1. Decide what you are actually buying
Four different products get sold under the same phrase. A media agency buys attention. A patient community borrows trust. A site network or record-mining platform finds patients who are already in the system. A bundled CRO sells you accountability for the whole study. If lead volume is not your bottleneck — if sites are drowning in referrals and still not randomizing — buying more volume makes the problem worse.
2. Ask for the one number that matters
Referral-to-randomization ratio, from three completed studies in your indication and your countries. Not impressions, not registrations, not pre-qualified leads. If the answer is that the data belongs to a previous client, ask for the range. A vendor that measures itself on randomizations will know it instantly.
3. Price everything as cost per randomized patient
Take total program cost, including media and internal site burden, divided by expected randomizations. Two proposals that look 40% apart on fees frequently land within a few percent of each other on this number — and occasionally invert.
4. Test the site burden honestly
Every referral is work for a coordinator who already has a full day. Ask how referrals arrive, whether the vendor books appointments or just hands over contacts, who chases no-shows, and what happens at 2am when a candidate submits a form. A vendor with no answer to this will generate referrals your sites quietly stop working.
5. Write the contract so performance is visible
- Define 'qualified referral' in writing, with the exact pre-screening criteria attached.
- Require weekly funnel reporting at referral, contact, pre-screen pass, consent, and randomization.
- Set a review gate at 90 days with an exit right, not a 12-month lock.
- Agree who owns the candidate data and what happens to it at the end.
- Put media spend on a separate line from fees so you can see the margin.
6. Check your own assumptions before you blame the vendor
A meaningful share of enrollment failures are baked in before a vendor is hired: an eligibility criterion that excludes most of the treated population, a visit schedule no working adult can meet, or a site list chosen on relationships rather than performance. If the forecast was wrong, no vendor can outrun it. Our sister publication Rock Enroll covers enrollment diagnosis and rescue for sponsors and CRO teams.
Then shortlist
Start from the full directory or a head-to-head comparison. Three vendors is the right shortlist. Five is a procurement exercise nobody has time to run properly.