How Many AI Projects Can One Part-Time Leader Actually Run

A project is more than a row in a spreadsheet. It brings decisions, meetings, reviews, risks, and follow-up with it. A fractional AI officer can carry several small items when they share a sponsor, data boundary, operating rhythm, and acceptance path. Two projects can still overload the same leader when each needs its own executive group, process owner, and risk conversation.

The useful unit is an ownership draw. Ownership keeps the portfolio coherent. It covers choosing what runs, refusing work with no owner, keeping the decision log current, and explaining what changed. Governance makes the work acceptable through policy, risk review, access decisions, evaluation gates, vendor questions, and escalation. Delivery means directing and reviewing a build or rollout. Discovery is the bounded work that turns an opportunity into a decision.

That split follows public role charters, not a vendor promise. The White House OMB memorandum on agency AI assigns senior AI leadership work across strategy, risk management, coordination, and oversight. The GAO AI Accountability Framework defines governance through roles, responsibilities, goals, and measurement. It is not a single project phase. A fractional leader inherits the same kinds of decisions, even when the organization is smaller.

The role needs one accountable center. Gyde's fractional Chief AI Officer terms lay out portfolio and governance ownership through monthly investment decisions, weekly initiative progress, monthly risk and vendor review, and quarterly executive reporting. That cadence exposes work a build stand-up can hide. The leader keeps the program legible to the people who authorize it.

Versantly's enterprise program describes a four-week operating cadence that moves from executive alignment to department activation before repeating impact reporting. Rawega's AI operating offer describes a weekly operating session, monthly leadership review, quarterly opportunity and impact review, and one primary implementation stream with smaller optimizations alongside. These are provider terms, not independent capacity studies. They show the recurring work a credible fractional engagement has to reserve, but they do not establish a market average.

The leader owns the portfolio boundary even when a delivery team executes the tasks. A request without a decision owner, measurable outcome, risk path, or time box is not ready for the active draw. Put it through discovery first.

Start with 32 available hours for the monthly planning case. That number is a planning model, not a claim that every fractional AI officer works exactly 32 hours. Provider terms run from a one-day-per-week rhythm to several days per month, and public provider pages do not share one definition of part time. Go Fractional's CAIO rate page presents both a 7-hour weekly average and a later 29-hour typical engagement figure. That contradiction is reason enough not to treat provider marketing as a benchmark.

Allocate eight hours to ownership, eight to governance, twelve to delivery oversight, and four to discovery. There is no spare hour on paper. Ownership and governance keep the role from becoming a project manager with no authority. Delivery gets the largest draw because acceptance reviews, dependency decisions, and release gates need concentrated attention. Discovery stays small because its job is to reach a decision before it turns into a second delivery program.

The BLS American Time Use Survey reports that full-time employees who worked averaged 8.4 hours on an average weekday they worked in 2024. That is descriptive labor data, not a fractional CAIO target. It supports one plain premise: hours are consumed by real workdays, not held in a frictionless pool.

Reserve time first for work that does not look like a project but keeps every project safer. Ownership covers the weekly portfolio decision, stakeholder alignment, a decision log, and an executive summary. Governance covers policy interpretation, risk and privacy coordination, evaluation gates, vendor review, and incident reserve. Delivery means reviewing scope, acceptance criteria, dependencies, rollout readiness, and the questions a build team cannot answer alone. Discovery means interviews, process mapping, opportunity sizing, and a written go or no-go recommendation.

The GAO generative AI accountability framework groups accountability around governance, data, performance, and monitoring. That lines up with this model's governance and delivery review work. The NIST AI RMF Playbook adds suggested actions under Govern, Map, Measure, and Manage. Neither document supplies a fractional staffing formula. We chose this allocation as a transparent planning case based on those responsibility structures and the cadence described by providers.

Treat the numbers as a starting budget. A low-risk internal knowledge workflow may need less governance and more discovery. A customer-facing recommendation workflow may need more evaluation and incident reserve before it earns a delivery slot. Change the mix before you raise the ceiling.

Meeting load stays on the calendar because the leader is fractional. O*NET's Chief Executives profile lists directing, coordinating, budgeting, reporting, and delegating as core tasks. Its detailed work context reports daily decisions and constant external contact for large shares of respondents. Those figures cover chief executives broadly, not AI officers. They still show why accountable leadership work grows with the number of stakeholders.

The BLS Occupational Requirements Survey for management occupations reports that 94.7% of management occupations involve supervising others, 97% provide pause control over the pace of work, and 88.5% allow self-paced work in the 2025 estimates. Pause control is not spare capacity. It lets the leader choose the sequence while carrying responsibility for decisions and people waiting on them. A meeting that ends without an owner creates follow-up work, even when the calendar invite was only thirty minutes.

Treat every recurring decision forum as an allocation. Weekly operating review, monthly leadership review, and quarterly executive review are three preparation and follow-through loops, not three isolated meetings. Add a delivery review for every active stream, a governance review for every material risk, and an escalation path for incidents. If those loops exceed the reserved ownership and governance hours, reduce active work or add delivery support.

A workable fractional engagement has a schedule people can see. Weekly work should answer what changed, what is blocked, what decision is needed, and what will be reviewed next. Monthly work should reconcile portfolio priority, investment, risk, vendor status, and capacity. Quarterly work should reset the opportunity set and show executives what was learned, what moved, and what should stop.

The provider pages spell out the cadence. The CAIO's terms describe weekly heartbeat, monthly review, and a range from 8 advisory hours per month to 40 to 60 hours per month for deeper involvement. Gyde describes weekly initiative progress, monthly risk review, and quarterly executive reporting. Versantly describes a four-week activation cycle, while Rawega names a weekly operating session, monthly leadership review, and quarterly opportunity review. These pages show operating shapes. They do not establish how many projects fit.

A missed loop matters more than a full calendar. If the leader skips a monthly risk review to attend a delivery meeting, the portfolio is already drawing more than the model allows. If discovery outputs sit unread because delivery is urgent, discovery is not active capacity. It is backlog theatre. Close it, reduce it, or resource it.

A fractional AI officer can direct a delivery stream without doing every build task personally. That distinction matters for non-software companies with no employee who writes software, firmware, or embedded code. The officer can own the business outcome, acceptance bar, risk path, and executive decision while a delivery pod handles implementation. The ceiling rises only when that support is real and the handoffs are clear.

Our four offers keep the boundary visible. CAIO Core is executive AI ownership from $5,000 per month, with builds separately scoped. CAIO + Delivery is from $10,000 per month and adds a delivery pod with one active build stream. A Starter build is $1,500 to $2,500 fixed, and a Production sprint is $5,000 fixed. The offers are not interchangeable. A project count cannot turn a fixed build into ongoing ownership.

Give proposed work a named delivery boundary. If it needs custom application work, security or legal specialists, or a second active build stream, make it a separate delivery decision. Do not hide it inside the fractional leadership draw. See how our AI ownership model works and our services for the capability boundary, then use the assessment to decide whether the fit is executive ownership, ownership plus delivery, or a fixed project.

Before adding a project, test five things. Is there one accountable business owner who can make decisions? Is there one measurable outcome with an acceptance test? Can the governance cadence handle its data, policy, and risk questions? Does it share dependencies and stakeholders with the current delivery stream? And can the leader say what slips if it starts?

A yes to the first four does not create free capacity. It tells you the work is shaped well enough to enter the model. The fifth controls the decision. Every new active draw must name the allocation it consumes. If the answer is discovery, state which review or delivery activity gets reduced. If the answer is governance, state the risk being accepted and who accepts it. If no one can name the tradeoff, the work is not ready.

The Information Asset Owner role guidance from GOV.UK shows that one person can carry multiple responsibilities. Coordination with data owners, security, privacy, and senior risk roles still exists. Combining roles can reduce handoffs. It does not erase the work.

Pause a discovery stream when it has not reached a written decision by its time box. Reduce delivery scope when acceptance review keeps expanding or when the business owner cannot attend the required checkpoints. Add delivery capacity when the leader is spending more than the reserved delivery draw on implementation detail, test triage, or release coordination. Add governance capacity when policy, privacy, security, or incident questions are waiting for a meeting that does not exist.

Use observable triggers. Missed weekly decisions, stale risk registers, unreviewed evaluation results, and two consecutive meetings without an owner tell you more than a general feeling that the team is busy. NIST's Generative AI Profile emphasizes risks that need to be identified, measured, and managed across the lifecycle. A skipped review becomes a recorded risk, so the controls make capacity pressure visible.

Do not ask the fractional leader to absorb an overloaded portfolio invisibly. Hold the line, stop a stream, narrow its outcome, fund a delivery pod, or move to a fuller ownership commitment. A clean no is part of the role.

Turn the model into a monthly decision routine. At the reset, write down each active draw before anyone proposes another one. For each delivery stream, record the business owner, acceptance test, next review, and the portion of the 12-hour delivery allocation it expects. For each discovery stream, name the question it must answer, the four-hour draw it receives, and the date its recommendation is due. Ownership and governance are not leftovers. If a decision log is stale or a risk review has no owner, reserve those hours before accepting a new request.

Then run three checks. First, compare planned hours with actual calendar and follow-up time. A meeting that generates unresolved questions has a cost after the meeting. Second, look for shared structure. Two streams can sometimes share a sponsor, data boundary, or review forum, but shared vocabulary does not make their decisions identical. Keep separate draws where the risks or acceptance tests differ. Third, choose the response in advance. A small overrun can mean a narrower scope. A repeated overrun means pause the lowest-value stream or add delivery support. A risk that cannot wait means governance takes the time and something else slips. Write that tradeoff down. The point is not to make 32 hours look neat. The point is to expose the next decision before urgency makes it for you.

The planning case assumes one accountable AI leader, 32 hours per month, one main executive sponsor, a bounded governance perimeter, and a delivery team that can execute scoped work without turning the leader into the implementer. The work is for a non-software company whose employees do not write software, firmware, or embedded code. Industry, headcount, title, and tool stack are not assumptions here.

These hours are illustrative. They are not a time study, staffing guarantee, survey result, or promise that every fractional AI officer should work the same schedule. Public role charters describe responsibilities. Labor data describes broader occupations. Provider terms show engagement patterns. None gives us a validated formula for simultaneous AI projects. We combine those sources into a decision aid and label the result as an inference. The Go Fractional terms are a useful warning: one page gives a 7-hour weekly average and a later 29-hour typical engagement figure. Those figures use different frames. We keep both instead of averaging them into false precision.

A project needs a boundary. A discovery stream may be a two-week process map with one sponsor and a written recommendation. A delivery stream may be a single workflow with a defined acceptance test and a pod that handles implementation. Those definitions do not transfer safely to a multi-department transformation, a portfolio with several regulated use cases, or a production system with a live incident queue. In those cases, governance may consume the same 32 hours before delivery is ready for review. Count a stream only when it has a start condition, a decision owner, an expected output, and a stop condition. Without those four fields, it is a standing request, not an honest project count.

At each monthly reset, inspect the decision log, meeting attendance, open risk items, delivery review queue, discovery outputs, and incident reserve together. A leader with enough hours on paper but no closed decisions has no usable capacity. A stream that reaches its acceptance test early can release hours back to discovery or governance. Capacity is a controlled budget, not a quota to fill. Record why each allocation changed, so next month starts with a decision trail instead of a memory contest. The record should show which commitments were completed, deferred, or changed by a new risk or dependency.

Use the model as a monthly conversation starter, not a promise. Write the next review date into each stream, name the person who can change the allocation, and record the evidence that would justify adding or removing work. That small discipline keeps a part-time operating model honest when demand changes.

The model excludes emergency response outside the agreed operating cadence, regulated legal advice, specialist security testing, custom software implementation, and work that requires a second executive portfolio. That work may still be appropriate. It needs its own owner, scope, or specialist support. The BLS Management Analysts outlook notes that analysts can work long hours under tight deadlines. A fractional schedule is not a promise of unlimited availability.