Why most enterprise humanoid pilots fail in month three

Across the enterprise humanoid pilots we have observed, the ones that fail tend to fail at the same point in the calendar and for the same reasons. It is almost never the platform. It is almost always the pilot design.

A humanoid pilot at an enterprise site is a specific artefact: three to six months, one platform, one operational role, a defined success metric, and an executive sponsor whose credibility is tied to the outcome. Enterprise pilots are how new operational technologies enter organisations, and enterprise pilots are also how organisations decide to stop pursuing operational technologies that failed to justify continued investment. Getting the pilot design right is therefore the most consequential decision the operations team makes about the entire humanoid programme, and getting it wrong ends the programme.

What we have observed across the pilots we have advised on or witnessed from adjacent engagements is that the pilots that fail tend to fail at the same point in the calendar. Not month one, when the deployment is new and everyone is on their best behaviour. Not month six, when the platform has settled into its role. Month three, when the initial enthusiasm has faded and the operational team is looking at the platform's actual return against the initial expectations.

The two failure modes we see repeatedly

The first failure mode is the metric mismatch. The pilot's success metric was set by the executive sponsor to be aspirational rather than realistic, and by month three the operational team can see clearly that the platform will not hit the metric. The team then either escalates the failure to the executive sponsor (which terminates the pilot early) or quietly redefines the metric (which typically triggers a re-review of the pilot at month six by an unimpressed executive team).

The second failure mode is the role drift. The pilot began with a defined operational role and a defined success metric for that role. By month three, the platform's role has drifted, usually because the on-site staff have discovered adjacent tasks the platform could plausibly perform, or because the executive sponsor has decided the initial role is too narrow to justify the deployment. Role drift feels productive at the time. It is almost always fatal to the pilot's outcome, because the platform is now being measured against a role its selection was not optimised for.

The pilot designs that succeed

The pilots we have seen succeed share three properties. First, the success metric is quantitative, modest, and measurable weekly. Not "improved guest experience" or "operational efficiency"; something specific like "reduced average check-in wait time by seventy-two seconds during the 4pm to 7pm arrival window." Modest, specific, and reportable.

Second, the pilot's role is defined precisely and defended against drift. The executive sponsor and the operations team both commit at the outset to not expanding the role during the pilot period, regardless of what adjacent tasks the platform could plausibly perform. Expansion is a separate decision, made after the pilot, based on the pilot's outcome. This discipline is difficult to maintain and is one of the things a good pilot design specifies in writing before the platform arrives.

Third, the pilot's timeline includes a formal midpoint review at month three, at which the executive sponsor and the operations team assess the platform's performance against the success metric and decide either to continue, to modify the metric with the executive sponsor's explicit sign-off, or to terminate the pilot. The midpoint review is what prevents the two failure modes above. Without it, the pilot drifts into either quiet failure or unproductive escalation.

The implication for the platform selection

None of the above is about the platform. The platforms currently in the market are, for the most part, capable of performing the roles that a well-designed pilot commissions them for. Platform selection matters at the margin, but at the pilot level, the design of the pilot itself matters far more than the specific platform chosen. Operators considering their first humanoid pilot should spend more time on the pilot design than they spend on the platform shortlist, and they should assume that any of the top five platforms in their category will produce a defensible pilot outcome if the pilot is designed properly.

This is not the message the platform vendors will convey. It is not in their interest to convey it. It is the message the operator should internalise before commissioning any pilot at all, and it is a substantial part of what our Assess engagement is designed to deliver.