AI Unit Economics Sprint · 10 working days

Improve AI gross margin without silently degrading product quality.

For B2B software teams with production LLM or generative-AI workloads: measure total cost per successful business task, test at your quality threshold and get an implementation-ready plan.

Two founding engagements available for projects starting by September 30, 2026.

Schedule a fit call →

Duration: 10 working days
Scope: Up to 3 workflows
Founding price: $12,500 upfront

The problem

A cheaper model is not necessarily a cheaper product outcome.

Provider invoices and token prices show only part of the system. Retries, fallbacks, supporting infrastructure, orchestration, human review and failed attempts all affect what it costs to complete useful work.

Cost reductions can also move failure elsewhere: lower task completion, more review, worse latency or subtle quality loss. The sprint puts cost and quality on the same decision frame before recommending a change.

total measured workflow cost ÷ successful business tasks = cost per successful business task

Qualification

Who the sprint is for

The best fit is a B2B software company with approximately 20–300 employees, live generative-AI workloads and a decision that can be bounded to no more than three production workflows.

A good fit when

LLM or generative-AI workflows are already in production.
Model API or inference expenditure is at least $20,000/month, or an equivalent reliability or gross-margin problem is material.
A technical owner and business sponsor can make decisions during the ten working days.
Usage, cost and representative quality evidence can be inspected through an approved route.

Probably not a fit when

The workload is still a prototype without production evidence.
The request is a company-wide AI strategy or an audit of the entire AI estate.
No one can define or approve an acceptable quality threshold.
The expected outcome is implementation or a guaranteed saving rather than a diagnostic.

The process

Ten working days, from evidence to decision.

Scope, access and data-handling boundaries are agreed before day one so the diagnostic can stay focused.

Days 1–2

Define the unit and map the workflows

Agree what a successful business task means, set the quality threshold and trace the production path for each workflow, including retries, fallbacks and human review.
Days 3–5

Establish the baseline

Reconcile the available usage, billing, infrastructure and outcome evidence. Calculate cost per successful task and make assumptions or measurement gaps explicit.
Days 6–8

Test practical alternatives

Evaluate a prioritised set of relevant changes to models, routing, prompts, caching, batching or workflow design against representative cases and the agreed operational constraints.
Days 9–10

Prioritise and hand over

Turn the evidence into an ordered implementation plan, with dependencies, measurement requirements and validation checks, then hold a decision-focused read-out.

Deliverables

What your team receives

The output is a decision package grounded in your workflows and threshold, not a generic benchmark deck.

A traceable baseline

Workflow maps and a unit-economics model for each scoped workflow, including the task boundary, included costs, assumptions and known gaps.

A quality decision rule

A documented successful-task definition, quality threshold and representative evaluation set agreed with the people who own the outcome.

An alternatives test matrix

A like-for-like record of what was tested, what met the quality threshold, the operational trade-offs and where the evidence is inconclusive.

An implementation-ready plan

Prioritised changes with technical steps, dependencies, risks, instrumentation and validation checks that an engineering team can estimate and execute.

The sprint includes the final report and a stakeholder read-out. Implementation is deliberately separate, so the recommendation can be assessed before the team commits to a build.

Security posture

Keep sensitive work under customer control.

The working approach is agreed before day one and kept as narrow as the evidence allows.

Read-only and offline analysis wherever practical.
No production changes during the Sprint.
Data minimisation and anonymised samples preferred.
Sensitive work can run inside the customer’s environment.

This is a technical diagnostic, not a formal security or compliance assessment.

Scope and price

A bounded founding engagement.

$12,500 plus applicable taxes, where required. Payable upfront.

The fee is credited against a qualifying follow-on implementation.

The Sprint covers ten working days and no more than three production workflows. Implementation is not included and remains a separate engagement.

$12,500

Founding Sprint price

Next step

Start with a short qualification call.

We will confirm the workflows, available evidence, quality ownership, security constraints and start readiness. If the sprint is not the right scope, I will say so.