Moving from scattered pilots to a sustained, governed AI preparedness capability requires sequencing, not a single leap. This page lays out a phased four-quarter roadmap that starts by fixing the surveillance data foundation, adds early-warning and forecasting models on proven streams, layers in governance and workforce reskilling, and ends with a scaled, audited capability funded to persist through quiet years. It gives public-health leaders a concrete quarter-by-quarter plan with entry and exit criteria for each phase, so investment builds cumulatively rather than resetting at every outbreak.
Preparedness is built in phases, or it is not built at all
The pattern is familiar and costly: a crisis triggers a surge of AI pilots, the outbreak recedes, funding evaporates, and the next event finds the capability rebuilt from scratch and too late. Breaking that cycle means treating AI preparedness as a phased build with cumulative milestones, where each quarter's work rests on the last and survives the quiet years. The 100 Days Mission is explicit that the response clock only holds if surveillance, sequencing, and forecasting are already in place when a threat emerges, not stood up after. A roadmap is how you get there deliberately.
The sequencing is not arbitrary. Data foundation must come first because every model rests on it, and skipping it means burning later quarters on reconciliation instead of forecasting. Early-warning and forecasting models come next, on the streams that are already clean, to prove value and build trust. Governance and workforce reskilling layer in alongside, not after, because a model no one trusts or is trained to use is shelfware. Scale and sustained funding come last, once the capability has earned its place and the metrics prove the return. The history of pandemic funding shows why the last phase matters most: after the 2009 H1N1 and 2014 Ebola scares, preparedness budgets rose sharply and then eroded within a few years, so the world met COVID-19 with capabilities half-built and half-forgotten. A roadmap that ends without committed multi-year funding simply reruns that cycle. Each phase has entry and exit criteria so no one declares victory prematurely, and the final gate is the one that keeps the capability alive between crises.
A four-quarter phased build with clear exit criteria
Sequence the work so each phase depends on the one before it and ends with a measurable gate before the next begins.
| Phase | Focus | Exit criteria |
|---|---|---|
| Q1 Data foundation | Interoperability, timeliness, and lineage across core surveillance streams | Two streams meeting readiness thresholds with tracked provenance |
| Q2 Early warning | Wastewater and countermeasure forecasting pilots on clean streams | Backtested models beating baseline detection lag in shadow mode |
| Q3 Governance and workforce | Review board, reliability thresholds, reskilling frontline staff | Approval gates live, target roles trained and using tools |
| Q4 Scale and sustain | Additional streams, multi-year funding, audited operations | Capability audited, funded on a multi-year basis |
Execute the roadmap with discipline at each gate
- Start with the data foundation and refuse to advance until at least two surveillance streams meet interoperability, timeliness, and lineage thresholds.
- Prove early-warning value on clean streams first, requiring backtested models to beat your baseline detection lag in shadow mode before any live deployment.
- Layer governance and workforce reskilling in parallel with model work, not after, so trust and oversight grow alongside capability rather than lagging it.
- Set explicit entry and exit criteria per phase and hold the gate, resisting pressure to scale a capability that has not yet cleared its checkpoint.
- Secure multi-year funding as the exit criterion for the final phase, so the capability persists through quiet years rather than resetting at the next outbreak.
How phased builds derail
- Skipping the data foundation to reach modeling faster, then burning later quarters on reconciliation the earlier phase should have solved.
- Bolting governance on at the end instead of building it alongside the models, forcing a painful retrofit of oversight onto live systems.
- Scaling before the capability has cleared its exit gate, spreading thin resources across use cases none of which is yet trusted.
- Declaring victory at launch with no multi-year funding, guaranteeing the capability atrophies before the next event tests it.
Track progress against the gates, not activity
- Phase gate completion: whether each phase met its exit criteria before the next began, tracked as a hard yes or no.
- Streams operational: number of surveillance streams meeting readiness thresholds and feeding live models.
- Detection lag improvement: reduction in median detection-to-alert time across the roadmap, measured end to end.
- Funding horizon: length of committed funding for the capability, targeting multi-year continuity as the final exit criterion.
Frequently asked questions
Why not build models first and fix the data later?
Because every model rests on the data, and unreconciled data means spending later quarters on plumbing instead of forecasting. The roadmap puts data foundation first for a reason: a forecasting pilot built on fragmented streams stalls in reconciliation, while the same pilot on two clean, lineage-tracked streams proves value fast. Fixing data later is fixing it never, once the crisis pressure fades.
Can we compress this into fewer than four quarters during an emergency?
You can accelerate within phases, but do not skip the gates. Each exit criterion exists because advancing without it creates downstream failure: scaling untrusted models, retrofitting governance onto live systems, or forecasting on unready data. Prepare the fast paths in advance, pre-authorized data agreements and pre-set reliability thresholds, so speed comes from readiness rather than from cutting the checkpoints that keep the capability defensible.
What makes the final phase different from just running the pilots?
Sustainability and audit. The earlier phases prove value; the final phase makes the capability persist and hold up to scrutiny. Its exit criterion is multi-year funding and an audited operation, which is what breaks the crisis-surge-and-atrophy cycle. A capability that clears every technical gate but has no committed funding is still structurally too late for the outbreak it was built to catch.
Related reading
Go deeper on this sector and topic.