The procedure is the product

Research sites misallocate their largest cost. Coordinator labor gets scheduled against individual studies, whose demand is lumpy and unpredictable, when demand at the portfolio level is smooth enough to staff against. The fix is operational rather than technological: decompose every schedule of assessments into procedure units, route each unit to the lowest qualified tier, and coordinate the schedule centrally. For a mid-size independent site, the recoverable value runs $110,000 to $400,000 a year — and the change requires no regulatory approval and no sponsor contract amendment.

How sites organize work today

Walk into any research site and ask how work gets organized. Coordinators are assigned to studies. Maria has the cardiology trial and two vaccine studies; James has the oncology protocol because he’s been here longest. When Maria’s enrollment surges, she drowns. When James’s study pauses for an amendment, the site eats his salary.

Nobody designed this arrangement; it accumulated because studies arrive one at a time and assigning a person to each is the obvious move. The consequence shows up in every site P&L: a site that staffs to each study’s peaks carries unrecoverable slack, and a site that staffs lean breaks the moment two studies surge at once.

The work arrives one procedure at a time — sites staff for it one study at a time.

Decomposing the schedule of assessments

Every SoA in every protocol is a bill of materials built from repeating units — blood draws, ECGs, vitals panels, consent discussions, dose administrations, rater-certified assessments. Decomposition means breaking each protocol’s SoA into these discrete units and tagging each with four attributes: required skill tier, duration including documentation, equipment and room type, and any certification or protocol-specific training attached.

The output is a long and fairly tedious table, but it changes what a site can see about itself in two ways.

Aggregate demand becomes visible. Summing procedure demand across the portfolio by week and skill tier replaces a stack of per-study guesses with the site’s real workload, and that workload is far smoother than any single study’s curve — smooth enough to staff against.

The skill-tier mismatch becomes visible. A coordinator who owns a study end to end performs everything the study generates, which means credentialed senior staff spend part of every week on work a trained medical assistant could carry, and another part on scheduling logistics that belong to nobody’s clinical training. The mismatch is a work-design problem: route each unit to the lowest qualified tier and reserve senior people for the judgment work — consent, adverse event intake, clinical source documentation, the patient relationship — that actually requires them.

The worked example

Take a recognizable composite: a multi-specialty independent site running 24 active protocols across dermatology, cardiometabolic, and vaccines. Three PIs, eleven coordinators, two medical assistants, a part-time phlebotomist, a regulatory manager, a recruitment lead.

One caveat is worth stating up front: the figures below are planning estimates, with the reasoning shown. The decomposition exists so a site stops needing anyone else’s benchmarks, including these. The table produces your actual numbers in weeks, from documents you already own: a protocol decomposes in 4–8 hours of skilled work, so the full 24-protocol build runs 120–180 hours, or a two-week sprint with extraction tooling doing the first pass.

Expect 55–70% of procedure instances to concentrate in roughly a dozen unit types — draws and sample processing, vitals and ECGs, questionnaires, scheduling and reminders, drug accountability. That concentration is measured in instances rather than hours; these are the short units, so they account for a smaller share of labor hours — and the economics run on hours. The below-tier share of coordinator time sums from two sources: hands-on hours inside those high-frequency units, plus the scheduling and logistics work attached to every visit. A third of senior coordinator time below tier is a defensible planning estimate, and the table replaces it with your own figure.

At one-third below tier, eleven coordinators carry ~3.5 FTE of work that belongs at a tier costing half as much. Routing it is itself a job — figure half an FTE of central coordination at $40–50K loaded. Two paths from there:

Against an implementation cost of one person’s quarter plus incremental training, either path pays back inside the first year — the cost path within months.

Two implementation costs need to be counted just as honestly.

Scheduling infrastructure. Most site CTMS platforms treat scheduling as a per-protocol visit calendar; where resource features exist, they rarely extend to skill-tier routing across a portfolio. The workable pattern today runs the decomposition table and a capacity schedule alongside the CTMS — clumsy, but workable. A purpose-built scheduling layer for sites is an obvious product gap, but its absence is no reason to wait.

Training, delegation, and turnover. Under the legacy model, three people sit on a study’s delegation log; under a pooled model, eight might, each with documented protocol training before touching a procedure. Training burden scales with pool size × protocol count, so the pool cannot be “everyone does everything.” Tiered pooling keeps it manageable: float staff only across high-frequency, low-specificity units, and cap pooled training per protocol to what coverage requires. Turnover raises the stakes — when a cross-trained tech leaves, her training investment across every delegation log leaves with her. Plan for the higher per-departure cost explicitly, keep at least two trained people per pooled unit type per protocol, and treat the retention upside (a visible skills ladder below, less burnout work above) as something to measure rather than assume.

Reallocation or new model?

Which one it is determines who has to say yes, and the answer differs by layer:

The interior layer is resource allocation, nothing more. The regulatory object is the investigator and the conduct of the study — tasks performed by qualified people, delegated in writing, supervised by the PI, documented contemporaneously. FDA’s 2009 guidance on investigator responsibilities frames supervision and delegation at the level of tasks and individuals, which is precisely the resolution a decomposed operation runs at. The only regulatory exposure is the one the site already carries: whether documentation holds up under inspection. Run properly, a decomposed operation holds up better, because delegation maps to named competencies instead of a job title and an assumption.

Sponsor friction is real, but it comes from convention rather than from any contract. What sponsors buy is protocol-compliant conduct; no CTA requires dedicated staff. The friction arrives through monitors trained on the legacy pattern, who will notice an eight-person delegation log and ask why. Get ahead of the conversation: a one-page staffing-model description in qualification materials, the training matrix offered at the site initiation visit, the delegation log organized by competency so a CRA can audit it in minutes. Present the model as a quality system and monitors respond with curiosity; let one discover it mid-study and the response is findings.

The boundary layer is where permission genuinely enters. Floating staff across locations raises per-site delegation, licensure, and the 1572 one-site-or-two determination — a network’s regulatory architecture has to be designed deliberately rather than extrapolated from a spreadsheet. Capacity pricing asks procurement organizations to change how they buy, a multi-year motion regardless of the arithmetic. Remote PI oversight has room under ICH E6(R3)’s risk-proportionate posture, with the burden of demonstrating real supervision sitting squarely on the operator. The layers are severable: capture the interior value this quarter, and let its operating data become the evidence the boundary conversations later require.

What it changes downstream

Budgeting. A site that knows its loaded cost per procedure unit builds study budgets bottom-up in hours instead of negotiating from the sponsor’s template and a guess — and knows which protocols are quietly unprofitable before signing. No sponsor format change required; the site simply stops being the party that doesn’t know its own costs.

Feasibility. Procedure-level data answers feasibility questionnaires from capacity arithmetic rather than optimism: the rater-certified hours and infusion-chair availability either carry the proposed enrollment curve or they don’t. Sponsors have spent two decades treating site-reported feasibility as fiction; the remedy is sites that have the data to answer honestly.

Growth. A site organized around studies grows by adding studies and hoping coordinators absorb them. A site organized around procedure capacity grows by adding capacity where the data shows constraint — new rooms, embedded clinic space, additional locations — without rebuilding its operating model, because coordination already runs centrally.

The site simply stops being the party that doesn’t know its own costs.

Where to start

For the site above, in order:

1. Decompose the top 10 protocols by visit volume. They cover most of the procedure demand.
2. Pool three unit types only — phlebotomy with sample processing, vitals with ECG, scheduling with reminders — across no more than 4–5 cross-trained staff. The dozen-type steady state comes later.
3. Leave the judgment work untouched. Consent, AE intake, rater-certified work, and the patient relationship stay with named coordinators. Patients keep the person who knows them and answers the phone — that person gains the time to do it well. What moves are the tasks where the patient experiences the procedure rather than the relationship.
4. Rebuild delegation logs and the training matrix before the first pooled visit, not after.
5. Measure coordinator utilization one quarter before and one quarter after.
6. Decide with numbers: extend the pool, take on the studies the freed capacity supports, or both.

The checklist is the easy half. People hired to own studies are being asked to own patients and competencies instead, and that transition needs to be led rather than announced — it is the hardest and least technical part of the work.

What matters more than the sequence is the recognition underneath it. Sites have spent decades pricing and staffing themselves by the study, while the actual work — the thing sponsors pay for and patients experience — happens one procedure at a time. The sites that learn to see their own production at that resolution will run cheaper, prove quality instead of asserting it, and answer the industry’s oldest question — can this site really deliver? — with arithmetic. The rest will keep assigning Maria another study and hoping.