Scenario controls
Recommend to the board: one screen, ready to forward
The move and what it is worth, written from the scenario above. Copy it into an email or print it as a one-page brief. No need to re-operate the tool.
Projected median time-to-therapy, next 18 months
Queue model: stage utilization ρ drives wait non-linearly (W ∝ ρ/(1−ρ)). The cliff is the point: wait time is fine at 75% utilization and catastrophic at 92%. The dashed line is the one-day brand promise.
Why this matters
Growth is the stated #1 priority, and the brand promise (therapy in ~1 day, not a week) is a queueing promise. At 10x provider growth, the human-in-the-loop review layer is the binding constraint, and headcount planning that looks fine on averages fails on the utilization curve. This model makes the break-point visible 6 to 9 months before it happens, and prices the two levers (hiring vs. automation) in the same units.
Stack rank: reviewers offset by month 12
Every automation is priced in the only currency the build-vs-hire conversation needs: reviewer headcount it replaces at months 4, 8, and 12 of your scenario. Because volume compounds at your growth rate, automation appreciates: the same shipped feature offsets more people every month. Hiring doesn't.
| Automation candidate | +cov | HC off M4 | HC off M8 | HC off M12 | eng-wks |
|---|
Uses the scenario set on the Simulator tab. Clinical-content steps (†) are shown at ramped value: they ship behind a human-review gate with explicit exit criteria, so early offset is discounted by design.
The math, exposed
Two numbers per project, both derived from one measurable input: manual minutes recovered per case. Divide by the 22-minute case to get coverage points added; multiply by monthly case volume and divide by effective reviewer capacity to get headcount offset. No weights, no composite score. Arithmetic anyone in the room can check.
One-pager: the capacity crystal ball
Forus's promise is speed. Speed is a function of queue utilization, and utilization at 10x/yr growth is a moving target that breaks suddenly, not gradually. Most ops planning models averages; queues punish averages.
- An 18-month forecast of PA volume → reviewer utilization → projected time-to-therapy, with the break-month surfaced as a single number.
- Hiring and automation expressed in the same currency (reviewer-equivalents), so build-vs-hire is an arithmetic, not a debate.
- An automation backlog ranked by hours-returned-per-engineering-week, risk-discounted for clinical content.
Same model fed by real workflow telemetry: arrival rates by drug × payer, measured service-time distributions per step, reviewer rosters. Add survival-curve treatment of in-flight cases and per-queue (not pooled) utilization. Two weeks to a live internal tool; the forecast updates nightly.
I led product analytics for the launch of Uber Market (capacity, courier supply, and throughput forecasting under hypergrowth) and ran forecasting and experimentation for Meta AI on smart glasses. I've spent a career making "when does this break" a number instead of a feeling. Jeff Pinto · jeff@jeffpinto.com · jeffpinto.com
Sources & method
What is grounded:
- 22-minute manual case: calibrated to CAQH's measured manual prior-authorization effort of ~24 min (16 min via payer portal). CAQH Index Report 2024. caqh.org/…/CAQH_IndexReport_2024_FINAL.pdf
- ~6 PA-requiring scripts / prescriber / month: order-of-magnitude consistent with the AMA Prior Authorization Physician Survey (~39 to 40 PAs per physician per week across all drugs and services; the specialty-Rx subset is a fraction of that). AMA, 2024-2025 eds. ama-assn.org/…/prior-authorization-survey.pdf
- One-day promise vs. the pre-Forus baseline: biologic time-to-therapy averages ~42 days (insurance approval ~21.5d + specialty-pharmacy fill ~20d); a required PA adds ~4 days at median and ~23 when first denied. Burton et al., J Allergy Clin Immunol 2019/2021 (PMID 33404389); Arthritis Care & Research 2020 (PMC7062557). The chart's "pre-Forus week" and "1-day promise" lines bracket this real range.
- Why the queue is the binding constraint: 93 to 95% of physicians report PA-associated care delays; AMA PA Survey 2024-2025. Same URL as above.
What is a labeled assumption (no credible public anchor):
- Reviewer throughput: modeled at 7.5 productive review-hours/day x an 85% effective-capacity factor (approx. a handful of 22-min cases per reviewer-day). Per-reviewer PA quotas are not published by PBMs or UM vendors; this is an internal planning assumption you would replace with measured service-time telemetry in production.
- Utilization cliff at 85%: standard queueing result (W ∝ ρ/(1−ρ)); the specific 85% threshold is a planning convention, not an industry-published number.
Figures are illustrative for a demo. Cited ranges are payer- and disease-specific (much of the denial/turnaround literature is Medicare-Advantage-specific) and would be re-grounded against Forus's own telemetry before any operational use.