HBI Deals+Insights / Interviews

What makes a successful investment in AI in 2025 — in conversation with L.E.K.’s Klaus Boehncke and Guillaume Duparc

AI has become somewhat of a buzz word in recent years, and we at HBI are no strangers to hearing about its implementation in healthcare. But what impact is it actually having on healthcare businesses, and what potential does it hold?

HBI spoke to Klaus Boehncke, Digital Health Lead for L.E.K. Consulting and his Healthcare Partner colleague Guillaume Duparc at L.E.K. — who are due to moderate a session on ‘What is AI’s equity story?’ at the HBI conference — about the AI opportunities that exist for healthcare businesses and investors and how to tap into them, overcoming shortfalls, and why 2025 is a defining year for AI. 

What are the AI opportunities for healthcare businesses and their patients?

While Boehncke emphasises the equity story depends on the company and use cases in question, in general he says you can think of AI impacting both the front (clinical) and back (administrative) office of healthcare provision.

“In healthcare, what we see happening is AI impacting the administrative processes first because it’s the easiest area to implement, it is often less regulated, and usually it matters less if the AI doesn’t get things 100% right, because humans don’t do that either. If you have an AI system taking bookings and it’s wrong occasionally, it’s not going to be life threatening for the patient.” The majority of call centre calls can be successfully automated.

Boehncke points to a European unicorn like Doctolib that allows patients to book appointments online, which recently acquired a German AI service (aaron.ai) that now allows the platform to take voice phone calls from patients on behalf of doctors.

Duparc adds, “We are also seeing companies leveraging this type of technology for processes that are getting closer to clinical care. For example, Optegra, a leading international ophthalmology healthcare provider, is using a voice-based GenAI solution to conduct pre and post-operative assessments with their patients over the telephone, without human intervention. After initially running a successful pilot test with 2,000 patients, the system is now fully live. Patients love the experience and give it an exceptional Net Promoter Score (NPS) score of 97%.” The vast majority of calls are fully automated.

Boehncke says over the next couple of years we will see AI making more inroads in full clinical applications. He cites a recent randomised clinical trial led by Stanford University, where medical doctors were asked to review patient test cases. The group using traditional reference materials received a reasoning score of 74%, the group with ChatGPT-4 assistance scored 76%, but the AI by itself was 92% accurate. “There was a lot of discussion in the research community around the result, and why the doctors plus AI group did not fully leverage the capabilities of the system. Scepticism over potential hallucinations (i.e. wrong results) may have played a role in this instance, but we can easily imagine such technology being accepted and used more widely in the future.”

The impact of AI in terms of value creation

Duparc says that value creation in the private sector is currently focused more on administrative than clinical processes (for example call centre services enhanced via AI, faster creation of documents), and depends on specific use cases which in total can impact around 20% of productivity.

On the clinical side, it also depends on the specific process. “In some instances we are already beginning to see fully autonomous AI solutions. For example, Oxipit received European approval for autonomous reviews of chest X-ray images in certain types of cancer screening, with significant cost impact by removing expensive specialists from the loop.” 

Examples of successful investment in AI

Aside from Optegra, Boehncke adds, “Another impactful example is Diaverum, part of M42, one of the globally leading dialysis providers. One risk of dialysis is thrombosis. Diaverum’s patient journey is hyper-digitised and they have used this extensive clinical data from their patients and trained a system that is able to predict thrombosis one week before it happens. So you can now adapt how the patient is treated to avoid this happening. This can potentially save the patient’s life and avoid a hospital stay, making this a big win for everyone involved. 

“Another example of impactful AI use cases is a company called Rapid AI, one of the leaders in stroke care, diagnosis and treatment support. Through their software, they are able to extend the [often very time critical] treatment window for stroke in some cases by 24 hours, in which case life saving procedures can still be performed. From a patient, staff and hospital perspective, a real triple win.”

Considering shortfalls when investing in AI

Boehncke isn’t aware of any instance where the application of AI has gone dramatically wrong and patients’ lives have been impacted. “But what we have seen is patient data being used, for example, to train AI or shared with third parties without patient consent. There are of course lots of good ways of using data with patient consent and, according to our research, it’s actually relatively easy to get patient consent to share their data particularly for difficult to treat diseases including cancer or heart disease,” he explains. 

He flags there’s lots of ways to handle data appropriately, including synthetic data generated from real data, federated data access, and sharing anonymised data with approval. 

“I think you just need to be open and transparent with patients to avoid any backlash.” 

Duparc also advocates for standardising AI use by healthcare providers. “Technology evolves rapidly. If every doctor uses a different AI that they personally like then monitoring and compliance becomes very difficult.” He suggests hospital leadership contracts a curated set of AI solutions to prevent these risks. AI marketplaces can also help manage this.

What should CEOs and investors implement when investing in AI?

Duparc points out that it is really key to identify the most important use cases first:

“It is crucial to get an understanding of where AI can be deployed for maximum impact, across administrative and clinical processes as well as operations. This can help automate and enhance patient processes as mentioned in the Optegra example above, as well as develop new models of care.

“It is also important to develop a relevant data strategy. Besides using domain-specific healthcare AI models, you can create much better results when local clinical data is added to the mix, e.g. via a technology called Retrieval Augmented Generation. That makes an appropriate data strategy at a group level really important.”

Boehncke says that these types of efforts are best coordinated by creating a ‘centre of excellence’ for AI. 

“It ideally should include the company leadership, and, at minimum, the clinical leads and key opinion leaders within your hospital group, for example. That will enable you to create a core group of people that regularly pilot and evaluate the leading solutions.

“Such a centre of excellence setup is also what Diaverum did when they developed their ‘AI factory’ that developed the thrombosis detection solution mentioned above.”

Duparc adds, “Another good example of this is what Humanitas hospitals have done. They have created an AI centre that combines data scientists, medical doctors, digital experts and managers to develop decision support for the group.

“And then, after vetting the technology, very rapidly deploy it, making sure it’s centralised, and you’re taking people along on the change journey. You want to work with physicians, clinicians and surgeons and enable them to be their best by using the best possible tools, but also make them aware of the limitations of these technologies.”

Boehncke and Duparc acknowledge that change management in medical healthcare provider workflows is very hard, because administrators and doctors have learnt how to work the current IT systems, and any change needs to be as user friendly as possible and driven by clinicians themselves.

What current market trends are you seeing with AI?

Duparc mentions that both AI and major data platform technologies are starting to go mainstream and cites the NHS as an example:

“Besides the much publicised £330m Federated Data Platform that is being built by Accenture and Palantir, the NHS has also announced very recently an £11m AI trial to enhance breast cancer screening. Currently two radiologists are needed to evaluate every scan, and with AI support the aim is to assess the images with just one specialist.” 

He adds that there is now good evidence of AI’s impact in imaging. Recent clinical trials in Sweden and Germany have shown that this technology contributes to early detection of cancer while significantly reducing clinician workload and without increasing false positives. There is also more scrutiny on the algorithms on potential biases.

Boehncke mentions that private operators consider administrative use cases (e.g. call centre operations) and automation of clinical paperwork (letters, orders, referrals), diagnostic imaging algorithms to manage load and support decision making. Digital first providers are also working on deploying at greater scale for the asynchronous operations.

He adds, “On the clinical side with AI, solutions that are integrated with the electronic medical record (EMR) are likely to gain more traction in the future. For example, you can look at the voice AI solution from Nuance (part of Microsoft) which has been integrated into several EMRs. The advantage is that the AI system can draw on the data from the EMR in addition to information from the clinician and the patient.

“If you are an investor into a standalone voice solution, you will need to work on partnership agreements and make sure your AI ends up being integrated into some of those clinical systems, because ultimately the utility will be much higher.”

Duparc flags, “Again the public sector is leading some of the clinical testing of these technologies. For example, in the UK, Greater Ormond Street Hospital (GOSH) is piloting an AI tool that actively listens to doctor/patient conversations and automatically drafts clinical notes for the EMR and patient letters for the clinician.”

A defining year for AI

In L.E.K.’s Look Forward 2025 video, they describe this year as the “crunch” year for AI. 

“We’re moving out of the cycle where you have huge hype for something and then have the disillusionment, real applications with impact are already here,” says Boehncke. 

Duparc concludes, “What we will likely be seeing in 2025 is a lot more of these real-life impactful applications, like the examples that we gave [Optegra, Diaverum, Nuance], which are just the tip of the spear. This year we will see much more AI adoption in many more healthcare provider settings. A proper data and AI strategy will become essential for everyone.” 

We would welcome your thoughts on this story. Email your views to Hannah Millington or call 0207 183 3779.