can pay back in months, while copilots and custom models sit at the long end.

Sources last checked: 23 September 2026. Every external figure on this page links to the publisher that produced it, and was re-read at that source before publication.

• The rule: time to ROI = build clock + ramp clock + conversion clock. The AI mostly sets the first two; your business sets the third.
• Short end: database reactivation and AI appointment setting, where the conversion clock is your own sales cycle, measured in your CRM.
• Middle: customer-service deflection, agency and BPO replacement, where the saving waits on a staffing plan or a contract notice period.
• Long end: employee copilots and custom predictive models. Gartner puts the average at 8 months from AI prototype to production, before any value accrues.
• Why the surveys disagree: Google Cloud reports 74% of executives saw ROI within a year; Deloitte reports 6%. They asked different people different questions.

How long does AI take to show ROI, by use case?

The table below is a planning tool, not a measured benchmark. Each planning range is derived by adding the three clocks described in the next section; where a published figure exists for that use case, it is in the evidence column and linked below the table. The column that matters is time to first revenue or saving, because that is the date your CFO will hold you to. It applies across the systems in our AI for business overview, not only to sales.

Sources for the evidence column: Klarna’s 27 February 2024 release (the US$40m is Klarna’s own estimate, published a month after launch); Brynjolfsson, Li and Raymond, Generative AI at Work; the MIT NANDA GenAI Divide report; the UK Government Digital Service Copilot findings; Gartner and Deloitte as linked in this page. For the net-new outbound row, see how long AI outbound takes to ramp.

The fastest AI payback is almost never the most sophisticated AI: it is the use case whose conversion clock is shortest.

The three-clock rule: why the same AI pays back in one quarter or three years

The three-clock rule splits any AI ROI timeline into three sequential waits, and only the first two belong to the technology.

1. Build clock — contract signed to live in production. Integration, data, security review, sign-off.
2. Ramp clock — live to a stable leading indicator: booking rate, deflection rate, documents processed without a human.
3. Conversion clock — stable indicator to cash. For revenue use cases this is your sales cycle. For cost use cases it is a notice period, a contract term or a hiring plan.

Time to first revenue = build + ramp + conversion. Payback month is later again: it is the month cumulative gross margin passes cumulative cost, which the four-number AI business case for a CFO sets out line by line. A vendor can shorten the build and ramp clocks. Nobody can shorten your conversion clock except you, which is why an AI timeline promise that never asks about your sales cycle or your contracts is not a timeline.

Which stage of an AI rollout slips the most?

In large organisations, the build clock is the one that slips. A Gartner survey of 644 respondents from organisations in the US, Germany and the UK (fourth quarter of 2023) found that on average only 48% of AI projects make it into production, and it takes 8 months to go from AI prototype to production. The same survey named difficulty estimating and demonstrating value as the top adoption barrier, cited by 49%.

Size makes it worse. The MIT NANDA report found that top-performing mid-market companies averaged 90 days from pilot to full implementation, while enterprises, defined there as firms with over US$100m in revenue, took nine months or longer. What fills those months is rarely the model. It is CRM write-back, calendar and identity integration, list extraction, territory ownership and third-party security review, each owned by a different team, and each able to stop the build clock on its own.

In an enterprise, the AI build is usually the shortest line on the timeline; integration and approvals are the long ones.

Why do AI ROI surveys disagree: 74% within a year, or 6%?

Three widely quoted studies appear to contradict each other. They measure different things, and the gap between them is itself the most useful fact about AI ROI timing.

Read together: a survey of organisations that already have AI in production, asking whether any use case returned anything, gets a high number. A survey asking when a typical use case paid back gets two to four years. A study measuring P&L impact six months after the pilot is looking before most payback periods in the Deloitte data would have ended. None of the three is wrong; quoting one without its window is.

“We’ve gone live with AI appointment setting — when will I see revenue?”

Appointment automation is the short end of the table because its ramp clock is short and its conversion clock is a number you already have: median days from first meeting to closed-won in your CRM. Here is the payback month worked end to end, with illustrative inputs to replace with your own.

• One-off setup and integration: $20,000. Running cost: $15,000 a month.
• 40 booked meetings a month; 70% held (an illustrative show rate, not a LeadsNow figure); 20% of held meetings close; $6,000 gross margin per deal.
• Gross margin per monthly cohort: 40 × 0.7 × 0.2 × $6,000 = $33,600.
• 45-day sales cycle: each cohort’s margin lands two months after its meetings.

Cumulative cost at month n is $20,000 + $15,000n; cumulative margin is $33,600 × (n − 2). At month 4 that is $80,000 of cost against $67,200 of margin. At month 5 it is $95,000 against $100,800: payback in month 5. Stretch the sales cycle to 120 days, so margin lands four months after the meetings, and the same system pays back in month 9 ($155,000 against $168,000), not month 5. The AI did not change; the conversion clock did.

The same arithmetic explains how we report our own results: LeadsNow’s 7x average sales lift is read at month 6, as trailing three-month closed-deal revenue against the three months before launch, and the methodology page discloses that the median is closer to 4x. A lift figure read at month 1 would miss most of the revenue, because the conversion clock would not have run. If the vendor bills per booked appointment or on revenue share (for AI appointment setting we charge 5–20% of the sales we help generate), the fixed-cost line changes shape; the conversion clock does not.

What AI ROI looks like when the saving is a cost, not revenue

Cost-side AI has no sales cycle, but it has an equivalent: the date a bill actually gets smaller. Klarna’s assistant handled 2.3 million conversations, two-thirds of its customer-service chats, in its first month, and Klarna said it did the equivalent work of 700 full-time agents. The profit improvement was still an estimate for the year ahead, because capacity only becomes cash when staffing or outsourced volume is reduced.

Copilots are the extreme case. The UK government’s trial of Microsoft 365 Copilot with 20,000 staff, 30 September to 31 December 2024, found an average self-reported saving of 26 minutes a day. That is real time, and it reaches the P&L only if a hiring plan, an overtime budget or a contractor line is cut to match.

An AI saving that cannot be pointed to on a specific invoice or a specific hiring plan has not yet started its conversion clock.

Frequently asked questions

How long does AI take to show ROI on average?

Most organisations report satisfactory ROI on a typical AI use case within two to four years, per Deloitte’s October 2025 survey of 1,854 executives. Use cases with a short conversion clock, such as appointment setting on existing leads, can pay back within one sales cycle of going live.

Is 12 months a realistic payback target for an AI project?

For most AI projects, no. Deloitte found only 6% of respondents reported payback in under a year, and just 13% even among the most successful projects, against the seven to 12 months typically expected of technology investments. A 12-month target is realistic only where the build, ramp and conversion clocks together fit inside the year, which in practice means a short sales cycle or a short notice period.

How long does it take an AI project to get into production?

An average of 8 months from prototype to production, and only 48% of AI projects get there, per a Gartner survey of 644 respondents published in May 2024. Google Cloud’s 2025 study found 51% of organisations with generative AI deployed took an application from idea to production in 3–6 months.

Why does my AI pilot show no ROI after six months?

Often because six months ends before the conversion clock does. The MIT NANDA report that found 95% of organisations getting zero return measured ROI impact six months after the pilot. Check whether your leading indicator is moving before concluding the pilot failed.

Does AI customer service pay back faster than other use cases?

It can, when the saving attaches to a contract or staffing line that actually shrinks. Klarna reported its AI assistant handled two-thirds of customer-service chats in its first month and estimated a US$40 million profit improvement for 2024. That figure was Klarna’s own forecast, not an audited result.