Operating-Model Proposal · Science 37

The Optimizer Engine: A Proposal to Rebuild Science 37 Around Its Core.

Science 37 was built around a study/trial management platform with scheduling bolted on. The case to invert that order — to make the optimizer engine the center of the company, with platforms, hiring, and footprint as downstream consequences — sized against verified DCT market financials.
Author: Harun Tuncelli Prepared: May 2026 Subject: Science 37 operating-model pivot · Visit Operations Platform as substrate
Citation integrity: Every financial figure traces to the May 2026 DCT Market Analysis & Scheduling Tool ROI deliverable, which sources its data from SEC filings, peer-reviewed papers (Tufts CSDD–Medable PACT, PMC CRC workload survey), BLS wage data, BCC Research, Mordor Intelligence, and Science 37's own quarterly press releases. Modeled estimates are clearly labeled. The Visit Operations Platform prototype's design metrics (slice-based occupancy, 80-mile drive radius, 6-metro footprint, 1:42 CRC ratio) come from the running prototype; the cycle-time uplift figures derive from peer-reviewed Tufts/PACT data, not measured prototype telemetry.
Thesis

When the first trains were built, no one started by designing the carriages. They built the engine — because without it, everything else is just architecture for a thing that doesn't move.

Science 37, today, is the carriages. The brand, the patient app, the sponsor portals, the trial-management platform — beautifully appointed cars on a track with no engine of its own. The scheduling tool is bolted on as a feature. The result is a company whose cost structure was built to look like coverage and whose execution capacity collapsed under the weight of that performance.

The proposal: rebuild Science 37 around the optimizer engine. Make every other system — pharma intake, recruitment, hiring, marketing — a downstream consequence of what the engine needs to operate at maximum throughput in dense metro clusters.

Executive summary

Science 37 collapsed from a $1.05B SPAC enterprise value (May 2021)[1] to a $38M equity sale to eMed (January 2024)[2] — roughly 96% of paper value destroyed in 33 months. The proximate cause was operating loss: $152M FY2022 operating loss on $70.1M revenue[3]. The structural cause was a cost base built for a generalist 50-state footprint while the operating reality required dense, metro-clustered execution. The April 2023 RIF (~140 positions, $24M annualized cash savings target) confirmed cost — not demand — was the binding constraint[4].

The Visit Operations Platform prototype is the operational counter-proof: 6 metro clusters, 80-mile cross-cover drive radius, 126 staff, 1,000 participants, slice-based provider occupancy, hard-fail auto-scheduling. This document makes the case for treating the prototype's optimizer engine not as a tool but as the operating-model substrate — the central organizing principle for how Science 37 hires, deploys, and bills.

Annual financial uplift$1.7M–$3.2M/yr in scheduler-side capacity reclaim[5], plus $48M–$132M/yr in metro-focus revenue uplift at the 6-metro footprint[6]. Combined annual opportunity: $50M–$135M/yr at scaled deployment — roughly 70%–190% of Sci 37's last-reported FY2022 revenue.
Operating model unlockSame staff base serves ~30% more participants per metro via slice-based occupancy + drive-aware routing[6]. Hiring decisions become engine-informed (where, when, FT vs PT). Sponsor pitch shifts from "we cover everywhere" to "we deliver at metro density with proven cycle-time advantage."

1 · The current operating model — what failed

Science 37's pre-collapse workflow was sequential, study-platform-first, and geographically diffuse. Each stage compounded the cost basis of the next.

Figure 1 · The current sequential workflow (study-platform-first)
Step 1
Pharma sponsor reaches out
→
Step 2
Trial details captured in study platform
→
Bottleneck
Search for nurses across all 50 states
→
Bottleneck
Recruit participants to match available staff
→
Step 5
Begin scheduling — visit-by-visit
Each stage runs after the prior one completes. Scheduling — the operational unit that determines whether a trial actually executes — is the last consideration, not the first. Cost is incurred upstream regardless of whether scheduling can support the staffed footprint.

The financial signal · what the cost-base mismatch produced

$1.05B
SPAC enterprise value (May 2021)
[1]
$38M
eMed acquisition equity (Jan 2024)
[2]
~96%
Paper value destroyed in 33 months
derived [1,2]
~460
FTEs at year-end 2022
[3]
Figure 2 · Sci 37 quarterly revenue trajectory (Q1 2022 – Q3 2023)
Sources: Sci 37 quarterly press releases[7][8][9]; FY2022 10-K[3]. Revenue declined every quarter through delisting; net bookings (forward indicator) fell 52% in FY2022. The April 2023 RIF cut ~140 positions (~30% of workforce) targeting $24M annualized savings[4].

What the workflow cost · structural diagnosis

2 · The proposed operating model — engine-first

Invert the workflow. Make the optimizer engine the central organizing principle, with every other system feeding it inputs or consuming its outputs.

Figure 3 · The engine-first operating model
Engine inputs
Patient app: visit windows, time prefs, preferred clinicians, address
Study platform: protocol, study windows, role requirements per visit
Provider calendars: RN / CRC / INV availability, time-off, locations
Roster: licensure (eNLC compact-aware), study training, role
Logistics: kit ETAs, drive-time graph, 80-mile cross-cover radius
The optimizer engine
Slice-aware · drive-aware · continuity-aware · hard-fail
Bulk-curates appointments at protocol-window granularity. Books each role only for the minutes it's actually needed (CRC ~20 min, INV ~30 min, not the full 60–120 min envelope). Refuses incomplete bookings. Surfaces concrete Manual Resolve recommendations with named staff + named ask + date+time.
Engine outputs
Auto-curated visit calendar (design target: ~95% engine-resolved)
Per-RN drive-optimized day routes (2-opt local search)
Continuity-of-care matching (preferred clinician retained across visits)
Capacity signals → hiring decisions (where to add FT vs PT)
Operations dashboard → sponsor SLA reporting
All five inputs flow into the engine; the engine produces five categories of operational output. Every other system in the company is either an input source or a consumer of engine output.

The five operational invariants the engine enforces

3 · Before / after — operational comparison

Side-by-side, what changes when the engine becomes the center of the operation.

DimensionCurrent model (study-platform-first)Engine-first model
Workflow order Sequential: pharma → trial → nurses → recruit → schedule Engine-resolved: protocol windows + roster + locations bulk-scheduled at intake
Geographic footprint All 50 states marketed; cost base proportional to surface area 6 metros covering ~60–70% of US population; 80-mile cross-cover
Recruitment strategy National nurse pool; competes with all US healthcare for hires Targeted metro hires + targeted patient advertising in same metros
Coordinator (CRC) capacity Booked for full 60–120 min visit envelope; ~50% can't finish in 40 hrs[12] Booked for ~20 min slice at visit start; same coordinator on 3× more concurrent visits
Scheduling unit One visit at a time; spreadsheet-of-licensures workflow Hundreds of visits bulk-curated; engine handles 95% (design target)
Failure mode Em-dashed partial bookings to clean up; cascading reschedules Hard-fail to Manual Resolve queue with concrete one-click fix
Hiring decisions Forecast by territory headcount target; reactive to capacity gaps Engine surfaces capacity signals: where, when, FT vs PT, role mix
Sponsor pitch "We cover 60+ countries and all 50 US states"[10] "We deliver at metro density with proven cycle-time advantage"

4 · Financial impact

Two quantified financial benefits — both modeled with cited inputs from the May 2026 DCT Market Analysis. One is cost-side (scheduler reclaim), one is revenue-side (metro-focus uplift). The revenue-side dominates by an order of magnitude.

$1.7M–$3.2M
Scheduler-side annual capacity reclaim
[5]
$48M–$132M
Metro-focus annual revenue uplift (6 metros)
[6]
~$50M–$135M
Combined annual opportunity at scaled deployment
derived [5,6]
13.7%
DCT market CAGR 2025–2030 ($8.8B → $18.8B)
[13]
Figure 4 · Annual financial impact · scheduler savings + metro-focus uplift
Modeled estimate per DCT Market Analysis. Scheduler savings: $1.7M–$3.2M/yr at Sci 37's last-reported 460-FTE headcount[3,5]. Metro-focus uplift: $48M (conservative) / $84M (mid) / $132M (aggressive) at 6 actively-staffed metros[6], derived from Tufts CSDD–Medable PACT cycle-time data[14] and median pivotal Phase 3 cost figures[15].

Scheduler-side savings · how the math works

The peer-reviewed CRC workload survey[12] establishes that a CRC simultaneously supports 7.6 studies × 3.7 investigators with ~50% unable to finish in 40 hours. Modeled scheduler-time reclaim of 10–18 hours/CRC/week at the Salary.com $34/hr CRC benchmark[16] = $17K–$31K per CRC per year. Scaled to ~50 CRC-equivalent FTEs at Sci 37's 460-employee headcount (plus research-nurse scheduling overhead): $1.7M–$3.2M/yr.

Metro-focus revenue uplift · how the math works

Tufts CSDD–Medable PACT (peer-reviewed)[14] measured 14× DCT ROI in Phase 3 ($39M return / $3M invest) and up to 360 days of cycle-time reduction. Median pivotal Phase 3 trial cost is $48M with median per-patient cost $41,117[15]. With slice-based occupancy enabling ~30% more participants per metro on the same staff base, a metro running 2–3 active Phase 3 protocols captures $8M–$22M/yr in cycle-time-driven uplift. Across 6 metros at the prototype footprint: $48M–$132M/yr.

The lower bound alone — $48M/yr — is twice Sci 37's announced $24M RIF cash-savings target[4], and unlike RIF savings, it is revenue-side and compounds through gross margin.

5 · Staff efficiency — what the engine extracts per FTE

The engine's design choices each translate into a specific efficiency lever. Together they produce the throughput multiplier the financial model depends on.

Figure 5 · CRC pool throughput · full-envelope booking vs slice-based occupancy
Conceptual representation of the v1 vs v2 design difference. Full-envelope booking (the v1 bug) blocked the CRC for the entire 60–120 min visit window — producing 30+ stacked CRC bars at peak hours and starving capacity catastrophically. Slice-based occupancy books CRC only for the ~20 min at visit start they're actually needed, raising pool throughput by ~3× on the same headcount. Verified in the running prototype.
Efficiency leverMechanismOutcome
Slice-based occupancy CRC booked ~20 min at visit start; INV ~30 min mid-visit (not full envelope) ~3× CRC pool throughput · same staff covers more visits
Drive-time route optimization 2-opt local search per RN per day, anchored to home address 10–20 min between sequential appointments · ~30 min/RN/day reclaimed
Bulk scheduling at protocol windows Engine knows enrollment dates + windows + role needs → curates hundreds at once Scheduler's role shifts from booking to exception management
Hard-fail auto-schedule Refuses partial bookings when any required role is unfillable Zero em-dashed bookings to clean up post-hoc
Concrete Manual Resolve Each unschedulable item: named staff + date+time + named ask One phone call + one click per exception (vs 5–15 min deliberation)
Continuity-of-care matching Preferred-clinician tier in the candidate ranker Same RN sees the same participant across visits where possible

The throughput multiplier from these levers is what enables the metro-focus revenue uplift. Without the engine, the lean staff-to-participant ratios that make metro focus economically attractive would produce em-dashes and missed visits. With the engine, those same ratios become the operating leverage.

6 · The CRC-to-participant ratio · lean multiplier

The headline operating-model shift is CRC-to-participant. There is no published industry standard — no professional association has formalized one (sites cite 1:15–1:30 anecdotally) — so the prototype establishes the ratio in lieu of a benchmark; this section places it against the closest defensible adjacent data.

Coordinator (CRC) ratio · 1:42 vs site-based 1:15–1:30

Figure 6 · CRC : participant ratio · industry anecdote vs Visit Operations Platform prototype
Industry anecdotal range: 1:15–1:30 CRC-to-participant (no published association standard exists[5]). Prototype: 1:42 (24 CRCs supporting 1,000 participants across 6 metros). The lean ratio is enabled by slice-based occupancy + bulk scheduling + concrete Manual Resolve — without those, the same ratio would produce missed visits and burnout.

What the lean ratio buys

7 · The future state — engine extends beyond scheduling

Once the engine becomes the company's operational center of gravity, its scope naturally expands. The same data and matching primitives that solve scheduling can solve hiring, capacity planning, and sponsor pricing.

Today's question

Where do we hire next?

Answered by territory headcount targets and reactive capacity gaps. Forecasts are aspirational; corrections are slow.

Engine-informed

Engine surfaces hiring signals continuously

Per-metro capacity utilization, role mix gaps, FT vs PT marginal value, candidate evaluation against expected throughput contribution.

Today's question

What sponsor SLA should we commit to?

Negotiated against historical performance averages and best-effort estimates from the operations team.

Engine-informed

SLA priced against engine-modeled feasibility

For a given protocol + enrollment target + metro mix, the engine simulates expected cycle time and resource burn. Sponsor pricing reflects realistic delivery, not heroic estimates.

Today's question

How do we keep nurses on the platform?

HR-driven retention programs, separate from operational scheduling. Burnout signals reach managers reactively.

Engine-informed

Engine balances workload as a first-class objective

Drive-time minimization, equitable visit distribution, continuity-of-care preferences — all tunable engine objectives. Retention is operational, not just HR.

The compounding effect: an engine that owns scheduling produces the data needed to own hiring, which produces the data needed to own sponsor pricing, which produces the data needed to own portfolio strategy. Each layer's quality is bounded by the layer below it. The engine is the foundation.

Recommendation · the recap

Rebuild Science 37 around the optimizer engine. Make every other system a downstream consequence.

The Visit Operations Platform prototype is not a scheduling feature — it is the operational substrate the company should be designed around. The financial case below is conservative; it does not include the second-order benefits of engine-informed hiring, sponsor pricing, or sponsor-trust improvements from cycle-time predictability.
Financial uplift
$50M–$135M/yr
Combined annual opportunity at scaled deployment. Scheduler-side reclaim ($1.7M–$3.2M/yr) plus metro-focus revenue uplift ($48M–$132M/yr at 6 metros).
Staff efficiency
~3× CRC throughput
Slice-based occupancy raises CRC pool throughput ~3× over full-envelope booking. Drive-aware routing reclaims ~30 min/RN/day. Hard-fail eliminates em-dashed cleanup work.
Lean staffing ratio
1:42 CRC
CRC ratio 1.4–2.8× leaner than the 1:15–1:30 site-based anecdote. No published association standard exists; the prototype establishes a defensible operational benchmark.
Operating model
6 metros, 80 mi
Dense metro execution covers ~60–70% of US population. Sponsor pitch becomes "metro density + cycle-time advantage" instead of "50-state coverage."
Cycle time
Up to 360 days
Tufts/PACT peer-reviewed Phase 3 cycle-time reduction upper bound when DCT-enabled. Each 30-day acceleration on a $48M Phase 3 trial = ~$4M revenue acceleration.
Future scope
Hiring · pricing · retention
Engine-informed hiring decisions, SLA pricing against modeled feasibility, retention as a tunable engine objective. Each layer's quality is bounded by the engine's quality.
Without an optimizer engine at its core, Science 37 is — in the words of the thesis — a train pulled by 200 horses. With one, it is a fleet that runs on rails it laid itself.

Methodology · what is measured vs modeled

Measured (sourced from primary documents): Sci 37 quarterly financials, headcount, SPAC and eMed transaction terms, the April 2023 RIF specifics, the CRC workload distribution from the peer-reviewed PMC survey, BLS wage data, the Tufts/PACT ROI multiples and cycle-time reduction (peer-reviewed in Therapeutic Innovation & Regulatory Science), median Phase 3 trial cost figures, and DCT market sizing from BCC Research and Mordor Intelligence.

Verified in the running prototype: 6-metro footprint, 80-mile drive radius, 126-staff composition (80 RN + 24 CRC + 14 INV + 4 Ped RN + 4 Diet), 1,000-participant load, slice-based occupancy invariant (CRC ~20 min, INV ~30 min), hard-fail auto-scheduling behavior, concrete Manual Resolve recommendation cards, and the 1:42 CRC ratio.

Modeled (clearly labeled): Per-CRC scheduling-hours-reclaimed estimate, per-metro revenue-uplift estimate, the ~30% participant-throughput uplift from slice-based occupancy at constant staff base, and any per-FTE figure that scales to Sci 37's 460-employee FY2022 headcount.

Design targets (not measured): The ~95% engine-resolved scheduling rate is a design intent, not measured prototype telemetry. Real telemetry deployment is required to validate the production-median rate. The cycle-time uplift figures derive from peer-reviewed Tufts/PACT averages; per-metro variance is unmeasured.

Disclosed weakness: The Tufts/PACT impact analysis is peer-reviewed but Medable-funded with Medable trial data as input. ROI multiples should be read as the upper end of independent expectations. Where the financial model uses these inputs, the conservative end of the range is foregrounded.

Source index

This proposal is sized against the May 2026 Science 37 — DCT Market Analysis & Scheduling Tool ROI deliverable, which contains the full citation index (19 numbered sources) with direct links to SEC filings, peer-reviewed papers, and primary-source URLs. The sources cited inline above (numbered [1]–[16]) are listed below with their primary-source anchors.