A fractional chief of staff has two days a week to fix a decision system that operates every day. If every awkward request waits for your next visit, you have added another approval queue. The first-month objective should be a decision rule that works during your absence, with evidence showing where intervention is actually needed.
Treat “misalignment” as a diagnosis to disprove. The same missed commitment can result from conflicting objectives, missing information, overloaded approvers or a founder who retains authority while describing it as delegated. Each requires a different intervention.
Days 1–5: test delegation against an exception
Ask the principal to walk through a customer request that would displace committed engineering work. Who can reject it? Who may accept the displaced commitment? What happens when the principal is unavailable?
McKinsey's chief-of-staff research emphasizes explicit scope and decision rights. Make that operational through a delegation agreement with three decision classes:
| Class | Decision rule | Escalation condition |
|---|---|---|
| Standard | Product lead accepts work within an agreed capacity envelope | Request exceeds the envelope |
| Exception | Founder chooses which existing commitment moves | Affected owner disputes feasibility |
| Emergency | Named deputy applies a temporary limit | Founder reviews at the next defined checkpoint |
The deputy must receive real authority. “Check with the founder” leaves the dependency intact. Test the agreement with a lucrative urgent request before announcing it; exceptions reveal the actual operating model faster than routine cases.
Days 5–10: reconstruct the queue, including unfinished decisions
Sample recent requests from intake through commitment. Include rejected, withdrawn and still-open requests; examining only completed work hides the oldest waits. Use original timestamps rather than people's recollections of how long approval “usually” takes.
Separate four intervals: preparing evidence, waiting for input, waiting for a choice and waiting for implementation capacity. Mark returns for rework. A decision that takes ten days with two hours of analysis has a different problem from one requiring ten days of technical investigation.
John Little's account of Little's Law relates average work in a stable system, throughput and elapsed time. In a fictional approval stream, 12 open requests on average and four resolutions per week imply roughly three weeks in the system. Count the same request class and boundaries throughout. A growing backlog violates the steady-state interpretation; one month's snapshot is insufficient to establish the average.
Use this as a consistency check. If the founder believes decisions take two days while the inventory implies weeks, reconcile missing requests, inconsistent definitions or a changing arrival rate before recommending another meeting.
Track the age of open requests separately from completed-request duration. Prioritizing only new urgent cases can improve the reported completion average while old requests accumulate. Give an aging request a named escalation owner and a decision deadline, including an explicit rejection when appropriate; removing it from the board without a disposition only hides the inventory.
Distinguish overload from ambiguous authority
For each repeated wait, ask a counterfactual: if the authorized person answered immediately, could the work proceed? If not, approval latency is only part of the problem. Missing estimates, implementation capacity or contradictory objectives remain.
Bain's RAPID analysis distinguishes recommending, providing input, agreeing, deciding and performing. Apply that distinction to the evidence trace:
- A complete recommendation waits untouched: investigate decision capacity and delegation.
- Requests repeatedly return for missing delivery estimates: repair the input standard.
- Two leaders can each veto without resolving the tradeoff: clarify their authority boundaries.
- Approved requests wait for the same engineer: reduce accepted work or reallocate capacity.
A fractional schedule adds batching. A Tuesday-only review makes Wednesday requests wait even when the judgment takes minutes. Create an asynchronous route for standard cases and reserve your live time for exceptions. Do not promise a two-day response while requiring attendance at a weekly forum.
Days 10–20: change one mechanism in a live session
Ask sales, product and delivery to complete independent pre-work on the same request class. Compare their accounts against the trace. Distinguish a dispute about the evidence from a dispute about which customer promise deserves priority.
Use a Teams Facilitator SWOT session to examine strategic constraints, or Business Model Canvas if the pattern exposes unpriced delivery work. Bring your decision trace as supporting context; the queue analysis and delegation agreement are documents you maintain separately.
For the fictional team, propose: “A request enters approval only with an effort range, contribution estimate, capacity owner and named displaced commitment.” Decide who checks completeness. Otherwise the new intake form becomes another unowned queue.
Run two actual requests through the rule during discussion. Have the delivery owner explain what they would do on a day you are absent. This tests whether the agreement transfers into work, beyond whether people understand the wording.
Make overrides observable without turning them into bureaucracy
A founder override can contain useful information: an unusually valuable customer, a changed objective or an urgent cash constraint. Require the override to name its reason, affected commitment and person informed. Separate those exceptions from ordinary requests when reviewing the intervention.
Repeated overrides suggest either the rule excludes legitimate business needs or delegation never became real. Ask which explanation fits the cases. Adding stricter reminders to a rule the principal does not accept will consume your limited time without changing decisions.
Keep the exception record short enough to use during the actual decision. A lengthy retrospective form encourages people to reconstruct a tidy justification afterward.
Days 20–30: assess flow and hand over the control loop
Compare arrivals, resolutions, open-request age, rework and overrides for the same request class. If a pre-existing backlog was cleared, separate that clearance from the response time of new requests. A lower average can also conceal one very old unresolved case.
Read the measures together. Faster approval with more delivery failures is a transfer of delay downstream. Fewer escalations with unchanged founder interventions may mean requests moved into private messages. Review a small sample of underlying cases alongside the dashboard.
At month end, leave a named rule owner, a deputy, an exception route and a review trigger such as repeated capacity breaches. Your next engagement day should examine exceptions and new evidence, rather than authorize every routine request.
Use the fractional CXO workshop guide to test the difficult tradeoffs, and run a guided session once the decision mechanism requiring repair is clear.