borghei/Claude-Skills / research-ops/research-finance
research-finance
VerifiedResearch budgeting and funding operations — study budget construction, cost per participant and per insight, burn against milestones, and portfolio prioritisation by decision value. Use when costing, tracking, or ranking research.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 15.1 KB |
| assets/sample_budget.json | config | 853 B |
| assets/sample_burn.json | config | 1.6 KB |
| assets/sample_portfolio.json | config | 2.4 KB |
| assets/study-budget-template.md | markdown | 4.5 KB |
| references/funding-and-portfolio-allocation.md | markdown | 15.4 KB |
| references/research-cost-models.md | markdown | 15.6 KB |
| scripts/burn_vs_milestone_tracker.py | script | 11.9 KB |
| scripts/portfolio_prioritizer.py | script | 12.3 KB |
| scripts/study_budget_builder.py | script | 12.4 KB |
references/research-cost-models.md
View on GitHub# Research Cost Models
Reference for building research budgets that fund what the work will actually
consume: cost structures by method, what drives each unit cost, and the line
items that are reliably forgotten.
## 1. The four cost blocks
Every research budget decomposes into four blocks. Most under-budgeting comes
from putting a cost in the wrong block, not from omitting it entirely.
| Block | Scales with | Examples |
|-------|-------------|----------|
| **Per participant** | Enrolment or screening volume | Incentives, assessments, labs, imaging, transcription |
| **Per site** | Number of sites | Startup, training, initiation visit, close-out |
| **Per site-month** | Sites × duration | Coordination, monitoring visits, site fees, storage |
| **Fixed** | Nothing | Protocol development, statistics, data management, final report |
The classic misclassification is **site coordination billed as fixed**. It is a
per-site-month cost, and on a 30-month study across six sites it is 180
site-months — often one of the largest lines in the budget. Booked as fixed, it
is understated by roughly an order of magnitude.
The second most common is **monitoring billed as fixed** when it scales with
both sites and duration.
## 2. Screening and yield
Participant-stage costs divide into those incurred on **everyone screened** and
those incurred only on **those enrolled**.
```
participants_screened = participants_enrolled / (1 - screen_failure_rate)
```
| Screen failure rate | Screened per 100 enrolled |
|--------------------|--------------------------|
| 10% | 112 |
| 20% | 125 |
| 30% | 143 |
| 40% | 167 |
| 50% | 200 |
Charged on screened: screening assessments, screening labs, consent time, and
the site's screening effort.
Charged on enrolled: the visit schedule, incentives, treatment, and follow-up.
The same logic applies outside clinical work under a different name. In product
research the equivalent is recruiting yield: if 30% of screener respondents
qualify and 70% of qualified participants actually attend, you pay for screener
completes and no-shows regardless.
| Research recruiting stage | Typical yield |
|--------------------------|---------------|
| Invitation → screener complete | 5-20% |
| Screener complete → qualified | 20-50% |
| Qualified → scheduled | 50-80% |
| Scheduled → attended | 70-90% |
Multiply these through: reaching 8 attended sessions typically means several
hundred invitations. Budget the funnel, not the sessions.
## 3. Cost drivers by method
### Interview-based research
| Line | Driver | Notes |
|------|--------|-------|
| Recruiting | Sessions × yield | The dominant cost for hard-to-reach populations |
| Incentives | Sessions | Scales steeply with seniority of participant |
| Moderation time | Sessions × (session + prep + debrief) | Budget 2.5x session length in total |
| Transcription | Session hours | Cheap; do not skip it — it makes synthesis auditable |
| Synthesis | Sessions | Roughly 1-1.5 hours per session of analysis |
| Readout | Fixed per study | |
**The cost nobody counts:** moderator and notetaker time. A "cheap" 8-session
study consumes roughly 40-50 person-hours across preparation, sessions,
debriefs, synthesis, and readout. Where that time is not budgeted, it is taken
from something else.
### Survey research
| Line | Driver | Notes |
|------|--------|-------|
| Panel sample | Completes × incidence | Cost rises sharply as incidence falls — a 5% incidence audience can cost several times a general-population complete |
| Instrument design | Fixed | Underestimated; a good instrument takes days |
| Programming and testing | Instrument length | |
| Incentives (owned list) | Completes | Much cheaper than panel |
| Analysis | Fixed plus subgroups | Each reported subgroup adds analysis cost |
**Incidence is the dominant driver.** A survey needing 400 completes from an
audience representing 4% of the panel requires screening roughly 10,000 people,
and screening costs money even when the respondent does not qualify.
### Experiments
| Line | Driver | Notes |
|------|--------|-------|
| Engineering build | Variant complexity | Nearly always the largest cost |
| Instrumentation | Number of new events | |
| Analysis | Fixed | |
| Opportunity cost of traffic | Duration × exposed users | Real when the losing variant underperforms |
**The cost nobody counts:** engineering time. Experiments carry no external
invoice, which makes them look free next to a study with a vendor quote. Count
the engineering days at a loaded rate, or the portfolio comparison is
systematically distorted in favour of experiments.
### Clinical studies
| Line | Driver | Notes |
|------|--------|-------|
| Site startup | Sites | Contracts, ethics, training, initiation |
| Per-participant grants | Enrolled × visits | The visit schedule drives this directly |
| Screening costs | Screened | Grossed up for screen failure |
| Site coordination | Site-months | Per-site-month, for the full duration |
| Monitoring | Site visits | Scales with sites, duration, and risk-based monitoring plan |
| Central lab / imaging | Participants × timepoints | |
| Data management and EDC | Participants and CRF complexity | Build cost plus per-participant |
| Biostatistics | Fixed plus interim analyses | Each interim is additional |
| Medical writing | Fixed | Protocol, CSR, publications |
| Regulatory and ethics | Sites and countries | Multi-country multiplies submissions |
| Insurance | Participants and risk class | |
| Pharmacovigilance | Duration and event rate | |
## 4. Contingency, overhead, and escalation
### Contingency
| Study type | Recommended contingency |
|-----------|------------------------|
| Short, single-site, well-precedented | 5-8% |
| Standard multi-site study | 10-15% |
| First-in-population, novel endpoint, or new geography | 15-25% |
| Multi-country regulatory study | 15-20% |
**[RECOMMENDED]** 10% is the practical floor for anything multi-site. Below
about 8%, every protocol amendment becomes a change request, and change requests
consume management attention out of all proportion to their value.
### Overhead
Overhead (indirect cost recovery) varies enormously by organisation and funder,
and some funders cap it or exclude specific categories. Two rules:
1. **State whether the quoted number is inclusive or exclusive of overhead.**
The same study is quoted at very different totals depending on this, and the
mismatch surfaces after award, which is the worst time.
2. **Check what the overhead base is.** Some organisations apply it to direct
costs only, some include contingency, and some exclude participant incentives
or equipment. The base changes the total materially.
### Escalation
For studies over 24 months, include an explicit escalation line. Site grants,
staff costs, and vendor rates all rise over a multi-year study, and a budget
fixed at year-one rates is short by the time it reaches close-out.
## 5. Cost per insight
Cost per insight is blunt but forces a comparison across methods that otherwise
get evaluated in isolation.
```
cost_per_insight = total_cost / decision_ready_insights
```
Two disciplines make it meaningful:
- **Count only decision-ready insights** — those that met a stated confidence
bar. Counting every observation makes any method look efficient.
- **Count all costs including internal time.** Otherwise methods with external
invoices look expensive relative to methods that consume salaried hours.
### Method comparison
| Method | Relative cost per insight | Why |
|--------|--------------------------|-----|
| Existing-artefact analysis (tickets, calls, logs) | Lowest | The evidence is already paid for |
| Instrumentation analysis | Low | Assumes instrumentation exists; otherwise the build cost dominates |
| Interview round | Moderate | Recruiting and internal time dominate |
| Survey | Moderate | Panel cost dominates; falls sharply on an owned list |
| Experiment | Moderate | Engineering time dominates and is usually uncounted |
| Diary study | High | Long duration, high attrition, heavy analysis |
| Multi-site clinical study | Highest by orders of magnitude | Regulatory and site infrastructure dominate |
The ordering is more useful than the absolute numbers. Its main practical use is
forcing the question: **has the cheap evidence been mined before the expensive
evidence is commissioned?** In most organisations the answer is no.
## 6. Commonly omitted line items
Run this list against any draft budget.
- [ ] Screening costs grossed up for screen failure
- [ ] Recruiting funnel costs, not just completed sessions
- [ ] No-show and replacement costs
- [ ] Site coordination as per-site-month for the full duration
- [ ] Monitoring scaled to sites and duration
- [ ] Data management build **and** per-participant costs
- [ ] Statistics beyond the initial analysis plan
- [ ] Each interim analysis costed separately
- [ ] Transcription and translation
- [ ] Internal moderator, analyst, and engineering time at a loaded rate
- [ ] Close-out: database lock, site close-out visits, archiving
- [ ] Final report and any publication costs
- [ ] Record retention for the full required period
- [ ] Insurance
- [ ] Currency and inflation escalation on studies over 24 months
- [ ] Protocol amendment contingency
- [ ] Regulatory submission fees per country
- [ ] Participant travel and reimbursement
- [ ] Equipment purchase, calibration, and disposal
The last-third of that list — close-out, retention, reporting — is where studies
most often run out of money, because it falls at the end when the budget is
already committed and the appetite for a supplementary request is lowest.
## 7. Build versus buy for research capability
| Capability | Build in-house when | Buy when |
|-----------|--------------------|----------|
| Recruiting | You have an owned list and recurring need | Hard-to-reach audience, or one-off study |
| Moderation | Research runs continuously | Specialist domain or language needed |
| Survey panel | Never — panel access is not worth building | Always |
| Transcription | Never | Always; it is cheap and it is not your differentiator |
| Analysis | Always — this is the capability worth owning | Only for specialist statistical work |
| Clinical monitoring | Large ongoing programme | Single study or few sites |
| Data management / EDC | Rarely; validated systems are expensive to build | Almost always |
The general rule: **buy the infrastructure, build the judgement.** Panels, EDC
systems, and transcription are commodities with real economies of scale. The
ability to choose the right method and interpret the result is the capability
that compounds, and outsourcing it hollows out the function.
## 8. Rate benchmarking
Rates vary widely by geography, seniority, and specialism, so treat these as
structural guidance rather than as figures to quote.
| Cost | Scales primarily with |
|------|----------------------|
| Participant incentive | Seniority and session length. Executive and clinical-professional incentives are multiples of consumer rates. |
| Recruiting fee | Incidence and reachability, not sample size. A low-incidence B2B audience can cost several times a general-population recruit per participant. |
| Panel complete | Incidence and instrument length. Screening out non-qualifiers is billed. |
| Site grant per participant | Visit count and procedure intensity, not therapeutic area alone. |
| Monitoring visit | Travel and site count; risk-based monitoring reduces frequency, not unit cost. |
| Statistical work | Design complexity and number of analyses, not sample size. |
### Getting comparable quotes
Vendors scope differently, which makes headline quotes non-comparable. Fix the
scope before requesting quotes:
- State the exact deliverable, not the activity ("8 completed 60-minute sessions
with recordings, transcripts, and a themed summary" — not "user research")
- State who owns recruiting and who pays incentives
- State the number of revision rounds included
- State the data ownership and retention terms
- Ask for the price of one additional unit (session, complete, site) so you can
model scope changes without re-quoting
That last item is the most useful and the most often omitted. Marginal unit
pricing is what lets you decide mid-study whether two more sessions are worth it.
## 9. Cost control during the study
| Control | Mechanism |
|---------|-----------|
| **Scope freeze after fielding starts** | Instrument and protocol changes mid-field invalidate collected data and cost twice |
| **Marginal-unit pricing agreed up front** | Lets you scale within a study without re-negotiating |
| **Invoice against milestones, not calendar** | A vendor invoicing monthly for a stalled workstream is a cost with no delivery |
| **Cap revision rounds contractually** | Unlimited revisions are where fixed-price research work overruns |
| **Weekly recruiting-funnel review** | Recruiting shortfalls are cheap to fix in week one and expensive in week four |
| **Single named budget owner** | Diffuse ownership is the most reliable predictor of overrun |
### When to stop spending
Define these thresholds before the study starts:
- **Saturation reached** — two consecutive sessions with no new theme. Continuing
to fill a booked schedule past saturation is pure cost.
- **Question answered** — if the evidence is already decisive, stop. Studies
routinely run to their planned n after the answer is unambiguous.
- **Question superseded** — the decision was made, deferred, or the situation
changed. Stop immediately.
- **Recruiting failure** — if the funnel cannot deliver the sample, re-plan
rather than extending indefinitely at full burn.
The saturation stop is the most commonly missed. Sessions 9 through 12 of a
homogeneous-segment study frequently produce no new themes and full cost.
## 10. Attributing shared costs
Research functions carry costs that no single study owns: panel subscriptions,
research repositories, analysis tooling, and the research-ops staff themselves.
Three allocation approaches:
| Approach | Mechanism | Best for |
|----------|-----------|----------|
| **Central overhead** | Held centrally, not charged to studies | Small portfolios; keeps study budgets simple |
| **Per-study allocation** | Divided by study count or by study cost | Medium portfolios; makes true study cost visible |
| **Usage-based** | Charged by sessions, seats, or participants consumed | Large portfolios; drives efficient consumption |
**[RECOMMENDED]** Per-study allocation by study cost for most organisations. It
makes the true cost of a study visible without the administrative overhead of
usage metering, and visible true cost is what makes the cost-per-insight
comparison across methods honest.
Whichever you choose, be consistent, and state the method alongside any
cost-per-insight figure. A number computed excluding shared costs is not
comparable with one that includes them, and mixing the two invalidates exactly
the cross-method comparison the metric exists to support.
## 11. Budget review checklist
- [ ] Every cost assigned to the correct block (participant / site / site-month / fixed)
- [ ] Screening costs grossed up by 1/(1 − screen failure rate)
- [ ] Recruiting funnel yield modelled, not just final sessions
- [ ] Site coordination charged per site-month across the full duration
- [ ] Internal time costed at a loaded rate
- [ ] Close-out, reporting, and retention explicitly funded
- [ ] Contingency at or above 10% for multi-site work
- [ ] Overhead base stated; inclusive/exclusive made explicit in the quoted total
- [ ] Escalation line present for studies over 24 months
- [ ] Cost per enrolled participant computed and benchmarked
- [ ] Cost per decision-ready insight computed
- [ ] Fixed costs below about 55% of the total, or the concentration justified
- [ ] Every assumption traceable to a source or a comparable study
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |