LLM or MD: Who Owns the Patient Journey?

Keneally (BCG, The So What) — LLM or MD: who owns the patient journey?

▶ Watch on YouTube · Boston Consulting Group · 21:17

Ben Keneally, BCG’s Asia Pacific health care services leader, explains why AI is now many patients’ first stop — often before they see a doctor. He argues that pairing AI with personal health records improves outcomes and equity, and that health leaders, not LLMs, must shape how this shift unfolds.

You’ll Learn: Why AI is becoming the default first step in care, especially in markets with limited access to clinicians. How pairing AI with personal health records improves treatment compliance and catches errors earlier. Why driving AI into end-to-end care pathways takes leadership, not just frontline experimentation.

TL;DR

A 21-minute episode of The So What from BCG on the Boston Consulting Group channel, published 14 July 2026. Host Georgie Frost interviews Ben Keneally, managing director and partner and BCG’s Asia Pacific leader for health care services. It accompanies BCG’s report Consumers Are Ready for AI-Enabled Health Care. Health Systems Need to Be, Too. (linked, not ingested). Five claims:

1. AI is the first step in care. “The biggest, fastest change we’ve seen in hundreds of years in how people access health care.” People use LLMs for assurance and to decide the next step, “a step on a pathway to treatment.”
2. Adoption is highest in developing nations. In a BCG survey of about 15 countries and thousands of internet-connected people, developing nations led, because free helplines and clinics are scarcer there. Within developed countries the wealthier use AI more; migrants with language barriers are a prominent use case.
3. From chat to agents. Next is an “always-on health agent” connected to wearables and records that advises on diet, sleep and exercise, books appointments and prompts immunisations, which raises open questions on who provides it (national systems or private firms) and how it is licensed and assured.
4. Pair the model with the person’s own records. Patients with access to their records catch errors in their treatment, comply better with it and know what to do when symptoms change; an LLM that explains the records adds context, “and we know that that leads to … substantially better health outcomes and, in fact, more equitable health outcomes.”
5. Health systems should shape it, not stop it. Two routes, “both/and”: influence the public models (“how do I make my information, my best practice guidelines … available to those models”) and bring LLMs behind the firewall, weighted to local guidelines and population. The unlock is leadership: frontline experiments need no one else’s agreement, while an end-to-end pathway needs everyone on board.

Key claims

Risk, in proportion

Asked about dangerous advice: “it happens from humans just as much as it happens from technology.” He cites an average of 18 years for best practice to move from the lab to most patients, a gap AI could compress. Commercial providers have “an incentive not to end up getting sued,” and regulators want good advice too, so he expects the risk to be manageable; the larger risk is settling for symptom checkers when “the opportunity here is so much bigger.”

Who owns the patient relationship

It is “a critical strategic question,” and it is not yet clear how consumer AI will make money: subscription, or free versions with advertising. Either way, providers and payers need a strong consumer experience and must make sure that when an LLM advises, it draws on their advice and reputation.

AI plus clinicians

AI can do supervised jobs on behalf of clinicians, such as applying simple algorithms to test results, but “health is a human-to-human contact service.” The aim is to use AI to remove administration and gatekeeping (for example, routine three-monthly check-ins for chronic disease) so that clinicians’ time goes into the relationship. Workforce shortages make this a necessity rather than an option.

Dynamic-capabilities reading

• contextual/external-triggers: changing consumer behaviour, with patients going to AI before they see a clinician.
• strategic-renewal/business-model: the question of who owns the patient relationship, digital front doors with AI built in, and value-based care as the logic AI could serve.
• contextual/internal-enablers: executive leadership needed to redesign an end-to-end pathway, against frontline experimentation that needs no one’s agreement.

How it connects

• answer-engine-optimization — health systems trying to have their guidelines picked up by public models, the same problem as a brand’s.
• automation-vs-augmentation — AI plus clinicians, with administration automated and care kept human.
• ai-agents — the personal health agent as a consumer agent with access to private data.
• enterprise-ai-adoption — point-solution experiments against leadership-led pathway redesign.
• responsible-ai — licensing and assurance of health agents, and a risk weighed against human error.

Related in this wiki

• Amiel & Sanders (HubSpot UNBOUND) — “by far the most important factor” in AI answers is how often you are mentioned across the web; Keneally’s health systems face the same question about their guidelines.
• Peron (Philips) — radiologists were not replaced, and patients fear losing face time; Keneally argues AI should give it back.
• Parikh (Hippocratic AI) — a working version of agents that call the patients hospitals do not reach.
• Goel (BCG, The So What) — the same series and host, four months later, on what AI takes from the workforce.
• ai-in-health-care — concept page (created 2026-11-07) that sets this source beside the hospital and health-system surveys.

What was actually ingested

The full episode (21:17, length_seconds: 1277) through the yt-dlp fallback from the human-written track, 104 lines. The survey (about 15 nations, thousands of respondents) and the 18-year figure are reported as stated; the report was not read. No stills: two webcams.

Linked entities and concepts

• Boston Consulting Group — publisher and Keneally’s employer. Updated in this ingest.
• answer-engine-optimization, automation-vs-augmentation, ai-agents, enterprise-ai-adoption, responsible-ai — updated in this ingest.
• dynamic-capabilities — the three cells tagged above.

Dangling (single-source mention, deferred): Ben Keneally, Georgie Frost (host; not in author:).

Scope and reliability

A consultancy partner promoting a report, on his firm’s channel. The survey is described but not shown, and its sample is internet-connected people only, as he notes. The outcome claims about patients reading their own records refer to “fantastic research” that is not named. He is candid that the agent future raises unanswered questions on regulation and privacy.