A consulting owner I spoke to recently was puzzled. His firm had invested seriously in AI tooling, his delivery teams were demonstrably faster, his clients were happier with the turnaround, and yet his margins had barely moved. He had run the numbers three times.
His mix was 80 per cent project revenue. Once you see it through that lens, the puzzle dissolves.
What does “recurring vs project revenue with AI” actually mean?
Project revenue is paid once for a defined piece of work with a beginning and an end. Recurring revenue is paid month after month for ongoing access to a capability or relationship. What changes under AI is that your tooling, talent and operational governance now carry costs that keep running between engagements. The way your revenue is shaped decides whether those costs amortise or accumulate on the P&L.
The cost structure is the part many owners underestimate. SFAI Labs’ implementation research suggests a typical mid-sized AI project costs around 100,000 US dollars to build with another 25,000 in annual maintenance, and that is before you count the AI-capable hires HybridHero reports as roughly 28 per cent more expensive than equivalent technology roles. TSIA’s State of Managed Services 2026 has a phrase worth holding onto: operational debt. An AI system never quite finishes. Models drift, prompts need refining, governance tightens, vendor terms change. In a recurring model, that ongoing cost has somewhere natural to live. In a project model, it has nowhere to go.
Why does this matter for your business?
It matters because AI is reshaping the economics underneath your P&L faster than the spreadsheet reveals. The visible part is delivery speed, and that is real. The invisible part is a permanent layer of cost in tooling, talent and governance that your revenue model may not be designed to absorb. Recurring revenue amortises that cost across many months and many clients. Project revenue lets it erode the margin on each engagement in turn.
The numbers across the research are consistent. Deloitte’s 2025 global AI survey found that 85 per cent of organisations increased AI investment in the previous year, yet only six per cent reported payback in under twelve months. PwC’s research on the same window found 81 per cent of C-suite leaders saying meaningful returns are at least a year away. An MIT NANDA analysis of 300 public AI projects found roughly 95 per cent of organisations reported no measurable return yet on their AI investments. AI delivers its compounding return when it lives inside a workflow that recurs.
SFAI Labs’ benchmarks show the upside cleanly. Well-scoped AI projects deliver positive ROI in six to nine months and quick-win use cases like document processing or support automation in three to six. Notice the framing. Those are scenarios where the AI is embedded in something ongoing.
Where will you actually meet this in practice?
You will meet it in the gap between what your delivery team tells you and what your management accounts tell you. Delivery reports speed, capacity, fewer late nights. Finance reports flat or worse gross margins, rising tool spend, and a wage bill that has crept up to bring AI-capable people on board. That gap is the project-resets-AI dynamic showing up in your books, and it tends to widen month by month until something gets redesigned.
You will also meet it in the client conversation. AI has trained your better clients to expect speed, and a meaningful share of them now ask for the discount that comes with it. HybridHero’s review of consulting firm margins names this directly: mid-market and boutique firms are absorbing the AI tooling and talent cost while clients increasingly push for fee reductions justified by AI efficiency. The faster you deliver, the harder the next pricing conversation gets, unless you have rebuilt the offer so the client is paying for the ongoing outcome rather than the hours saved.
Operators in managed services have run furthest with the implication. ITPro’s 2026 analysis describes MSPs moving away from reselling AI tools, where margins are thin, and instead embedding AI into their internal service desks and remote monitoring, then repackaging those AI-enabled workflows as outcome-priced managed services. The same logic is playing out in accountancy through Thomson Reuters’ subscription-led pricing and in agencies through the retainer mix that Predictable Profits found at 90 per cent of digital firms.
When should you act on this, and when can you leave the mix alone?
Act when your recurring share sits below roughly a third and you are investing seriously in AI tooling or hiring AI-capable people. That is the threshold where the maths gets materially harder, because there is too little stable revenue to amortise the costs against. Leave the mix alone if you are above 60 per cent recurring and margins are tracking sector benchmarks. Many firms sit between the two, and the question is one of pace.
Three moves carry the practical weight. The first is to productise the repeatable slice of your work and put a retainer around it. Parallax’s research on productised services reports that best-in-class professional services firms now generate around 80 per cent of revenue through productised offerings, which is the same shape SFAI’s high-payback AI use cases take: standardised, reusable, refined over time. The second move is to package follow-up implementation as recurring support rather than letting it disappear at project close. The post-engagement tuning, monitoring and optimisation, all the things that already sit in your operational debt anyway, become billable when you frame them as the ongoing service they already are.
The third move is to restructure the pricing conversation around an outcome that genuinely needs continued attention, then bill for that attention monthly. Bain’s 2025 technology deals research identifies value-based pricing tied to measurable outcomes as the emerging standard for both software and services, and AI is what makes the outcomes measurable in the first place. When to leave it alone is equally important. Some work is genuinely one-off. A merger integration, a one-shot system migration, a fixed-scope diagnostic. Forcing a retainer onto work that has not earned the right to recur is how you end up with the David C. Baker critique of monthly recurring revenue: relationships that amount to selling time on a calendar, which clients resent and providers under-serve.
Related concepts worth holding alongside this
Three ideas sit close to this one and are worth knowing by name. Productisation is the structural bridge between bespoke project work and a service that repeats. Outcome-based pricing is the language for billing the recurring relationship around what the client gets rather than the hours you spend. Managed services thinking, especially TSIA’s framing of AI operational debt, names the ongoing cost layer your AI investment now carries.
Revenue mix benchmarks are worth keeping in view as you redesign. Teamwork’s agency profitability work suggests 60 to 70 per cent retainer is the healthy zone for professional services. Predictable Profits finds that crossing 60 per cent recurring substantially improves the odds of a revenue-multiple valuation at exit. Mosaic’s consulting benchmarks point to EBITDA margin above 20 per cent, gross margin above 50 per cent, utilisation in the 75 to 80 per cent band, and overhead near 30 per cent of revenue as the operational shape that lets a firm absorb AI investment without it eating the bottom line.
The thread that runs through all of these is the same. AI is a cost layer you take on, not a tool you install. The shape of your revenue decides whether that cost layer compounds into a moat or accumulates into a drag. The consulting owner I started with did not have an AI problem. He had a revenue architecture problem, and the AI was making it visible. Book a conversation if you want to look at your own mix and work out what to do next.
