Fractional CTO for healthcare and hospitality SMEs.
Home · Method · Healthcare · Hospitality · Engagement · About · Book a discovery call
Two weeks. The locked strategy becomes a roadmap engineering leadership can run: milestones, one owner, a budget band, and dependencies.
Strategy ends with a handoff block. Translate opens by linking to that block and repeating only those lines: the guiding policy in one sentence, the bets in order, the refusals, and the open questions. The diagnosis stays on the strategy page. This page does not reopen it.
Melissa Perri’s distinction is the working rule. A roadmap states an outcome and a hypothesis. A feature Gantt locks a team into a solution it has not validated. The strategy document is a different artifact.
Week 1. One card per bet. Half a day with engineering leadership and the people likely to own the work. Each coherent action becomes one initiative.
Fields on the card:
Place each card in Now, Next, or Later by how well the problem is understood. Now is work the current team can run inside the appetites. If the bet needs a hire, the hire is the Now card. Next holds problems still being narrowed. Later holds the remaining strategic problems, unfunded.
Week 2. Two reviews, then publish. Engineering walks every Now card and marks the dependencies they will actually honor. The board sees one page: the three columns, the owner, the budget band, and a line that Later is not a commitment. Publish the roadmap with a first review date. Six weeks matches the larger appetite and is a check-in, not a quarter plan.
Janna Bastow’s format puts validated work in Now, sharper problems in Next, and unshaped bets in Later. The columns are confidence, not Q1, Q2, and Q3. Dates go on a card only when an external constraint exists: a regulation date, a season opening, a contract.
Appetite, from Ryan Singer: estimates start with a design and end with a number. Appetites start with a number and end with a design. For a small team, the small batch is about one to two weeks. The big batch is about six weeks. Write the appetite on the Now card before solution detail. Next and Later get no appetite yet, because the problem is not narrow enough. Shape Up’s six-week cycle, cool-down, and betting table are an operating system. They do not fit inside this two-week translation. Only the appetite does.
When two Now cards need the same people, use cost of delay in words. Donald Reinertsen’s rule is to divide the cost of delay by duration. Joshua Arnold’s working definition: value and urgency, usually as money per week, divided by how long the job blocks the team. Ask finance for the weekly leak already in the strategy memo. Ask engineering for duration. Divide.
Skip the variant that adds relative scores and divides by job size. Use Intercom’s RICE score only when finance cannot yet price the leak, and only to sort Next and Later. Reach times impact times confidence, divided by effort. Do not publish both a RICE rank and a cost-of-delay rank for the same card.
GitLab’s handbook records the directly responsible individual: one person accountable for the outcome, who consults others and then decides. For a team of 5 to 30 engineers, put one named person on each Now card. Add a consulted line only for the HIPAA or PCI boundary. A full RACI matrix for every task will not be maintained. The fractional CTO is not the owner unless the contract keeps them in the role after this step.
On the Now service, baseline change lead time, deployment frequency, and change fail rate with the people who ship it. DORA’s guide says to start with a conversation or the Quick Check, to keep the numbers on one service, and to avoid a target such as “daily deploys by year end.” The board tracks the business metric. The delivery numbers stay diagnostic.
Intake, Now. Outcome: share of referrals for one service line that reach an attended visit. Hypothesis: the loss is the unowned handoff, visible within two weeks by counting last quarter’s referrals. Appetite: two weeks. Owner: the practice manager. Dependency: a report from the EHR they already have. Kill criterion: if the drop-off is not in the handoff, stop and return to the diagnosis. Next is the double-entry removal. Later is any model on eligibility. Budget band: current team.
Arrival, Now. Outcome: share of direct bookings at one property that reach a ready room without a desk override. Hypothesis: housekeeping never receives the arrival time. Appetite: six weeks. Owner: the front-office manager. Dependency: read-only access to the PMS arrival feed, and no new card data store. Kill criterion: if the feed cannot be read without a PMS replacement, stop and take that finding back to the strategy. The PMS replacement stays declined.
Engineering reviews the Now cards in full: appetite, dependency, owner, kill criterion. They review Next only at outcome and hypothesis. Later is a problem list. Their question is whether the current team can finish Now without dropping the live service.
The board reviews one page with the same three columns, the same owners, and the same budget bands. Their question is whether the money and the sequence still match the strategy they accepted. They do not review task breakdowns. A change to a bet’s outcome, or a reversal of an option the strategy declined, goes back to the strategy page. Reordering inside Now, using cost of delay, stays in this review.
DevOps receives the published roadmap, the Now cards, the refusals, and the first review date. The 90-day block holds that roadmap still. New ideas go on a list for the day-90 note. They do not enter the month.
The call is to decide whether this engagement fits. The answer can be no.