Inside T-Pro's Enterprise Ambient AI Platform

T-Pro is going through a rapid growth phase, and I had a chance to get exclusive platform access and dive deep into what they built. I spent 10 days in the platform. I also spoke with five members of the team. One hour with Jonathan Larbey, CEO. One hour with Ben J., CCIO. One hour with the implementation team, who walked me through the full platform end-to-end. Here is what I found.

TLDR: T-Pro is not best understood as a standalone ambient scribe, but as an enterprise clinical documentation infrastructure layer. As note generation becomes easier to replicate, the harder problem shifts to workflow integration and AI governance. That is where T-Pro looks strongest.

I have structured this breakdown into six parts:

T-Pro is a healthcare AI-powered documentation platform founded in 2010 by Jonathan Larbey. Headquartered in Ireland, it operates directly across the UK, Germany, Australia, and New Zealand, with global reach through partnerships spanning East Asia and beyond. With over 1,400 client organisations and 137,000 monthly clinical users outside of partnerships, T-Pro has quietly become one of the most embedded documentation platforms in the NHS today.

They have been doing AI scribing since 2019, well before the current wave of ambient voice technology companies existed, and they specifically specialise in enterprise-scale healthcare deployment.

Two things stood out before I even got into the platform. Without receiving a penny of VC funding, T-Pro reached this scale by growing revenue profitably at 22-24% year-on-year. While many companies raise heavily and burn cash to acquire customers, T-Pro took the opposite path.

What surprised me most came from Ben. Unlike most vendors who licence their speech recognition AI model from a third party, T-Pro owns, trains and governs theirs entirely in-house. That gives them direct visibility into model performance, accuracy, and governance that they can share transparently with clients.

Before taking you with me into the platform, let me first break down what they have built.

1) Speech Recognition

T-Pro Speech Recognition converts clinical speech into structured documentation in real time across mobile, desktop, and web. It works online and offline. It is not limited to the Scribe. This is the layer that runs across the platform and powers the scribe, dictation, document editing, and any clinical correspondence wherever speech input is needed.

Most ambient scribes licence their speech recognition from a third party, meaning the quality of their core output depends on a layer they do not fully control. T-Pro built and trained theirs in-house. The transcript this model produces feeds the LLM that structures the final document. Get the transcript right, and everything downstream improves. That ownership means they can measure performance by accent, specialty, and setting. Ben gave a specific example: if the Spanish-accented clinical workforce is underperforming, they can identify it and retrain.

Under the EU AI Act, demonstrating how you manage bias requires data, accent performance by demographic, specialty-specific error rates, training data provenance. A vendor relying on a third party is dependent on what that third party chooses to share. T-Pro does not have that problem. We will come back to this in the governance section.

2) Scribe

T-Pro Scribe is their flagship product. It captures clinical consultations in real time through web browser or mobile app and generates multiple AI-generated documents from a single encounter.

Every output, whether a GP letter, clinical note, or patient letter, is generated directly from the transcript rather than from a secondary document. It integrates with EPR patient schedules, and documents route automatically through configurable workflows to admin teams and back to the clinician for approval before filing.

3) Document Workflow

This is T-Pro's core platform capability - the foundation the scribe was built on top of. It manages every stage of clinical documentation from creation through to distribution: capturing, prioritising, routing, reviewing, approving, and releasing correspondence across teams, sites, and services. Every step is auditable. Turnaround targets can be set and monitored.

I see this as one of T-Pro's moats. And it is the section of the platform most likely to be under-appreciated by someone evaluating it as just an AI scribe. This infrastructure orchestrates the entire documentation process across every member of the team, integrating into varying clinical workflows. It is what separates a scribe from an end-to-end documentation system.

4) eClinic Manager

eClinic Manager is T-Pro's virtual care platform built for enterprise deployment across one-to-one, group, and video/phone consultations. It integrates directly with EHR schedules. Patients receive a secure link with no logins or downloads required, then a pre-session check to reduce DNAs and are routed to a virtual waiting room that lets clinicians track who is ready before the session begins.

Outcomes and follow-up tasks are captured within the platform, meaning virtual encounter documentation routes through the same governance and distribution workflows as everything else.

5) Connect

T-Pro Connect is the electronic document distribution layer built on enterprise-grade HL7 messaging. It automatically routes and delivers documents to GPs, external services, and other recipients in the correct format, whether the organisation is operating in a fully digital or hybrid paper environment. Distribution is auditable, traceable, and scalable across sites.

Beaumont Hospital saves over €160,000 per year by sending outpatient letters electronically rather than by post using Connect. This product represents the final mile in T-Pro's platform logic. Generating a good note is step one. Ensuring it reaches the right person, in the right format, at the right time, automatically, is the step that most ambient scribes never address.

This matters beyond cost. I have personally encountered patients whose care pathways were interrupted for months because a letter was never processed or never reached their GP. No referral chased. No follow-up booked. The clinical decision was made, but the chain broke at distribution. That is not a documentation failure. It is a patient safety issue, and it is one that a system like Connect directly addresses.

One thing that stood out during my time in the platform was how deliberately access is structured. I received the first three levels of access and will walk you through my experience in the workflow section. But first, let me break down what each level means.

T-Pro operates on a four-tier access model.

L1 - Clinician: Records consultations, generates documentation, approves outputs. The interface is deliberately simple. The clinician's job is to see the patient and approve the output, not to manage the system.

L2 - Admin / MedSec: Reviews and edits documents, adds attachments, confirms patient details, and finalises for distribution. This is where workflow complexity sits.

L3 - Manager: Real-time oversight across their department. The dashboard surfaces exactly where bottlenecks are forming - which clinicians have letters pending, which departments are missing turnaround targets, where support or training is needed, and where capacity gaps are emerging. A manager can see all of it in real time. And act.

L4 - Master: Full platform access across live and test environments. Handles billing, implementation sign-offs, and organisational-level configuration.

That granularity matters in an NHS environment where access control is part of clinical AI governance, not just IT hygiene.

Before accessing the platform, the implementation team set up a dedicated test environment with simulated patient profiles and configured the Clinician and MedSec accounts for me. What follows is my experience across the scribe and documentation workflow products:

I logged in as a clinician. In a live integrated deployment, my patient schedule would pull directly from the EHR. For this session, it is simulated as a night shift in the ED.

Carol is next. I click on her name and start the consultation. When I started recording, the screen turned red. Deliberately. When I am making eye-contact with the patient, it’s visible in my peripheral vision throughout - a constant reminder that the encounter is being captured.

During the consultation, I added contextual notes. This is one of the features that does not get enough attention. In a consultation, there are things you observe but would not say aloud. A differential that crosses your mind whilst the patient is talking. Safeguarding concerns. The patient's appearance. In Carol's case, I noted safeguarding concerns and that she appeared dehydrated.

I recorded the consultation and selected two output templates simultaneously. The clinical note and the GP letter were both generated independently from the same transcript. Not from each other, which is what prevents cascading omissions. In a referral chain, an omission in the note that carries into the GP letter is not a documentation error. It is a patient safety issue.

After the consultation, I decided to add antibiotics. I used dictation to capture it. No typing. No clicking. It was added to the record.

I approved both documents. I routed the discharge summary as high priority to the medical secretary given the safeguarding concerns, and approved the clinical note as final, which was filed directly to the EHR.

Now I switched accounts. Doctor's hat off. Secretary's hat on.

As a secretary, I go for the high priority item first. My role here is to finalise the discharge summary document for the clinician's approval and ensure the safeguarding workflow is completed on my end.

In my hospital, posting GP letters is standard practice - so I also need to ensure this one is not sent to Carol's home address due to her specific safeguarding concerns. I configure that on the T-Pro platform and add a note to her file. I go through the document, edit it, complete the safeguarding steps, and route it back to the author clinician for approval.

Doctor's hat back on. I reviewed and approved it.

Two observations before moving on. The consultation can be recorded in more than one way. The phone scribe (which I also tested) is an option on the same account, different device. Particularly useful on remote visits or where a computer is not accessible.

On dictation: T-Pro's dictation is structured, mapping directly into the template you have configured. I do not prefer dictating personally. But I know many clinicians who do, and for them, structured dictation into a pre-built template is a meaningful improvement over free-form recording.

As L1 (clinician): Straightforward. The platform stays out of your way and presents clear choices afterwards.

As L2 (admin): A different environment entirely. Letter queue, editing tools, distribution settings, routing paths. The L2 interface requires onboarding. It reflects the genuine complexity of what it is managing. A dedicated project manager handles that setup. One thing became clear navigating both sides of this platform: T-Pro is not a plug-and-play tool. It is an infrastructure deployment. It is what turns the platform into a documentation operating system.

As L3 (manager): Matheus from the implementation team walked me through the manager interface during our demo session. In a live deployment, the dashboard gives real-time visibility across the entire department - every document in the queue, at every stage, with full clarity on which clinicians have letters pending and which teams are hitting or missing their turnaround targets. Configuration of routing rules, account creation, and turnaround targets all live at this level. This is where the platform moves from a documentation tool to an operational management system.

Different departments route documentation differently. L3 is where that configuration lives and where these workflows are built.

An example of a workflow that can be created by the manager:

The platform augments the entire chain. Not just the note. The whole workflow. And it does not sit in isolation - it connects directly into the systems clinicians already use.

So, this is all great - I get the full workflow with integrations and see how the products are clinically grounded. What matters now is understanding how T-Pro governs the AI behind it. Ben J. had a lot to say on this.

Ben has made it a personal mission to raise clinical safety standards across the industry, not just within T-Pro. He attends hazard workshops. He chairs the Digital Health Networks CSO Council and has made it T-Pro's position to engage with every NHS trust's clinical safety process as a genuine partnership, not a formality.

The variation he sees across trusts in clinical safety rigour is, by his own assessment, a concern. Some trusts do not run hazard workshops at all. The clinical safety controls T-Pro provides cannot function effectively if the trust has not done its own requirements work first. This is an industry problem T-Pro is actively trying to solve through its clinical fellowship programme which consists of four fellows in total, two via the NHS AI Fellowship Programme, giving clinicians direct visibility into how the platform is built, tested, and governed.

T-Pro's VERA system scores documentation outputs against a modified PDQI-9 framework across ten metrics, including a hallucination checker and an omission checker. Most evaluation frameworks in this space focus on accuracy and hallucination. Recent research has found that omissions can be equally or more prevalent, a risk that is less visible but clinically significant. VERA accounts for both.

Pre-deployment, VERA is live in the Playground environment. Templates are tested against mock transcripts, scored, and refined before any clinician goes live. The L4 Master controls this test environment, evaluating template performance and reviewing VERA scores before any deployment proceeds. The traceability feature highlights exactly which section of the transcript produced each sentence in the output, addressing one of the most persistent clinical concerns about AI scribes: how do I know it did not hallucinate this? VERA gives a line-by-line answer.

Ben shared that automated surveillance is in final validation for post-deployment use. VERA will run continuously, flagging hallucinations and omissions automatically. Editing rates will be tracked as a proxy for model drift, so if clinicians start editing more after a draft is produced, the platform flags it, investigates, and takes corrective action before it becomes a clinical risk. That is not reactive quality management. It is preventive.

T-Pro maintains a full medical device lifecycle. Every step from design through development, QA, and deployment is documented. Each identified risk connects to a training control. Each control connects to a module in the T-Pro Academy. The chain is traceable end to end. Post-market surveillance via automated VERA scoring is in final validation. When live, it closes the loop entirely. That is what clinical AI governance looks like in practice, not a certificate on the wall, but a process that runs continuously.

So, what does the process of moving such a system into an organisation actually look like?

T-Pro does not offer a sign-up button. Before any deployment, T-Pro's project team conducts detailed mapping of clinical pathways, document types, routing requirements, and workflow configurations, specialty by specialty. This takes three months, by design.

Before go-live, templates are tested in the Playground environment. Ben walked me through a precise example. An ambulance assessment template scored 48 out of 50 on VERA. The system flagged that crew member names were missing from the output. Before deployment, T-Pro checked with the customer: is that a requirement in your documentation? It was. The template was updated. The score went to 50 out of 50. The point is not the score itself. It is that a gap in clinical documentation was caught and corrected before a single patient record went out.

At this stage, T-Pro shifts into ongoing support. A dedicated support team triages tickets, while a training team provides on-site support where needed. The T-Pro Academy delivers role-specific training mapped to identified risks, and usage dashboards allow the team to monitor adoption across services. If one clinician is using the platform in only 20% of clinics while another is using it in every session, that gap becomes visible. The team can then intervene with targeted support.

That monitoring also gives NHS leaders something more concrete than anecdotal feedback. The numbers an NHS executive can take to a board: utilisation dashboards, heat maps of approval activity showing whether clinicians are approving inside or outside working hours as a proxy for burnout, PDQI-9 benchmarks before and after implementation, and turnaround time metrics. Ben mentioned how one of their NHS trust clients reduced clinic letter turnaround from 27 days to 6 days after deployment.

AI literacy remains a separate challenge. It varies enormously across NHS organisations. The clinicians who could benefit most from these tools are often the least confident using them. An ambient scribe enthusiastically adopted by tech-savvy consultants while quietly failing the rest of the team is not solving the documentation burden. It is widening the inequality within it. T-Pro's Academy is being built with this in mind, but the problem is industry-wide and no single training platform solves it alone.

T-Pro is not building for the individual clinician who wants to sign up in five minutes. That market exists and is well-served by other product-led growth companies. T-Pro is building for the NHS trust that needs to deploy at scale, govern responsibly, and show measurable return to an executive board.

A three-month implementation timeline, a dedicated project team, specialty-by-specialty pathway mapping - these are all part of the product. As Jonathan mentioned, the alternative is a faster, cheaper deployment that generates notes and leaves every workflow problem exactly where it was.

That focus extends beyond the product into the markets they choose to operate in. T-Pro does not chase the US, which runs on different workflow structures and reimbursement models. That discipline is what allows them to go deep where they do operate rather than spreading thin across markets the platform was not built for.

One of the things I respected most across both conversations was the willingness to name where the platform falls short. Jonathan and Ben were transparent here:

From Jonathan Larbey:

From Ben J.:

The next phase for T-Pro is not more features. It is deeper process integration, post-market surveillance through VERA, and agentic AI.

As Ben framed it, the problem is not the note alone. A clinician sees a patient and then moves across multiple systems to complete everything that consultation generates - blood requests in one platform, imaging in another, follow-up scheduling somewhere else. The next step is to capture the clinical intent from the consultation itself and route those downstream actions automatically. Whilst SNOMED coding is live, clinical orders and further coding capabilities including ICD-10 and ICD-11 are in final stages of development.

Jonathan also shared two developments he was ready to make public.

The first is a formal partnership with OptumUK to launch EMIS Scribe. T-Pro's scribe technology is now built directly into EMIS, leveraging agentic capabilities to code and structure GP consultations directly into the record. Every GP practice running EMIS in the UK is a potential deployment.

The second is M&A. Acquisition targets are actively being pursued to broaden T-Pro's footprint across the NHS and beyond, with the explicit goal of accelerating deployment of technologies like Scribe at scale. Jonathan was clear that this is active, not aspirational.

What stood out to me was not only the workflow fit of T-Pro’s platform, but the maturity of the people behind it. Ben J. brings clinical informatics credibility that is rare in this space, and Jonathan Larbey was unusually transparent throughout - including on what had not worked. Selling certainty first and leaving the detail for later is not Jonathan’s approach, and that tells you something about the kind of company he is building.

The ambient scribe market is heading toward compression. A number of companies entered on the strength of a free trial, a polished interface, and aggressive user acquisition. AI scribing itself is becoming commoditised. The note is no longer the differentiator. Infrastructure is. And that is exactly where T-Pro has positioned itself. Over time, NHS buyers will ask a harder question: what has actually been embedded into workflow, and which of these vendors will still be standing in a few years?

That is where T-Pro looks strongest. Its defensibility is not just that it owns its speech model, though that matters. It is not VERA alone, though that is one of the most rigorous governance frameworks I have encountered in this space. It is the depth of integration. When document creation, letter routing, admin workflow, coding, and downstream distribution all sit inside the same platform, that becomes difficult to displace. That is not a tool. That is infrastructure.

One area I would still like to see develop further is greater self-serve configurability when it comes to the manager access, so that they can shape workflows themselves rather than requesting T-Pro’s team to make the changes. At the scale T-Pro is heading, that distinction will matter.

I am also excited to see how T-Pro develops its agentic AI capabilities over time. As the category matures beyond ambient documentation, the greater opportunity lies in helping clinicians move from captured intent to completed action in a governed way, and they already shared this is on the roadmap.

The companies that survive the coming consolidation will be the ones that went deep, not only wide. T-Pro went deep. And in a market where everyone is racing to generate the note, T-Pro already owns what happens to it after.

If you want to learn more or speak with the T-Pro team directly, you can reach them here.

_{*This deep dive is part of a sponsored collaboration with T-Pro, though the analysis is entirely my own. I maintained editorial independence throughout, and the company was open to all feedback - including limitations found. I regularly write about clinical AI tools, and this follows the same format and standards. If you or your company is interested in being featured in a neutral and credible way for an executive-facing healthcare audience, click here.}