Industry Report — IR-003

Site Selection in 2026: New Tools, an Old Problem, and Little Independent Evidence

Published November 2026 — At least eight companies announced AI-driven site selection or feasibility tools in 2026. This report reviews what each one claims, what data it uses, and how those claims compare with the published evidence on why sites under-enroll.

Executive Summary

A Crowded Year for Site Selection Technology

Choosing sites has long been slow and unreliable. A 2016 Tufts CSDD benchmark put site selection at about 3.2 months on average and described the path from identification to activation as a one-year cycle. In 2026, CROs, data companies, and site networks all announced tools intended to shorten it.

This report lists those announcements and sets them against the independent evidence. We have marked company claims as claims. We did not find independent validation of any of the results reported by vendors.

Key Findings

8
Launches Reviewed

Site selection or feasibility tools announced or promoted in 2026 by ConcertAI, PSI CRO (twice), Ryght AI, Fortrea, H1, WCG, and Velocity Clinical Research.

30–40%
Non-Performing Sites

A 2025 review in Applied Clinical Trials reported that many trials continue to see 30% to 40% of sites fail to perform.

0.79
Best Published Model Accuracy

A 2024 PLoS ONE study by Janssen researchers reported an AUC of 0.79 for telling the top 30% of sites from the bottom 30%. The authors described the model's overall explanatory power as low to modest.

0
Independently Validated Vendor Results

None of the performance figures in the announcements we reviewed has been confirmed by a party other than the vendor or its customer.

What Was Announced

The tools fall into four groups, defined by whose data they use.

1

CROs Using Their Own History

PSI CRO promoted its SYNETIC platform in February 2026, saying it covers more than 500,000 institutions and scores sites on enrollment history, therapeutic expertise, and geographic performance. In October it launched an automated site selection workflow with Appian, built on PSI's own systems and data warehouse. Fortrea published its Feasibility Intelligence Engine in April, describing a single platform that combines internal trial history with external datasets. Neither company has reported measured results for these tools.

2

Data Companies Profiling Every Site

Ryght AI launched a free site search engine in April, which it says covers 100,000 site locations in 192 countries. It reported that an unnamed CRO identified 43 qualified sites in 26 days for an early-phase oncology trial. ASCO began using the platform in May for a breast cancer study. WCG promoted its Trial IntelX product, which ranks countries and sites by past performance and which the company says is built on more than 80,000 protocols. ConcertAI launched a platform in February that it says can cut site selection, activation, and recruitment timelines by 25% to 50%.

3

Platforms That Involve the Site

H1 launched its Site Network Suite in April, in which sites can claim and manage their own profiles, confirm their capabilities, and answer questionnaires. Ryght has said that sites can review and update pre-filled answers in its paid feasibility product. These are the first signs of sites being given a role in the data held about them.

4

Site Networks Scoring Themselves

Velocity Clinical Research described a Feasibility Agent in mid-2026 that produces ranked site recommendations with probability scores from the performance data of its more than 70 sites, in hours instead of weeks. The data is first-hand, and it covers only Velocity's own sites.

5

What the Evidence Says About the Problem

The most-cited figures are old. A Tufts CSDD analysis of trials run from 2008 to 2010 found that 11% of activated sites enrolled no one and 48% under-enrolled. Some 2026 marketing attributes those figures to 2024. More recent work reaches similar conclusions: the 2025 review found inaccurate feasibility projections to be the most common concern in the literature, and the Janssen study found that site questionnaires often overestimate recruitment.

6

What the Evidence Says About the Solution

The Janssen model beat a baseline built on historical medians, which supports the case for data-driven selection. Its authors also listed limits: recent data only, claims data lacking laboratory results and outcomes, and U.S. sites only. A sponsor panel in October 2026 put the wider point plainly: historical data is not evidence of future enrollment.

What This Means for Sponsors

Every tool on this list will produce a ranked list of sites. The differences lie beneath the ranking.

Ask Whose Data It Is

A CRO's tool reflects the sites that CRO has used. A network's tool reflects its own sites. A data company's tool reflects public records. Each has blind spots, and none is neutral.

Separate Reported From Verified

Trial counts and registry entries show that a site took part in a study. They do not show how it enrolled, how clean its data was, or whether the staff responsible are still there.

Measure the Outcome

Most 2026 claims concern how fast a site list was produced. Ask instead for the share of selected sites that met their enrollment targets, and keep your own record.

Conclusions

The tools announced in 2026 will make site identification faster, and several will make it broader by reaching sites a sponsor would not otherwise have considered. Both are improvements.

They do not yet show that the sites chosen perform better. The data behind them is largely historical, often self-reported, and usually held by a party with an interest in the result. Until vendors publish outcomes, sponsors should treat these tools as a faster first pass and continue to ask for verified, site-level evidence.

Where SiteScore™ Fits

Verified, Independent, and Comparable

SiteScore™ addresses the gap this report describes. It is independent of any CRO's portfolio, and it scores every site on one 0–100 scale across execution, operational depth, market access, financial discipline, and risk and compliance. Alongside each score, a Data Confidence measure shows how complete and well-evidenced the underlying data is.

A PreScore draws on public registries. A Validated Score is checked against the site's own documents and reviewed before it is issued. Sponsors can search scored sites, build shortlists, compare candidates side by side, and request validation for sites they are considering.

See How SiteScore™ Works

Standardized, document-verified scoring for clinical research sites, used by sponsors, sites, and investors.