A forward-deployed team can be fully booked and still allocate its capacity badly.

Every engineer may have a customer. Every account team may have an urgent request. Yet the portfolio can still fill with weak workflows, unresolved dependencies, work the customer could self-serve, and exceptions that teach the product nothing.

Utilization answers whether people are assigned. It does not answer whether those assignments deserve the full delivery system around them. I propose a different unit: the qualified deployment, a bounded workflow with enough importance, product fit, readiness, evidence potential, and ownership to justify embedded attention.

This is an operating model, not a disclosed industry benchmark. The public evidence supports the need for qualification without validating a universal score. An official Anthropic program-lead role says demand for its FDE work exceeds capacity and includes qualification, prioritization, partner routing, legal and finance coordination, and handoff. That capacity statement belongs to Anthropic; it should not be generalized to every provider. OpenAI's FDE description connects discovery and rollout to adoption, evaluation feedback, and reusable product learning. Practitioners from Ramp and Cursor also describe active scoping and self-service gates.

Qualification is part of delivery design because accepting one deployment consumes the option to pursue another.

Five gates, not a weighted score

The following gates are proposed practice. Do not total them into a synthetic score or pretend that a high rating in one cancels a hard failure in another.

A consequential workflow

A named owner can describe the current path, the affected users, the friction, and the consequence of leaving it unchanged. “We need an AI strategy” is not yet a deployment. A bounded document-intake, service, retrieval, or decision-support workflow can be.

If no one can define the workflow or make its business decisions, route the work to discovery—not an embedded build.

Product fit and reusable footing

The proposed solution should stand on real product capability. Customization may be necessary, but the boundary between reusable product, field-created delivery asset, customer configuration, and one-off code must be visible.

If the request requires rebuilding the platform around one account, product leadership must decide whether the exception is strategic learning, separately governed custom work, or a stop.

Dependencies that can become decisions

Data, systems, users, security, procurement, and legal review need named owners and plausible dates. Unknowns are not automatic disqualifiers. Unowned unknowns are.

A short diagnostic can convert uncertainty into a build route. Production capacity should not be reserved on the assumption that access and authority will eventually appear.

Evidence and learning value

The deployment needs a baseline, representative cases, an acceptance owner, and an expansion-or-stop rule. For a provider, it may also yield reusable evaluations, product requirements, or tooling. For the client, it should improve the ability to own the next decision.

If the team cannot say what the work should prove, urgency is substituting for qualification.

Ownership and transfer

Name the business owner, technical owner, acceptance authority, and post-deployment operator. Define which artifacts transfer and what residual support remains.

Transfer by design does not promise that support disappears. It prevents permanent dependency from being discovered after release.

The output is a route

A qualification gate should produce one of four routes.

Self-service fits when the platform is suitable, the workflow is understood, and the customer's team can configure and operate it with standard guidance.

Direct FDE fits when the deployment is strategically important, technically difficult, product-relevant, and benefits from a short escalation or feedback path to the provider.

Partner delivery fits when industry context, integrations, change, or sustained implementation capacity outweighs the need for continuous direct-provider attention.

Stop or reshape fits when the workflow, owner, rights, product footing, evidence path, or operating conditions are not credible. A diagnostic may resolve the gap. Otherwise, stopping protects both capacity and trust.

Routes can change. Discovery can make partner delivery feasible. A partner can produce evidence that earns direct provider attention. A client can learn enough to move from embedded support to self-service.

Keep a qualified deployment record

Before the work enters the schedule, record:

  • the bounded workflow and consequence;
  • business, technical, and acceptance owners;
  • reusable product footing and known exceptions;
  • critical dependencies with decision dates;
  • evidence plan and next decision;
  • chosen route and rationale;
  • transfer destination and stop condition.

This is not a sales scorecard. It is the portfolio record that explains why the delivery system accepted the work and what would change its route.

Review those records as a portfolio, not only at intake. Dependencies slip, evidence changes, and product capabilities move. A recurring capacity review should confirm the route, record new learning, and release attention when the original qualification no longer holds.

A defensible stop protects the portfolio

Commercial systems often treat a stop as lost momentum. A delivery system should recognize an evidence-based stop as useful. It is cheaper than learning through senior engineering, product escalation, and production risk that the workflow has no owner or the platform does not fit.

Clear routing also protects credibility. When every strategic-sounding request receives embedded attention, customers learn that qualification is theater. When the field team can explain the route, required evidence, and conditions for reconsideration, the portfolio becomes governable.

The scheduling question comes last. First decide whether the request is a qualified deployment, what it should prove, and who will own the result. Make the route visible to sales, product, delivery, and the client. Then place scarce capacity where it can create working evidence instead of a fuller queue.