Skip to content

Notes

Machines at work: from pilot to paid deployment

Pilots are easy to start and hard to convert. Agree with the customer, before the pilot starts, how each of the five tests will be met, and it can become your first paid site.

By Alessandro Singh and Rohith Varma, Updated , 5-minute read

In short

  • Pilots of robots, sensors and autonomous systems stall when nobody has agreed what happens after them: who buys, who integrates, who certifies and who services the machine.
  • Agree with the customer, before the pilot starts, what each of the five tests will need: proof in use, a buyer who pays, a way to deploy again, a way to support it and a route through the rules.
  • Write the pilot as the first phase of a contract, with the price and terms of the rollout agreed before it starts.

Why pilots stall

In Capgemini's 2026 survey, 79% of organisations were engaging with AI-enabled robots, but only 27% were deploying or scaling them. Nearly eight in ten executives said scaling was a challenge, mainly because their organisations lacked technology and operating readiness. Most of that readiness is decided before the pilot starts.Source 1

Robotics investors and engineers see the same thing. In 2024 Innovation Endeavors wrote that robotics is rich with pilots, but very few have converted to scaled contracts. In 2023 the robotics engineer Benjie Holson wrote that starving in "pilot purgatory" is the most common way robotics startups die.Source 2Source 3

The reasons repeat. The contact sits in an innovation team with no operating budget. The pilot measures the machine, not the customer's result. Integration was scoped for one site. Nobody priced the service. The buyer's procurement process starts only when the pilot ends.

Five tests, agreed before the pilot starts

Our thesis looks at every company through five tests. Applied to a pilot, each one is something to agree with the customer before the work starts.

  • Proof in use, in the customer's terms. The pilot measures a result the operating team already cares about, such as hours saved per shift, on site and for long enough to show uptime and how the system fails.Source 3
  • A buyer who pays. The person who will sign for the rollout owns the pilot, the money for the next stage has been identified, and even a friendly customer pays something, so it has a stake in the result. A pilot fee is a start, not a pass. Pilot revenue proves little on its own: in 2024 Innovation Endeavors warned that it can exceed $1 million and still mislead. Benjie Holson notes that what a friendly customer pays for a pilot says nothing about what the robots cost to run. The price covers hardware, installation, integration, training and service at each site, with a margin left over. Supplier onboarding, the security review and the approval steps are mapped at the start and run alongside the pilot. Public buyers add formal tenders.Source 2Source 3
  • A way to deploy again. The plan covers the equipment, software and data the system must work with, who does the integration, and what changes at the next site. In May 2026 the consultancy BCG put system integration, setup and engineering at about three-quarters of the cost of conventional robotics; it gives no source for the figure.Source 4
  • A way to support it. Response times, spare parts, remote monitoring and software updates are written down, and the uptime promised is one the company can keep at the tenth site as well as the first.
  • A route through the rules. The safety case, the standards that apply and the time to certify are mapped at the start. In the EU, the Machinery Regulation applies from 20 January 2027, and safety components whose behaviour evolves through machine learning need a third-party conformity assessment.Source 5Source 6Source 7

Write the pilot as the first phase of a contract: the site, the task, the length, the measure, the price of the pilot, and the price and terms of the rollout if the measure is met. A customer who will not agree the second half in advance is telling you something.

Have it signed by the budget holder who will pay for the rollout, and price the rollout per site, so the customer can see what the tenth site costs before the first one starts.

The first paid site proves the technology works. The next one, done without the founders in the room, proves the company does: that is the test in our thesis.

Questions to ask before the next pilot

For founders planning a pilot, and for investors checking a company's pipeline, nine questions under the five tests:

  • Proof in use. What result will the operating team measure, and what number counts as success?
  • A buyer who pays. Who signs for the rollout, and have they approved the pilot plan? What happens the day the pilot ends: a purchase order, a tender or another pilot?
  • A way to deploy again. Which systems does it connect to at this site, and what changes at the next one? Who installs and integrates at the next five sites?
  • A way to support it. Who answers the call when it stops, and how fast? What does one site cost to install and run, and what does the customer pay? Who pays for the machine: the customer, a leasing or financing partner, or the company's own balance sheet?
  • A route through the rules. What must be certified, by whom, and how long will it take?

If you cannot answer these in writing before the pilot starts, it is a trial with no route to revenue.

Some pilots are right to stop. A pilot that ends with a clear no is a result: count it, and do not stretch it into a second pilot.

How we help

From the founders' careersA company grown from 2 to more than 50 people, with its commercial function built from zero.

This is the commercial work we do with companies and the funds that back them: the buyers, the route to market and procurement in regulated markets, from the first site to the next.

Updated on 12 November 2026 to follow the thesis we published that day: the five tests carry its names, and they are agreed before a pilot, not passed in one.

More notes

  • Our thesis

    The second deployment, without the founders in the room

    The second deployment is the first one done without the founders. It rests on what they wrote down and who they trained.

    For founders and investors

    Updated

    , 8-minute read

  • AI you can rely on

    Relying on a vendor's AI: the evidence to ask for

    In a 2026 survey, vendors supplied over half of UK financial firms' most material AI. Firms still answer for it: ask for evidence that repeats.

    For enterprises, investors and founders

    Updated

    , 8-minute read

All notes

Newsletter

One email a month

Our newest notes, and what we are seeing in AI, machines at work and deep technology.

We use your email only to send the monthly notes. See our privacy notice.

Sources

In the order we cite them, with the day we read each one.

  1. 1. Two-thirds of organisations rate physical AI as a high priority for the next three to five years (opens in a new tab)

    Capgemini (a consultancy's survey of 1,678 executives), , Accessed , Back to the text, where source 1 is cited

  2. 2. Lessons learned the hard way: robotics (opens in a new tab)

    Sam Smith-Eppsteiner, Innovation Endeavors (an investor's essay), , Accessed , Back to the text: citation 1 of source 2, a citation 2 of source 2, b

  3. 3. So you want to do robots, part 3: how to escape pilot purgatory (opens in a new tab)

    General Robots (Benjie Holson, a robotics engineer's essay), , Accessed , Back to the text: citation 1 of source 3, a citation 2 of source 3, b citation 3 of source 3, c

  4. 4. The CEO's guide to physical AI (opens in a new tab)

    Boston Consulting Group (a consultancy's analysis; it gives no source for the figure), , Accessed , Back to the text, where source 4 is cited

  5. 5. Regulation (EU) 2023/1230 on machinery (opens in a new tab)

    European Agency for Safety and Health at Work, , Accessed , Back to the text, where source 5 is cited

Tell us what you are working on.

A founder replies within one working day.