The site selection visit request lands in your inbox six weeks before the sponsor wants first-patient-in. The coordinator who handled your last oncology study left in March. Your IRB slot for new submissions is already booked through mid-month. And somewhere upstream, an algorithm has already decided your site is ready.
That is the operational reality behind the ASCO and Ryght AI partnership announced this week, pairing the American Society of Clinical Oncology with an AI-driven site identification platform to select research locations for an upcoming clinical trial evaluating starting doses of CDK4/6 inhibitors in patients with metastatic breast cancer. The pairing makes strategic sense at the portfolio level. The implications for the sites being selected — and the coordinators who will carry this trial — are worth examining closely before the SIVs get scheduled.
What the Algorithm Sees vs. What the Coordinator Lives
AI site selection platforms work by ingesting published trial histories, investigator publication records, patient population proxies, and historical enrollment data. Ryght’s own published results are attention-grabbing: a global CRO used the platform to complete an oncology site selection campaign in 26 days — a process that would normally take six months — and secured 43 qualified sites against a target of 13, exceeding their recruitment goal by 330%. Those are numbers that make any CTM’s quarterly review look good.
But “qualified” in the platform’s output means something specific: the site has the patient population the trial needs, and an investigator with relevant publication history. What it cannot surface from a database is whether your oncology coordinator has three competing sponsor protocols running simultaneously, whether your pharmacy has IP storage capacity for another study, or whether the PI who looks great on paper has a site initiation visit attendance record that your CRAs would recognize as a warning sign. The algorithm selects on potential. The activation audit reveals the truth.
This gap matters more in oncology than almost anywhere else. CDK4/6 inhibitor trials in metastatic breast cancer require coordinators who can manage complex eligibility screening — prior treatment lines, performance status documentation, cardiac monitoring requirements, and concurrent medication reviews that generate deviation risk from day one. Sites I work with in this therapeutic area routinely report screen failure rates above 40% on initial enrollment waves, and the root cause is almost never the patient population. The root cause is a startup package that did not account for the actual protocol burden at the coordinator level.
That burden is not visible in historical enrollment data. A site that enrolled well on a previous CDK4/6 study may have done so with a different PI, a different coordinator team, and a different competing protocol load. The platform’s signal is real — it is just incomplete.
The Startup Gap AI Cannot Close
Here is the counterintuitive read on AI site selection: the faster you identify sites, the more dangerous the startup assumptions become. When a traditional manual process takes six months to build a site list, there is usually time baked into that delay for feasibility conversations that actually happen — clinical operations talking to coordinators, not just PIs. Compress that to 26 days and you have a list of algorithmically ranked sites, but you may not have sites that have been asked whether they want this study.
The FDA’s January 2025 draft guidance on AI and ML use in regulatory decision-making acknowledged AI’s role in streamlining patient recruitment and trial design, but did not define what constitutes a sufficient site readiness assessment when AI tools are used upstream. That regulatory silence means the SIV checklist — ICH E6(R3) section 5.22 on investigator qualification and section 8.2 on essential documents before trial commencement — is still doing all the quality gatework. The algorithm accelerates selection. The inspection standard does not move.
Across our network, the trials that hit first-patient-in within 45 days of SIV share one operational feature: the feasibility questionnaire was treated as a contract instrument, not a screening form. Every line item — available coordinator hours, pharmacy capacity, IRB pathway (central vs. local), and competing protocol schedule — was reconciled against the activation timeline before the budget was countersigned. When sites are selected by an AI platform and the activation timeline is already compressed to match the platform’s speed, that reconciliation step is the first thing that gets cut.
The Kitsa case study published by AWS reported 91% cost savings and a 96% reduction in data acquisition time compared to legacy manual processes. Those efficiencies are real. But cost savings in site identification shift the timeline pressure to site activation — and activation costs, the ones buried in startup burn, coordinator onboarding hours, and central IRB fee structures, do not compress at the same rate as a data pipeline.
What Changes Monday Morning
For sites receiving a selection notification from an AI-driven platform, the operational ask is simple: do not treat an algorithmically generated invitation as a readiness assessment. Before you accept the feasibility questionnaire, audit your current coordinator capacity against the protocol’s actual visit schedule. CDK4/6 trials with the complexity profile ASCO is describing — dose titration decisions, safety monitoring requirements, metastatic population documentation burden — will consume more coordinator hours than a standard oncology study. If your team is already at capacity on two active oncology protocols, that is information that needs to reach the sponsor before the SIV is scheduled, not after the first monitoring visit.
For sponsor-side counterparts and CTMs, the ASCO-Ryght model signals something worth building into your activation SOP: add an explicit feasibility reconciliation step that is triggered by AI-generated site lists specifically because those lists were built at speed. The site fitness data the algorithm used may be 12 to 18 months old. Population proxies do not reflect last quarter’s competing protocol activations. A single structured call — not a questionnaire, an actual call between the CRA and the site coordinator — before the SIV is scheduled is the cheapest quality checkpoint in your entire startup budget. At a typical Phase II oncology study cost structure, one failed activation costs more in reset time, budget amendments, and enrollment delay than every feasibility call you will run across the entire site network.
As of late 2024, only 11% of sponsors and CROs reported fully implementing AI and ML tools to support clinical trial activities, with another 22% reporting partial implementation. ASCO deploying Ryght AI for a cooperative group oncology trial is a signal that adoption is moving past early pilots into mainstream activation strategy. The sites that adapt fastest are the ones that build a standing readiness assessment process now — before AI-selected invitations start arriving faster than their startup infrastructure can absorb them. The algorithm will keep getting better at finding you. The question is whether your activation workflow is ready for the speed.
References
- FierceBiotech — “ASCO, Ryght AI team up for breast cancer trial site selection”
- Ryght AI — “Introducing RyghtSites.com: The World’s First Free AI-Powered Clinical Trial Site Search Engine”
- Applied Clinical Trials Online — “New Insights on the Impact of AI-Enabled Solutions”
- Realtime eClinical — “The FDA’s Draft Guidance for AI in Clinical Trials: Implications for Sites and AMCs”
- Amazon Web Services — “Kitsa transforms clinical trial site selection with Amazon Quick Automate”
