Gradient makes a window heat pump that heats and cools a room without a conventional building-wide replacement. In a New York public-housing pilot, the company's accessible installation design met a harder commercial test: how the equipment behaved across a season in occupied apartments.
The lesson for a hardware founder is to separate the gates. A fast installation can help adoption, but it does not establish seasonal reliability, low maintenance or cheaper bills. A pilot should collect evidence for each promise before a large contract is described as a completed rollout.
The window was a commercial constraint
Gradient's June 2024 engineering account describes efforts to simplify the equipment and its installation, including resident input and attention to condensation. Water created during operation is an everyday physical detail with a consequential location: it cannot simply become a problem outside the apartment.
A window-mounted system addresses a different adoption problem from a major building renovation. The product must fit existing openings, household routines and the work that maintenance staff can perform. Making that work easier is a meaningful design objective.
The company's fast-training and setup accounts are its own reports. They should not be read as a universal installed-project duration, especially when preparatory work is outside the timed step. A founder should state where the clock starts and stops.
NYCHA's August 2023 announcement describes the initial Woodside Houses window heat-pump installations and contracts with Midea and Gradient. The first installation announcement includes Midea equipment. It cannot be used as a count of Gradient units delivered, and a contract for future units is not a delivery total.
Occupied apartments supplied different evidence
A December 2025 NYSERDA evaluation examined a Gradient pilot in 12 apartments from mid-December 2023 to early September 2024. It says prototypes were removed for an updated model incorporating interface feedback and changes addressing condensate backup and corrosion.
The report also says large-scale Gradient deployment at NYCHA had slowed while outdoor condensation concerns were evaluated. That is the report's dated observation, not a claim about the company's position in October 2026.
Its weather-normalised, annualised model estimated lower net energy consumption but an annual energy-cost increase of $937 across the pilot, about $78 per apartment in 2025 US dollars. Local electricity and gas prices mattered. The estimate excluded equipment first costs and did not establish full lifecycle payback.
Those findings are not contradictory. A machine can use less energy while the energy it buys costs more. It can improve one part of the user experience while revealing another physical problem. A field pilot exists to uncover that combination.
The small setting and modeled annualisation limit generalisation. The report is stronger than an anonymous anecdote, but it is not evidence of the same bill effect in every building, tariff or climate.
Design a pilot with five separate gates
Use the following structure before choosing pilot size. It is a proposed commercialization tool, not Gradient's internal checklist.
| Gate | Evidence to collect | Decision it supports |
|---|---|---|
| Installation | Preparation, crew time and exceptions | Can ordinary planned crews repeat the work? |
| Comfort | Conditions and user feedback across relevant weather | Does the product meet the intended task? |
| Water and materials | Drainage, leakage, condensation and degradation observations | What needs redesign before volume? |
| Service | Callbacks, diagnosis and repair time | Can support scale with deliveries? |
| Economics | Local rates, usage, maintenance and capital scope | Which buyer benefit is supported? |
Each gate needs an owner and a decision rule written before the result arrives. That makes it harder to declare success by selecting only the most flattering measurement. The actual thresholds depend on the product and qualified engineering team.
The first three gates are not substitutes for applicable safety and installation requirements. This is a business evidence plan. Real installations and technical acceptance belong to the qualified people responsible for the equipment and site.
A quick install can accumulate slow callbacks
Consider a hypothetical 100-unit pilot. Assume the final installation step saves 45 minutes per unit against an alternative, at a fully loaded labor cost of US$60 an hour. Gross labor saving would be $4,500: 100 multiplied by 0.75 hours multiplied by $60.
Now assume 20 units require one extra visit costing $180 each in travel, diagnosis and work. Callback cost is $3,600, leaving $900 of that initial labor advantage. At 25 such visits, the advantage disappears. These are illustrative numbers, not Gradient's callback frequency or installation costs.
Parts, customer disruption and administration could add more. Some visits might occur even with the alternative, so a fair comparison counts incremental callbacks rather than treating every service event as a new product failure.
The purpose is not to make field service look unacceptable. It is to budget learning before scale. A founder might rationally accept high pilot service cost if it resolves a design uncertainty that would be expensive across thousands of installations.
Keep learning expenditure separate from normal delivery economics. Otherwise the pilot looks uneconomic for the wrong reason, or the scaled plan quietly assumes that experimental support will vanish.
Sell the evidenced benefit to the right purchaser
Energy use, utility cost, comfort and installation disruption are different benefits. A housing operator may value reduced building emissions even when the modeled bill does not fall. A resident may care most about controllable room temperature.
Ask who pays for equipment, energy and maintenance, and who receives each benefit. Then use the appropriate evidence. A percentage reduction in energy cannot answer a dollar-bill question without prices and usage.
Form Energy's duration economics supplies a neighboring discipline: define the customer's constraint before converting a technical advantage into a commercial promise. Here the constraint might be the existing building or installation process rather than cheap heat alone.
The Light Phone manufacturing story likewise shows why a working device needs a surrounding adoption system. For window equipment, that system includes maintenance staff, resident instructions and a response to seasonal exceptions.
Record the hardware revision in every pilot observation. A problem seen in a prototype is not automatically a defect in the later model, and a later design change does not erase the observation. Traceable versions let the team ask whether the change resolved the specific problem before expanding the claim.
Choose the next unknown before the next thousand units
Spend an hour dividing a product's launch claims into installation, operation, maintenance and economics. Beside each, write the strongest existing evidence and the missing observation. Select the uncertainty that would be most expensive to discover after volume production.
Give that uncertainty its own pilot gate. A small deployment that forces a design revision may create more value than a larger one that produces a pleasing installation statistic. The useful milestone is not merely equipment in place. It is evidence that the next installation can keep its promise through the season.
Discussion
Sign in with Google or just a name. No email link, no password to remember.