Robotics sits inside hardware go-to-market and then breaks three of its assumptions. The unit the customer actually buys is not the machine. Expansion does not compound the way it does in software, because every new site reopens questions the last site already settled. And the people who can stop a deployment mostly do not report to the person who wanted it. This note is about those three, in that order. The practice behind it lives at GTM consulting for deep tech and hardware.
Sell utilization, because that is what gets renewed.
Avatar Robotics announced a seed round on August 5, 2026, and the operating detail matters more than the round. Since launching in December 2025 the company says it has packed, sorted and helped ship more than 900,000 products, including for a leading global beauty retailer, through robots deployed in live customer environments, and it has begun a post-pilot expansion with a large warehouse operator. Its stated model puts remote human operators in the loop so customers get labor capacity immediately while autonomy improves on the data those deployments generate.
Whatever you think of the margin structure, notice what the customer is being sold: work completed, starting now. That framing changes which number decides the renewal. A machine sale is graded once, at installation, on whether the thing works. A capacity relationship is graded continuously on how much the machine actually ran, and utilization is where robotics deals quietly die a year after everyone declared them a success.
You sell a machine once. You sell its utilization every quarter.
Utilization is not a product metric. It is the sum of changeover time, exception handling, operator availability, maintenance windows and the days a line ran a different job. A robotics company that reports uptime and lets the customer discover utilization on their own has handed the renewal conversation to somebody in operations who has been keeping a spreadsheet. Instrument it first, report it monthly, and put the exception rate in front of the customer before they raise it.
The counterpoint deserves stating rather than buried. Keeping humans in the loop, as Avatar does, caps gross margin until autonomy carries the load, and a customer who buys capacity can stop buying it at the end of a term in a way that a purchased machine cannot be unbought. That is a genuine risk to enterprise value. It is also a structure that converts, and a purchased machine that never leaves the pilot bay converts at nothing.
Expansion is site by site, and that breaks the software model.
The expansion motion imported from software assumes the second unit is nearly free to sell. One integration, one security review, one contract, then seats. Robotics inverts it. The second cell on the same line is close to free. The first cell at a new site is close to a new sale, because a new site brings its own layout, its own throughput profile, its own safety review, its own maintenance crew, its own union or works council relationship, and often its own budget holder.
Three practical consequences follow, and each one is a planning error worth catching early.
Net revenue retention will look wrong. A robotics account expands in steps rather than curves, and the steps are timed by the customer's site calendar. Reporting a retention rate against a software benchmark invites a board conversation about a problem the company does not have.
Pipeline coverage built on one average deal has no referent. Coverage is arithmetic: required qualified pipeline is the inverse of the win rate on that pipeline, and at manufacturing's 19 percent median win rate that implies roughly 5.3x (Hardware GTM Benchmarks 2026, derived). The arithmetic holds. The inputs do not, because one signature is a multi-site capacity agreement and the next is a single cell, and averaging them describes neither.
The reference that wins the next site is operational, not logotypical. A named customer logo moves a first meeting. What moves a second site inside that same customer is the first site's utilization number and the name of the operations manager who will take the call. Collect the second thing deliberately, because nobody hands it over on request a year later.
Five people can stop it, and none of them report to your champion.
A typical B2B buying group runs 6.3 to 6.8 people (Hardware GTM Benchmarks 2026). On a robotics deployment the roster is unusually predictable, and each seat holds a different veto that the others never raise.
Operations owns the throughput target and loses production time to the trial. Safety or EHS owns the risk assessment, the guarding and the interlocks, and can hold a deployment for a review cycle measured in months. Maintenance owns what happens at 2am when the cell faults, and will ask about spares, mean time to repair and who gets called. IT or OT owns the network segment the robot sits on and will treat it as an attack surface. Finance owns the comparison against what the work costs today.
Most robotics pitches are built for the first seat and answer none of the other four. That is a large part of why median pilot-to-production conversion across hardware is 12 percent, roughly one pilot in eight, with most of the rest dying at the handoff from an innovation budget to a business unit (Hardware GTM Benchmarks 2026, IDC 2025). The hardware buying committee maps the general case; the robotics-specific addition is that safety and maintenance are not procedural formalities here, they are the two seats most likely to add a quarter to the calendar.
What a staged agreement looks like when a large buyer writes one.
On August 6, 2026 HII signed performance-based production agreements with Path Robotics and GrayMatter Robotics covering autonomous welding, grinding, blasting, painting, assembly and inspection across US Navy shipbuilding programs. The agreements run across seven years in two stages. In the development stage both companies qualify high-precision production techniques with HII and integrate them into an autonomous production line. In the delivery stage HII begins sourcing shipbuilding work through that line, contingent on favorable cost, schedule and quality performance.
The structure is the transferable part, and a company a fraction of that size can copy it. A qualification stage with defined readiness milestones, then a delivery stage priced on output and conditional on the first stage performing, lets a buyer commit to a direction before committing to a volume. That is usually the concession that unlocks a committee which cannot yet agree on a number, and it converts a binary yes-or-no into a sequence with a next step.
The same release notes that in 2026 HII plans to outsource more than 2.5 million hours of shipbuilding work, a 30 percent increase from 2025. The robots are being measured against hours the buyer already knows how to purchase elsewhere, which is a budget question rather than a product one, and it has its own note: find the budget line that already exists.
What to build instead: four moves.
Quote in the customer's operating unit. Hours, throughput, units processed, square feet covered, whatever the operations team already reports on a Monday. GrayMatter Robotics states it has processed over 30 million square feet of surface area across more than 20 industries and delivers up to 12x the throughput of skilled manual labor with a 95 percent reduction in rework. Those are commercial instruments as much as engineering ones, because they are denominated in something the buyer already measures.
Instrument utilization from day one and report it unprompted. Uptime is the number that flatters the vendor. Utilization is the number that decides the renewal. Own the second one publicly.
Write the production trigger into the pilot at signature. Name the success metric, the person who owns the rollout decision, the budget it comes from, and the date. A pilot that ends with a positive technical report and no named budget has not produced an opportunity, it has produced a reference the buyer can use with somebody else. Pilot to production walks the mechanism.
Run safety and maintenance as first-call stakeholders, not as sign-off. Bring the risk assessment, the guarding plan, the spares list and the mean-time-to-repair answer to the second meeting rather than the eighth. The seats that can add a quarter to your calendar should meet you before they are asked to approve anything.
Three questions, answered straight.
What does robotics GTM consulting actually cover?
The work defines the unit the customer is buying and prices in it; instruments utilization so the renewal is defended with data rather than argued; separates single-cell deals from multi-site capacity agreements in the forecast so coverage means something; maps the operations, safety, maintenance, IT and finance seats that each hold a veto; and writes the production trigger, decision owner and rollout budget into the pilot at signature. The firm-level engagement is described on the GTM consulting page, and the earlier-stage version is go-to-market consulting for hardware startups.
Why do robotics pilots stall after a successful technical trial?
Because the trial answered an engineering question and the rollout needs a budget answer, and those are held by different people on different cycles. Median pilot-to-production conversion is 12 percent (IDC 2025). Conversion is largely decided at pilot signature, when the success metric, the production budget and the decision date are either written into the agreement or left to be argued about later.
Why does a robotics company grow more slowly than its logo list suggests?
Because expansion is site by site rather than seat by seat. Each new site reopens layout, safety review, integration, maintenance cover and often a different budget holder, so an account expands in steps timed by the customer's site calendar rather than in a curve. The fix is to plan capacity and coverage against sites, and to collect each site's utilization number and operations reference before asking for the next one.
Keep reading.
This note sits inside a published method. Start with the page that ties it together, or go straight to the piece that names your stall point.