In February 2024, Nature Medicine published a study by Johanna Habicht, Sruthi Viswanathan, Ben Carrington, Tobias Hauser, Ross Harper and Max Rollwage on Limbic Access, an AI self-referral chatbot used in England’s NHS Talking Therapies services. Rather than acting as a therapist, the chatbot sits at the front of the system: it helps a person refer themselves into care, guides them through options, and gathers an initial mental-health assessment before a human clinician takes over. The question the study asked was simple and practical - does putting an AI at the intake door get more people into treatment?
The multisite observational study covered 129,400 patients across NHS services, comparing sites that adopted the chatbot with matched control sites that used standard self-referral. Services using the chatbot saw a 15 percent increase in referrals versus a 6 percent increase in the controls. The most striking result was who the increase reached: referrals from non-binary individuals rose 179 percent and from ethnic-minority individuals 29 percent, groups that typically face higher barriers to mental-health care. The authors framed this as narrowing an accessibility gap rather than just adding volume.
What makes the result notable is the design choice behind it. Limbic deliberately aimed the AI at the access problem - getting people in the door - rather than at the clinically and legally fraught task of delivering therapy. That positioning let it show measurable benefit while keeping humans responsible for treatment.
Why business readers should care: this is a clear case of AI creating value by removing friction at the start of a process rather than automating the expert work itself, and of measuring impact by who newly gets served, not just by throughput.