Inside Clementine AI €1.7M Bet on Emotionally Adaptive AI

9 min read

A caller phones a hearing clinic in the Netherlands. She is not ready to book. She wants to ask whether the ringing in her ears means what she fears it means, and whether saying it aloud commits her to treatment. The voice on the other end does not get tired or rush her. Its pace does not suggest another caller is waiting.

The caller is representative of the uncertainty Clementine AI says it is trying to handle. The voice is Matthew, an AI agent built by the Maastricht startup founded in 2024. Answering at 2 a.m. is the straightforward part. Clementine is also selling the behavioural judgement surrounding the call.

Clementine’s pitch extends beyond speed, uptime and call volume. It says it helps hearing-care businesses convert more leads through behavioural science. Founder Vince van de Weijer describes the goal as reducing barriers, especially stigma, that keep people from seeking care. Lead investor Healthy.Capital calls it AI-driven activation.

Clementine offers a useful case for a wider commercial shift. AI companies are beginning to package behavioural techniques, emotional calibration and conversation design as product capabilities. The emerging proposition is that software can handle the uncertainty surrounding a request as well as the request itself.

A call that begins before the booking

Call centres automated routing decades ago, and scheduling software absorbed much of the calendar. Conversations with hesitant or frightened customers remained harder to standardise. In hearing clinics, dental offices and debt-collection teams, experienced workers listen for resistance and adjust. They may soften a word, slow down or give someone more time before asking for a decision.

Organisational sociologists describe this as emotional labour, the work of managing tone and feeling as part of a job. It is often poorly recognised even when it shapes the outcome of a call. Two receptionists can follow the same script and leave a hesitant caller with very different impressions.

Voice AI companies are now trying to encode parts of that judgement into software. Most healthcare agents are still sold through operational measures such as answered calls, scheduling, system integration and cost per appointment. Clementine adds another layer by linking those functions to behavioural science and the barriers that delay care.

The idea extends beyond hearing care. Debt collection, legal intake and sales all contain conversations in which people hesitate, avoid disclosure or resist a decision. Gartner expects 40 percent of enterprise applications to include task-specific AI agents by the end of 2026. As those systems enter more conversations, behavioural design can become a feature companies buy rather than an informal quality of a good employee.

Answered calls provide the clearest evidence

Clementine raised €1.7 million in August 2026 in a seed round led by Amsterdam-based Healthy.Capital. Its clearest named reference involves Virsono and Earlens. A customer executive said during a Clementine promotional webinar that the deployment reduced missed inbound calls to zero. The result is useful and attributable, though it has not been independently audited. It demonstrates operational value and gives the wider proposition a base from which to develop.

Clementine also advertises faster engagement and higher response rates using industry benchmarks rather than a transparent, company-specific dataset. These measures reflect the outcomes buyers are likely to care about. More detailed reporting would show whether behavioural adaptation contributed to them.

A fuller category test would compare AI and human staff handling similar conversations. Conversion would need to be separated by the caller’s starting point, since someone ready to book presents a different task from someone still deciding whether to seek care. The same gap between an overall score and performance on consequential tasks appears in contract review. I could not find the required breakdown in the available Clementine evidence.

The product contains several layers

A behavioural-science product can sit across several layers of an interaction. The system needs reliable operational access, signals it can interpret, rules or models for choosing a response and an outcome against which that choice can be tested. Strong call automation can exist without accurate emotional inference, and a warm response can sound convincing without changing behaviour.

Health researchers have spent years documenting behaviour-change techniques. The BCT taxonomy, first published in 2013, provides a shared vocabulary for social support, credible sources, self-monitoring and beliefs about capability. That vocabulary could help turn behavioural science from a broad product label into something buyers can inspect and compare.

A buyer could ask which techniques a system uses, which signals trigger them and how the resulting responses were tested. I could not find that level of design detail in Clementine’s published information, leaving an outside reader unable to assess how those techniques are implemented.

Evidence is emerging in other markets

Research from another setting shows that the commercial premise is plausible. A debt-collection study reported that AI agents collected more than human workers when their emotional tone matched the borrower’s situation. The advantage ranged from 49 to 94 percent, using warmth for minor early delinquency and firmness after repeated delays. Hearing care presents a different decision, but the study shows that emotional calibration can affect a measurable outcome.

A 2025 preprint explores how language models might identify emotional cues and adapt a conversation around them. It remains an early design study rather than evidence of reliable emotion recognition in live hearing-care calls. Together, the research suggests a possible technical layer while leaving its performance in specific markets open for testing.

Clementine has not published which of these dynamics, if either, shapes its product. I found no named behavioural scientist, design specification or academic collaboration in the material reviewed for this draft. More disclosure would make the product easier to compare as the market develops.

Trust changes the interaction

A 2025 Frontiers paper offers a conceptual account of simulated compassion. It distinguishes outward signals of care, such as patient language and attentive phrasing, from the human experience behind them. For a product category built around emotionally sensitive conversations, that distinction affects disclosure, expectations and trust.

Marketing research provides a second design consideration. A five-experiment study found weaker customer responses to an AI sales recommendation when the agent’s profit motive became obvious. A related research paper examines the customer’s awareness of being persuaded. Neither study concerns Clementine, but both show why an emotionally adaptive product must account for who benefits from the interaction. Similar conflicts emerge when AI help serves the company deploying it as well as the person receiving it.

These studies do not determine Clementine’s performance. They identify variables that any behavioural-science AI product may need to manage, including technique, disclosure, perceived motive and the user’s starting state. A natural voice and a final booking count capture only part of that system.

Europe adds a test around vulnerability

There is a more concrete question specific to where Clementine is built. Article 5 prohibits AI systems that exploit a person’s age, disability, or economic situation to materially distort their behavior. The European Commission’s 2025 guidelines on the prohibition use, as an illustrative example, a therapeutic chatbot aimed at people with cognitive disabilities that could exploit their limited capacity to influence them toward buying expensive medical products.

Hearing aids can be expensive medical products, and hearing loss is more common in older populations. That combination does not place Clementine inside a prohibited category. Article 5 requires additional conditions involving exploitation, material distortion and harm. The Commission example is analogous rather than a classification of hearing-care agents. It nevertheless shows why a European company designing emotionally adaptive health conversations needs to document the signals it uses and the decisions those signals influence.

The practical issue is documentation. A company using behavioural science in health-related conversations may need to explain which techniques it applies, what safeguards govern them and how it distinguishes assistance from pressure. That demand forms part of the wider European debate over AI explanations. Systems designed to adapt around vulnerability require a more detailed account of their behaviour than products limited to scheduling.

The market becomes easier to assess when several companies solve a recognisable problem, buyers allocate a budget and outcomes can be compared. Clementine supplies the problem in hearing care, while debt-collection and sales research show related techniques producing measurable effects elsewhere. Common evidence and buying criteria are less developed.

The next step is practical. Companies selling behavioural science through AI will need to show which behaviour they are trying to change, how the system chooses an intervention and whether the result improves without damaging trust. The answers will determine whether this becomes a durable product category or remains a feature described differently by each company.

Methodology note: Company-reported results are identified as such and have not been independently audited.

FAQ