On October 28, Otto Group and Sereact announced a two-phase returns project in Poland. Ten dual-arm robots come first. Fifteen more follow “based on initial results.”

Under a Robotics-as-a-Service model, Sereact operates and maintains the systems. Otto’s employees take on “supervision, exception handling and quality checks.”

On November 2, Tutor Intelligence announced a $75 million Series B led by Andreessen Horowitz. The company says it sells robot work “by the hour or by the pick,” with no upfront purchase and no long-term contract.

Both models promise to shift some risk to the vendor. How much shifts depends on what the contract counts—and what remains on your side.

The Meter Has an Edge

Most coverage frames these models as risk transfer. No capex, pay for work done, easy exit. Inside the boundary the contract draws, that holds.

The boundary is where the cost hides. In Otto’s case, the announcement puts exception handling, quality checks and supervision on Otto’s side. That is people, shifts and floor space. None of it shows up in a per-unit price.

Tutor shows the other side. Trade coverage of the round describes remote human “tutors” who guide the robots. The coverage I reviewed doesn’t say how much billed work involves them, or whether the price changes as that share falls.

My read: a per-pick price may bundle some amount of the vendor’s own labor. As the buyer, you can’t see how much, or how it changes.

The Gate Has No Published Criteria

The meter also doesn’t price the decision to expand.

Otto’s second phase depends on “initial results.” The announcement gives no financial terms and no performance targets. GXO and Agility Robotics followed the same path: a late-2023 pilot, a June 2024 multi-year RaaS agreement, and now two more sites. I covered the roadmap side of that last week. Their release gives no uptime, intervention-rate, labor-hour or cost baseline.

To me, that’s the pattern. The commercial terms are moving faster than the evidence buyers use to judge them.

Why Buyers Should Care

An opex service contract can clear approval with less scrutiny than a purchased system would face. In Issue 58, I walked through the productive-hour break points that decide whether an owned robot pays back. A service contract can skip that test. You still carry the exception labor, the quality risk and the floor-space commitment.

If the billable unit is loosely defined, the vendor’s unit cost and your fully loaded unit cost can drift apart without anyone noticing.

In phased deals, the most important decision sits at the gate between phases. It is far easier to define that gate before phase one starts than after the robots are on your floor.

The Billable-Unit Test

Before signing a RaaS, per-hour or per-pick agreement, I would want six answers in writing:

1. Unit definition. Does the bill count attempted picks, completed picks, or picks that pass your quality check? Who settles disputed units, and on whose data?
2. Exception boundary. List every exception type: failed grasp, damaged item, unrecognized SKU, jam. For each, record whether the vendor or your staff resolves it. Then measure your site labor minutes per 1,000 billed units.
3. Fully loaded unit cost. Add the vendor fee to your supervision, exception, quality-check and space costs. Compare the total with your current cost per unit at peak volume, not just average.
4. Remote human content. What share of billed units involves remote human assistance? Does the price or service level change as that share falls?
5. Expansion gate. Before phase one starts, write down which metrics, over what duration and including which peak periods, trigger phase two. Write down who can decline it.
6. Exit and migration. What are the notice period and your obligations if you leave? Who pays to requalify the workflow if the vendor swaps hardware generations mid-contract?

Not all six carry equal weight. If you can’t answer the first three, the quoted price can’t be compared with your current cost. If you can’t answer the fifth, the vendor’s definition of success decides your expansion.

What to Reconsider

A rate card is not a cost model.

Ask for the exception log and a measured labor-minutes figure from a comparable site before you compare the price with anything. Write the phase-two criteria before phase one starts, including who can say no.

So when the invoice says 10,000 units, what share of that work did your own people finish?

3 Signals Worth Watching

1. The criteria behind Otto’s phase two.

The plan names “initial results” but publishes no thresholds. Watch for results or terms that explain what moves the project from 10 robots to 25.

2. How much of Tutor’s work involves remote humans.

The company says it is on track for more than 1,000 robots under contract by the end of 2027, and that is a company claim. Watch for disclosure of remote-assistance share, or customer-level results.

3. Whether GXO’s data becomes public.

Trade coverage says the economics of Digit are still being assessed, with uptime and throughput data collected before further expansion decisions. Watch for that data before the new sites start in early 2027.

If this question is active in your organization, tell me where the operating reality differs from the technology narrative.

​