Deep tech pilots fail to convert because most are run as free experiments with no economic buyer, no written pass test and no budget line for what comes after. Fix that before the pilot starts. Charge for it, write the success criteria, name the person who signs, and agree the contract terms that apply the day it passes.
How many pilots actually become contracts?
Few pilots become contracts. AI has repeated the pattern. Gartner predicted in July 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. IDC research with Lenovo, reported by CIO.com in March 2025, found that 88% of observed proofs of concept did not reach wide deployment: for every 33 launched, four reached production.
The MIT NANDA report of July 2025 is the harshest. It found 95% of organizations getting zero return from generative AI. For enterprise-grade tools, 60 percent of organizations evaluated one, 20 percent reached a pilot and 5 percent reached production. The method was a review of over 300 public initiatives, interviews at 52 organizations and 153 survey responses, so treat the figures as directional.
Some buyers do convert. Menlo Ventures reported in December 2025 that 47% of AI deals go to production, against 25% for traditional SaaS. In manufacturing, the World Economic Forum said in June 2026 that its Global Lighthouse Network now counts 238 sites that run advanced technology at scale. The Forum describes AI at the newest sites as moving from isolated pilots to a core operating capability.
Why do pilots stall after the technology works?
A pilot can meet every technical target and still die, because the commercial questions were never asked. Seven causes show up repeatedly:
- No economic buyer. The pilot was sponsored by an engineer or an innovation team that cannot sign a production contract. When the pilot ends, nobody with budget authority has a reason to act.
- No success criteria agreed up front. "Let's see how it performs" gives the buyer no point at which to say yes. Gartner names unclear business value as one reason for abandonment.
- No budget line. The pilot came from an experiment fund. Production needs a line in next year's operating or capital plan, and nobody asked for one.
- Unpaid pilots. IDC's research, as CIO.com reported, describes these projects as "highly underfunded or not funded at all". With no fee, nobody on the buyer side has to defend ignoring the result.
- Pilot scope that cannot scale. A pilot scoped for a lab or a single line often fails to anticipate the requirements of enterprise scale. A robot cell that works on one line with a dedicated engineer tells the buyer little about a full plant.
- Integration and security review left to the end. IT, OT and security teams see the product for the first time after the pilot, and their review restarts the clock. In defense, the DoD CIO states that contractors handling controlled information must meet a specific CMMC level as a condition of contract award. Phase 1 began in November 2025, Phase II was suspended in July 2026, and self-assessments remain.
- The champion leaves. One sponsor carries the pilot. They change roles, and the project has no owner.
Size of buyer matters too. MIT NANDA found in July 2025 that top mid-market performers took 90 days from pilot to full implementation, while large enterprises took nine months or longer.
Which fix matches which cause?
| Cause | Fix | When |
|---|---|---|
| No economic buyer | Name the person who signs the production contract and get them on the kickoff call | Before the pilot is scoped |
| No success criteria | Write a short list of measurable pass tests into the pilot agreement, with the data source for each | In the pilot agreement |
| No budget line | Ask which budget pays for production and when that budget is set | First commercial call |
| Unpaid pilot | Charge a fee that credits against the first contract | In the pilot agreement |
| Scope cannot scale | Run the pilot on the buyer's real system, with their staff, at the conditions of a full rollout | Scoping |
| Late security and integration review | Send the security questionnaire and integration plan in week one | Kickoff |
| Champion leaves | Get a second sponsor and a signer above the champion | Before the pilot starts |
How do you design a pilot that converts?
Run these six steps in order. Steps 5 and 6 run in parallel with the technical work.
- Make it paid. Price the pilot so that a manager with budget authority has to approve it. That approval is your first proof that money exists. Credit the fee against the first year of the contract.
- Write the success criteria. Put each pass test in the agreement as a number, a measurement method and a date. Agree what happens if a result is close: a short extension with a new date, or a stop.
- Name the signer. Write down the person who signs the production contract and the second sponsor. Meet the signer before kickoff and again at the midpoint review.
- Pre-agree the commercial terms. Attach the production price, volume, term and start date to the pilot agreement, set to trigger when the pass tests are met. The US Defense Innovation Unit uses this structure: its guidance for commercial companies (checked October 2026) states that, under 10 USC 4022, a successful prototype agreement can lead to a follow-on production contract without further competition.
- Fix the timeline. Set an end date and a decision meeting in the calendar on day one. The 90 days that top mid-market performers needed from pilot to full implementation, in MIT NANDA's July 2025 report, is a fair ceiling for most pilots.
- Start procurement and security in parallel. Send vendor onboarding forms, the security questionnaire, insurance certificates and the master agreement draft in week one. By the decision meeting, the only open item is the signature.
If a buyer refuses steps 1 to 4, you have learned something useful early. A buyer who will not pay, write criteria or name a signer is running research, and you should price and staff it as research.
How does this change for robotics, quantum and defense?
The same six steps apply, with different signers and pass tests. In robotics, the signer is often a plant or operations leader, and site acceptance tests are the natural pass criteria; our piece on how robotics startups sell to manufacturers and warehouses covers that buyer. In quantum, many first engagements sit in a research budget, so step 3 matters most: find out early whether any business unit will own a production line item. Our article on how quantum computing companies find their first enterprise customers goes further. In defense, the prototype agreement and its follow-on clause do much of step 4 for you, while the cybersecurity requirement makes step 6 a condition of award.
What should a founder do this week?
List every open pilot. For each, write down the signer, the pass tests, the budget line and the contract terms that apply on success. Any pilot with a blank is at risk, and the next call with that buyer should fill them. If you want an outside read on your pilots and pipeline, start with The Diagnostic.