What do you actually own at the end of an AI consulting engagement?

Every artifact produced during the engagement: the code in your repository under your organization, the brand spec and compliance overlay rules, every system prompt and the scoring rubric the gate runs, the auto-publish thresholds and routing configuration with the reasoning behind each number, the per-agent runbooks, the vendor evaluations and contracts, the recorded training sessions, and the golden set of approved and rejected outputs that makes the gate accurate. The golden set matters most and is the one most often retained by vendors, because it is what makes a replacement expensive.

What is the difference between the three tiers of an AI engagement, and which is wrong for me?

Tier 1 is a diagnostic over roughly three weeks ending in a ranked build sequence and one decision. It is wrong for operators who have already run a serious internal assessment. Tier 2 is a defined-scope build over four to eight weeks with a thirty-day operating tail; it is wrong for operators with no foundation layer and no appetite to build one. Tier 3 is an embedded fractional operator, one to two days per week with a six-month minimum; it is wrong for operations with no marketing team to embed inside, because fractional leadership multiplies a team rather than replacing one.

What questions should we answer internally before a first AI consulting call?

Twelve, and the most diagnostic are these. Export your location list from every system that holds one and count how many disagree on hours — that number is your real starting position. Name the system you consider the source of truth for a location record, then check whether anyone else in the building disagrees. Identify who approves a customer-facing message today and how long approval takes at the median. Name who would own the system after an engagement ends; if the answer is nobody, that is the first thing to solve, before any tooling decision.

What will an AI consultant decline to do?

We do not resell software or take vendor referral fees, so vendor recommendations are ours to defend and yours to overrule. We do not run engagements where the client does not own the artifacts. We do not deploy a surface agent onto no foundation because a timeline is tight, since that is the exact failure mode the method exists to prevent. We do not take engagements outside multi-unit franchise, multi-location retail, or DTC ecommerce, because the method depends on operator knowledge that does not transfer.

How is a fractional CMO engagement structured week to week?

One to two days per week embedded inside the marketing operation, with a six-month minimum. Month one covers the diagnostic and the first foundation agent against the same acceptance tests used in a defined-scope build. Months two and three deploy surface agents in dependency order, with each one measurably faster to stand up than the last. Months four through six add the measurement layer, per-location benchmarking, and the operating rhythm the team keeps afterward. Throughout, vendor selection and negotiation, team training, and a documented runbook that makes the role transferable to a permanent hire.

Scope sheet · Free · Before you talk to anyone

The Scope and Sequence Sheet

Every deliverable, the week it lands, what you own at the end, and the twelve questions worth answering before any call. It is written down so you can hold it against anything else on your desk.

Published August 21, 2026· 9-minute read

Why this exists

Consulting scope is usually vague on purpose. Vagueness protects the seller, because a proposal that names deliverables and dates can be held to them, and a proposal that promises transformation cannot. The practical result is that you cannot put two proposals side by side and compare them, which is exactly the condition under which the more confident presenter wins.

So here is ours, in writing, before any conversation. Use it as a template against other vendors if that is more useful to you than working with us. A scope sheet you can hold someone to is worth more than a proposal you cannot.

The three shapes, including who each one is wrong for

Scoping conversations go badly when the buyer and the seller are describing different shapes. Read the “wrong for” line first — it is the fastest way to rule two of these out.

Tier 1

AI Readiness Assessment

2-3 weeks

What it is: A diagnostic. Three weeks, three deliverable checkpoints, one decision at the end.

What it is not: Not an implementation. Nothing gets built. If you already know what to build and want it built, this is the wrong tier and we will say so.

Wrong for: Operators who have already run a serious internal assessment and have a ranked build list. You would be paying to be told what you know.

Week 1 — Source-system inventory and buyer-research review. Every system that holds a location record, a customer record, or a piece of brand truth gets catalogued, including the spreadsheets nobody admits to.
Week 2 — The four-question shape diagnostic and the five-component orchestration audit, run against your actual stack rather than a maturity model.
Week 3 — Written deliverable and the decision conversation. You leave with a ranked build sequence and an explicit recommendation, including the recommendation not to proceed if that is the honest answer.

Tier 2

AI Swarm Setup Sprint

4-8 weeks

What it is: A build. Defined scope, fixed timeline, with a thirty-day operating tail after handoff so your team is not alone the first month.

What it is not: Not an ongoing retainer, and not a staff-augmentation arrangement. It ends on a date, with your team operating what was built.

Wrong for: Operators without a foundation layer and without the appetite to build one. A surface agent on no foundation is a demo that strands at ten locations.

Weeks 1-2 — Foundation. Master record reconciliation and the brand spec written as scoreable dimensions rather than adjectives.
Weeks 3-4 — Compliance overlay and governance routing. Auto-publish thresholds set as numbers, with the escalation tiers written down.
Weeks 5-6 — First surface agent live against the foundation, gated, with the acceptance test passing before anything reaches a customer.
Weeks 7-8 — Emit layer, telemetry, and handoff documentation. Your team runs an output end-to-end while we watch, not the other way around.
Thirty-day tail — We stay reachable on the operating questions that only surface once real volume hits.

Tier 3

Fractional CMO with AI Swarm

6-month minimum, 1-2 days per week embedded

What it is: An embedded operator. Someone inside your marketing operation who owns the swarm as it grows, sits in your meetings, and is accountable for the number.

What it is not: Not an advisory retainer where you get opinions on a monthly call. The commitment is days per week inside the operation.

Wrong for: Operations with no marketing team to embed inside. Fractional leadership multiplies a team; it does not replace one.

Month 1 — Diagnostic and the first foundation agent, running against the same acceptance tests as Tier 2.
Months 2-3 — Surface agents deployed in dependency order, each one measurably faster to stand up than the last.
Months 4-6 — Measurement layer, per-location benchmarking, and the operating rhythm your team keeps after the engagement.
Throughout — Vendor selection and negotiation, team training, and the documented runbook that makes the role transferable to a permanent hire.

What you own at the end

All of it. The list below is not a summary of a longer list — it is the list. The last item is the one that matters most and the one most often quietly retained elsewhere, because it is what makes a vendor expensive to replace.

Code: Every integration, pipeline, and script written during the engagement, in your repository under your organization, committed as it is written rather than delivered as a zip at the end.
Prompts and specs: The brand spec, the compliance overlay rules, every system prompt, and the scoring rubric the gate runs. These are the assets that took the longest to get right and the ones most vendors treat as proprietary.
Configs and thresholds: Auto-publish thresholds, routing tiers, escalation rules, and the reasoning behind each number — so a successor can change them on purpose instead of guessing.
Playbooks: The runbook for each agent: what it does, what it reads, what it emits, how to pause it, and what breaks downstream when you do.
Vendor selections: The evaluation, the scoring, the contracts, and the reasoning. No referral fees, no reseller margin, no vendor relationship that survives the engagement.
Training: Recorded working sessions with your team, not a slide deck. The test is that your team runs an output end-to-end while we watch.
The golden set: The accumulated set of outputs you approved and rejected, which is what makes the gate accurate. It compounds over the engagement and it stays with you.

If an engagement ends early, for any reason and on either side, everything produced up to that point is already in your repository and your systems. There is no final handover event that has to happen for you to keep what you paid for.

The twelve questions

Answer these internally before speaking to anyone — us or otherwise. Most are answerable in an afternoon, and several will surface a problem worth fixing regardless of whether you hire help. The first conversation then starts at the real problem instead of spending an hour establishing what is true.

1How many locations, and how many of them have a manager who currently writes anything customer-facing?
2Which system do you consider the source of truth for a location record — and does anyone else in the building disagree with that answer?
3Export your location list from every system that holds one. How many disagree on hours?
4Which customer-facing surface produces the most volume per week: reviews, calls, posts, listings, or email?
5Who currently approves a customer-facing message, and how long does approval take at the median?
6What may not be said, and who decides? Name the regulation or the policy, not the department.
7Have you bought an AI tool in the last eighteen months? What is it doing today, and who owns it?
8When a location changes its hours, how many systems does someone update by hand?
9What does the marketing team measure weekly right now, and who reads it?
10If one surface got twice as fast at no loss of quality, which one would move the number you are accountable for?
11Who on your team would own this after an engagement ends? If the answer is nobody, that is the first thing to solve.
12What has to be true in six months for this to have been obviously worth doing?

Questions three and eleven do the most work. Three tells you the size of the foundation problem in a single number. Eleven tells you whether anything built will survive the engagement.

What we decline

We do not resell software or take vendor referral fees. Vendor recommendations are ours to defend and yours to overrule.
We do not run engagements where the client does not own the artifacts. If a procurement process requires us to retain the prompts or the code, we decline.
We do not deploy a surface agent onto no foundation because the timeline is tight. That is the failure mode this whole method exists to prevent.
We do not take engagements outside multi-unit franchise, multi-location retail, or DTC ecommerce. The method depends on operator knowledge we do not have elsewhere.

Common questions

What do you actually own at the end of an AI consulting engagement?: Every artifact produced during the engagement: the code in your repository under your organization, the brand spec and compliance overlay rules, every system prompt and the scoring rubric the gate runs, the auto-publish thresholds and routing configuration with the reasoning behind each number, the per-agent runbooks, the vendor evaluations and contracts, the recorded training sessions, and the golden set of approved and rejected outputs that makes the gate accurate. The golden set matters most and is the one most often retained by vendors, because it is what makes a replacement expensive.
What is the difference between the three tiers of an AI engagement, and which is wrong for me?: Tier 1 is a diagnostic over roughly three weeks ending in a ranked build sequence and one decision. It is wrong for operators who have already run a serious internal assessment. Tier 2 is a defined-scope build over four to eight weeks with a thirty-day operating tail; it is wrong for operators with no foundation layer and no appetite to build one. Tier 3 is an embedded fractional operator, one to two days per week with a six-month minimum; it is wrong for operations with no marketing team to embed inside, because fractional leadership multiplies a team rather than replacing one.
What questions should we answer internally before a first AI consulting call?: Twelve, and the most diagnostic are these. Export your location list from every system that holds one and count how many disagree on hours — that number is your real starting position. Name the system you consider the source of truth for a location record, then check whether anyone else in the building disagrees. Identify who approves a customer-facing message today and how long approval takes at the median. Name who would own the system after an engagement ends; if the answer is nobody, that is the first thing to solve, before any tooling decision.
What will an AI consultant decline to do?: We do not resell software or take vendor referral fees, so vendor recommendations are ours to defend and yours to overrule. We do not run engagements where the client does not own the artifacts. We do not deploy a surface agent onto no foundation because a timeline is tight, since that is the exact failure mode the method exists to prevent. We do not take engagements outside multi-unit franchise, multi-location retail, or DTC ecommerce, because the method depends on operator knowledge that does not transfer.
How is a fractional CMO engagement structured week to week?: One to two days per week embedded inside the marketing operation, with a six-month minimum. Month one covers the diagnostic and the first foundation agent against the same acceptance tests used in a defined-scope build. Months two and three deploy surface agents in dependency order, with each one measurably faster to stand up than the last. Months four through six add the measurement layer, per-location benchmarking, and the operating rhythm the team keeps afterward. Throughout, vendor selection and negotiation, team training, and a documented runbook that makes the role transferable to a permanent hire.

When you have answers to the twelve

Bring them. A first conversation that starts from twelve answered questions is a scoping conversation; one that starts from zero is an interview. Scope is set on the call, in writing, against the sheet above.

Send your answers, or book the call Read the deployment blueprint

Not sure which shape fits? The three-question diagnostic takes about a minute and no email.