AI Consultancy to Fractional Architect: When to Hand Off

2 August 2026 · Bedrock AI

Hand off from an AI consultancy to a fractional AI architect at the moment the question changes from “what should we do?” to “who owns this running in production?”. That point usually arrives once the diagnostic is written and the first workflow is scoped, roughly four to eight weeks into an engagement. Most businesses never make the switch, because nobody’s contract makes it anyone’s job to raise it.

What does sequencing a consultancy into a fractional engagement actually mean?

Sequencing means treating the consultancy engagement as the diagnosis and the fractional engagement as the ownership, deliberately, from the start. It is not a fallback for a project that went badly. It is a two-phase structure where an outside firm answers a bounded question with fresh eyes, and a named architect then holds the standing seat that turns those answers into a working system with someone accountable for it.

The pattern is well established outside AI. Analysts of the fractional model describe organisations that begin with executive consulting to gain clarity, then move into fractional leadership when they need embedded execution rather than more analysis (Shane Kinkennon). AI simply makes the seam more visible, because the gap between a strategy that reads well and a system that behaves correctly under real inputs is unusually wide.

What makes AI different is that the diagnosis has a short shelf life. A supply chain review holds its shape for two years. An AI opportunity map goes stale in months as model capabilities, pricing and the vendor landscape shift underneath it. A recommendation nobody owns is not merely unexecuted, it is decaying.

When is the right moment to hand off?

The handoff moment is when the remaining work is decisions, not findings. A consultancy is optimised to reduce uncertainty about the business: what the workflows are, where the cost and time actually sit, which use cases clear the bar. A fractional architect is optimised to make and defend the hundreds of small binding choices that follow: which model tier for which step, what the fallback does, who gets paged, what “good” means in the eval suite.

Five signals say the diagnostic phase has finished its useful life:

1. The findings memo has stopped changing. Two more weeks of interviews are producing confirmation, not new information.
2. The debate has moved to implementation trade-offs. Meetings are now about retrieval design, latency budgets or vendor terms rather than about which problem to solve.
3. Someone has asked “who is going to run this?” and got no answer. This is the single clearest trigger, and it is usually asked by the person who suspects it will land on them.
4. A pilot is being built with no named production owner. Building before ownership is assigned is how pilots become orphans.
5. The consultancy’s proposal for phase two is another phase of analysis. Firms are structured to sell the next scoped deliverable, which is a reasonable commercial instinct and a poor fit for the work in front of you.

If three or more of these are true, the engagement has already passed its handoff point and is running on momentum.

Why do most businesses skip the handoff entirely?

Most businesses skip it because no one in the room is contractually responsible for ending the engagement well. The consultancy is accountable for a deliverable and exits when that deliverable is accepted (Fractionus). The internal sponsor is accountable for the business outcome but usually lacks the technical standing to specify what a successor would need. So the report lands, everyone congratulates each other, and the implicit plan becomes “the team will pick this up”, on top of everything the team was already doing.

The consequences are measurable at the production boundary. A March 2026 survey of 650 enterprise technology leaders found that while 78% of enterprises had at least one AI agent pilot running, only 14% had scaled an agent to organisation-wide operational use (Zen van Riel). Analysis of that scaling gap identifies organisational ownership as the least technical of the barriers and arguably the most destructive: when IT and the business unit each assume the other owns the agent’s production behaviour, the evaluation harness never gets built, because it sits in neither team’s mandate (AI Assembly Lines).

There is also a quieter reason, and it is worth naming. Handing off looks like an admission that the expensive engagement did not finish the job. It did finish its job. Diagnosis is a different job from ownership, and pretending otherwise is what produces a strategy deck with a six-figure invoice attached and nothing running twelve months later.

What has to transfer at the handoff?

A handoff is a transfer of decision rights and context, not a transfer of documents. The failure mode is a well-produced deck landing in an inbox with no accompanying authority to act on it. Six things have to move for the sequence to work.

The second row is the one businesses most often lose. Consultancies deliver conclusions; the reasoning behind rejected options usually lives in an associate’s working file and leaves with them. Insisting that the decision log is a contractual deliverable, not a courtesy, is the single cheapest thing a buyer can do to protect the handoff.

What does the fractional engagement look like once it starts?

The fractional phase is a standing commitment measured in days per week, not a project measured in deliverables. Published guidance on fractional AI leadership puts typical engagements at one to three days a week against a monthly retainer, running twelve to twenty-four months, with the early months on governance and the first production workflow, and later months on building the internal capability that eventually replaces the fractional seat (Hatchworks, Stefan Finch).

That last part matters and is often misread as a sales problem. The best fractional engagements are explicitly designed to phase out: the architect builds the operating model, writes the decision records, trains the named owners, and leaves a business that can make its own calls. A fractional architect angling to become permanently indispensable has misunderstood the role.

This is where the diagnose-first philosophy earns its keep. The architect arriving after a good consultancy engagement does not start cold, but nor do they accept the map as given. They re-test the top two or three recommendations against what they find in the systems themselves, because a diagnosis assembled from interviews and one assembled from logs rarely agree completely. Architecture beats tool-buying precisely because it survives contact with the actual data.

When should you not hand off at all?

Skip the fractional phase when you already have the leadership bandwidth to execute, and only then. If there is a senior technical owner with genuine AI judgement, spare capacity and formal authority over the roadmap, a consultancy’s diagnosis plus that person is a complete arrangement. Adding a fractional architect on top creates two people with overlapping decision rights and a slower path to production.

The honest test is capacity, not competence. Most businesses do have someone capable. Very few have someone capable who is not already running three other things. Hiring embedded leadership when you need analysis, or analysis when you need embedded leadership, is how organisations spend six figures and get the wrong shape of help (Forbes).

FAQ

Can the same firm do both phases? Sometimes, and it is worth asking directly whether the named senior person in the pitch will hold the fractional seat. If the answer is that a different team picks it up, you are getting a second engagement, not a handoff.

How long should the overlap be? Two to four weeks. Long enough for the architect to sit in on the closing sessions and interrogate the reasoning while the consultancy is still contracted, short enough that you are not paying twice for the same weeks.

What if the consultancy has already left? Then the fractional architect’s first fortnight is a re-diagnosis, and you should expect to pay for it. Request the decision log retrospectively: most firms will supply it, and it materially shortens the ramp.

Does this work for a business with no AI running yet? Yes, and it is cheaper at that stage. Sequencing before anything is built means the architect shapes the first system rather than inheriting one that has to be unpicked.

Is a fractional architect just a contractor on a retainer? No. A contractor executes a defined scope you specify. A fractional architect holds decision rights and is judged on whether the system works in production, which is a different accountability entirely.

Bedrock AI maps your systems, team and workflows to show where AI actually pays, before you spend a pound building. Book a strategy call.