An enterprise contract should be rejected when fulfilling it would change the product’s purpose, users, or ethical boundary. Before Monday, the founder needs to separate the buyer’s legitimate need from the surveillance mechanism written into the deal, then offer a smaller version the team can defend.
In 2018, Google faced a larger version of that decision. Its work on Project Maven, a US Department of Defense initiative that used artificial intelligence to analyse drone footage, had triggered opposition inside the company. Employees signed a letter objecting to the project, and some resigned. The contract brought revenue and a serious customer, but it also forced Google to answer a harder question: what kinds of systems was it willing to build?
The requirement that changes the buyer
The AI networking product had been built to help employees find relevant people across an organisation. A product manager in Berlin could identify a colleague in Accra who had solved a similar problem. A new hire in Lagos could find the right person without asking through four layers of management.
The enterprise requirement altered that relationship.
The buyer wanted administrators to see who employees contacted, how often they interacted, which introductions they ignored, and how their networking behaviour changed over time. Those records would feed individual reports available to management.
Technically, this looked achievable. The product already handled connections and interaction events. The team could add reporting, permissions, retention rules, exports, and alerts.
But the user would change. Employees had been the users; management would become the real user, while employees became the observed population. Every product decision after that would drift towards richer monitoring because the buyer, renewal, and product roadmap would all point in the same direction.
That is the first test I would run on Friday: after accepting this requirement, who does the product primarily serve?
If the honest answer changes, the contract has proposed a new product.
Revenue does not contain the full cost
A founder with limited runway cannot discuss this as a clean ethics exercise. The contract might fund two engineers, extend the company’s operating time, and provide the reference customer needed for the next sale. Declining it could mean returning to smaller deals while payroll keeps moving.
The pressure is real. So is the implementation cost hiding outside the statement of work.
Employee surveillance introduces questions the current product may never have needed to answer. What exactly can a manager infer from a missed introduction? Can an employee challenge an inaccurate report? How long does the company retain interaction data? Can an administrator quietly expand monitoring after deployment? What happens when data collected in one office becomes visible from another country?
Those questions create product work, legal review, support cases, access controls, and sales commitments. The surveillance feature will keep asking for attention after the first invoice clears.
This is where I would write two roadmaps before making the call. One would show the next six months without the contract. The other would include every capability needed to deliver, govern, explain, and support the requirement. The comparison should include postponed customer work and founder time, not only engineering estimates.
A contract can extend cash runway while shortening product runway.
Project Maven forced the boundary into writing
Google’s Project Maven decision did not remain an internal disagreement over one contract. In June 2018, Google said it would not renew the contract when it expired. Sundar Pichai also published the company’s AI principles, including areas Google said it would not pursue.
The New York Times documented the employee opposition and Google’s decision not to renew. The important mechanism was larger than the outcome of a single procurement process. A customer request had exposed a boundary that the company had not made explicit enough before the work began.
A small AI company has less room for ambiguity. It cannot assume that a difficult requirement will remain isolated inside one customer account. Once the team builds manager dashboards, employee-level scores, and monitoring exports, the next enterprise prospect will see those capabilities. Sales will learn to demonstrate them. Engineering will maintain them. The product’s identity will change through ordinary commercial repetition.
That is the direct connection to the Monday deadline. The founder is deciding which future requests will become easier to accept after this one.
Turn the deadline into a narrower offer
I would not spend the weekend writing a broad values statement. I would convert the boundary into contract language and a product alternative.
The buyer may have a defensible underlying concern. Perhaps employees are failing to find expertise across departments. Perhaps introductions happen but produce no useful collaboration. Perhaps leadership cannot tell whether the programme deserves continued funding.
Those needs can be measured without individual surveillance. The founder could offer aggregate adoption reporting, voluntary feedback after introductions, department-level patterns with minimum group sizes, or a time-limited pilot that excludes individual performance reports. The proposal should state which data will not be collected and which inferences the system will not make.
This resembles the move in Chinedu’s smaller enterprise pilot: reduce the unknowns before they become permanent obligations. If cross-border access enters the design, the questions raised by a Berlin support dashboard also belong in the contract review.
If the customer refuses every version that protects employees from individual monitoring, the surveillance is the requirement. The founder then has a clear answer.
Before Monday, write one page with the prohibited capability, the buyer need beneath it, the acceptable alternative, and the revenue you are prepared to lose. Google needed employee resignations and public conflict before its boundary became explicit. A six-person team should put the boundary in writing before the first surveillance dashboard reaches the backlog.
Comments
No comments yet.