From Selling Machines to Guaranteeing Results: The Real Issue Is the Contract

On 3 November 2026, IIoT World published a summary of a panel from Industrial AI Summit 2026 on outcome as a service in manufacturing. The speakers were Boris Scharinger (Siemens), Jan Pingel (Ingersoll Rand) and Karthikeyan Natarajan, CEO of Infinite Uptime, moderated by Rob Tiffany of IDC. A caveat: the article is marked as sponsored by Infinite Uptime, although it states that it is editorially independent, and it was summarized with the help of AI tools. The claims should therefore be read as opinions voiced at the panel, not as independent research.

Three stages of risk transfer

The model described proceeds in stages:

• Product sale: the customer owns the machine and bears the risk.
• Availability guarantee: the supplier takes on the risk of downtime.
• Output as a service: the supplier owns the plant, bears all the operational risk, and the customer pays for what is actually produced.

The examples cited range from Rolls-Royce, which rents engines by the miles flown, to Siemens, which won a rail contract by guaranteeing the transport of one million passengers between Cologne and Düsseldorf every morning from 8 to 9. Ingersoll Rand supplies compressed air "over the fence": it installs compressors on adjacent land and feeds the air into the customer's plant.

Shifting capex to opex is not enough

According to the panel, in Europe the difference is not made by moving from an investment to a recurring fee. What counts is transferring to the supplier the risk of failure, maintenance planning and performance degradation. For small and medium gas turbines, Siemens Energy works with a 10-day maintenance window every two years and no unplanned downtime between windows. The financial exposure of a missed window is such that Siemens' internal audit was asked to review the predictive algorithms.

When predictive models underpin a contractual commitment, they become subject to liability and therefore to audit.

The cost of changing models

The problem is first of all economic. A vendor that sold a $100,000 machine and earned as much again from service over 10 years used to collect the full machine price in the first month. Under an outcome-based model, the first year brings in roughly a third of the previous revenue. Established manufacturers therefore face a multi-year revenue dip, while startups do not. Some turn to financial companies to structure the offer, but, according to the panel, it is those companies that capture much of the profit.

The data advantage

Pricing an outcome requires learning across fleets of hundreds of thousands of machines. Ingersoll Rand generates 50 to 100 insights per equipment type and thousands of predictions a week on failure probability, service needs and capacity signals: if all machines are running at 95%, for example, the customer probably needs more equipment.

This exposes a limit for third-party AI vendors: customers often forbid reusing their production data for other customers. The article cites a vendor left without a reusable product after 20 projects in three years. Manufacturers with their own installed base do not need this permission. A middle path is to share the interpretation rather than the raw data, for example "bearing issue, 10–14 days to failure".

What this means for plant managers

What follows is our own inference, not a claim made in the article. Offers based on availability or output will become more frequent, and the clauses on connectivity and data sharing will decide who can learn from plant data. Some practical checks:

• read the terms on data ownership and reuse carefully before signing;
• understand which data leaves the plant and in what form (raw or interpreted);
• maintain your own data layer, through an MES or IIoT platform, to measure availability and performance independently;
• ask how the predictive models underpinning the guarantee are validated.

Having your own data is not about distrusting the supplier, but about being able to verify what is guaranteed. For more on the technical side of predictive maintenance, see our article "Predictive Maintenance: Why It Stalled and Where the Value Lies Next".