In this field, that is the only question that matters. Six months ago the tools were different, the prompting techniques have been superseded, and the model your Medical Affairs team was taught to use carefully now behaves differently than it did in the workshop.
And if the answer is that you have not trained them yet, that is the easier of the two problems on this page.
The AI Enablement Institute is a continuous AI capability programme for pharmaceutical and biotech teams. Monthly live sessions with Dr Andree Bates. A library that grows every month. Function-specific tracks built around the work your team actually does — the same work Eularis has been doing inside pharma functions since 2003 — and a curriculum rebuilt as fast as the technology and the regulations move.
Because in AI, trained once means trained then. It does not mean capability.
• AI strategy and systems built inside pharma since 2003, across every major function
• 40+ public endorsements from pharma executives on LinkedIn
• Trusted by teams at the majority of the world’s twenty largest pharmaceutical companies
The first is the obvious one: AI training that worked, and then faded. Tools change, guidance moves, and what a team learned stops being true.
The second is quieter: A company-wide AI literacy programme or training, that never fitted the job. A company-wide AI literacy or fluency programme, refreshed on a schedule and properly documented, fixes the first completely and can leave the second entirely untouched.
Most large pharmaceutical organisations have made the first move — the maintained, company-wide rollout — and stopped there. Most smaller biotechs have the first failure in its purest form, and no one whose job it is to notice.
Memory erodes over months, and fastest on exactly the accuracy-critical judgement regulated work depends on. Relevance erodes faster: models are retired on published schedules, new ones arrive every few months, and guidance moves on its own rhythm.
Memory gives you months. The tools and the rules give you a quarter. Capability has to be maintained to the faster of the two, which is why ninety days.
The evidence behind both curves, including the two meta-analyses the memory curve is drawn from, is in section 2 of the white paper.
And it erodes unevenly. Daily habits hold. The workflows run quarterly — the payer dossier, the periodic safety update, the submission that comes round twice a year — go first. Those are exactly the tasks where AI saves the most time.
Most companies answer this with a workshop. It goes brilliantly, the scores are excellent, and ninety days later the team is working from a version of AI that no longer exists, with the confidence of people who have been trained.
In an unregulated industry that is inefficiency. In yours it is exposure.
Many organisations now do — under that name, or as AI literacy, or as AI fluency — and if that is yours, you are ahead of most of your peers. It solves one problem completely and another not at all.
It solves availability. Content does not age on a shelf. New joiners can be enrolled. There is a record.
It does not solve fit. A company-wide curriculum is, by construction, the same curriculum for everyone. It teaches what AI is, where it fails and what the policy says — identical for a medical writer and a supply chain analyst. The work is not. The judgement a regulatory writer needs to decide whether a generated summary faithfully represents its source has almost nothing in common with the judgement a brand manager needs to decide whether a generated claim is substantiated. Neither is taught by a course about AI.
The floor is common. The job is not. The generic courses teach what AI is, where it fails and how to work with it. Article 4 names that layer, a company-wide rollout delivers it, and if you have already done one you should keep it. What none of them can teach is your own job: whether this generated summary is faithful to its source, whether this generated claim is substantiated, what has to be in the audit trail once a model has touched a document inside the quality system. That is not a higher grade of AI skill. It is a different subject, and it is the only one that changes what happens on a Tuesday.
The tick in the box is real. It is just not a tick in the box that was asked about — and that is not a failure of the programme. It is what a company-wide programme is for. It was never going to teach anybody their own job.
Which is the part worth sitting with. Neither failure is really a failure of training. Both are what happens when capability is bought the way training is bought — as an event, from a catalogue, against a discretionary line, by someone whose job is learning rather than the work.
The question is not which AI course to buy next. It is who understands your function’s work well enough to build the thing that would hold.
It is called an Enablement Institute rather than a training programme for a literal reason: the aim is a team enabled to do its own job properly with AI, compliantly, not a team that has completed a course.
For over twenty years Eularis has done AI strategy and AI builds inside pharmaceutical and biotech companies — discovery, clinical, regulatory, medical, pharmacovigilance, quality, supply chain, market access, commercial and the corporate functions. Hundreds of implementations, on six continents, for the majority of the world’s twenty largest pharmaceutical companies. Not courses about AI. Systems that had to work, inside workflows that were audited.
The Institute is downstream of that work. Every track exists because a function asked the same question often enough that it needed an answer which outlived the meeting. The judgement layer exists because we have watched where AI-assisted output actually fails once a regulator reads it. The vendor evaluation course exists because we have sat on the buying side of these tools and seen how the claims hold up.
That is a different starting point from a learning platform that has commissioned some pharma content, and it shows up in the only place it matters: whether a module tells your team what AI is, or tells them what to do differently on Tuesday.
One consequence worth naming. AI in pharma is not only language models. Several of the methods that still lead in discovery — variational autoencoders, generative adversarial networks — predate the current generation of tools by a decade or more, and nothing has replaced them. Dr Bates was applying AI to pharmaceutical problems in 2003. It is worth asking any provider when their AI experience begins.
The Digital Omnibus on AI — Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026 and in force from 27 July — rewrote Article 4 of the EU AI Act in full, replacing a duty to ensure a sufficient level of AI literacy with a duty to take measures supporting its development. National supervision under the Act began the following month.
That reads like a relaxation. It is a change of kind. Reaching a sufficient level can be discharged by a single intervention. Taking measures supporting development is an obligation of effort — and effort is not a state you arrive at. It is demonstrated continuously, by what is done and what can be shown to have been done.
The new text goes further, and says in terms that it does not require anyone to guarantee a specific level of AI literacy in any individual. Read quickly, that is a second relaxation. Read properly, it is the same change again. If no outcome can be required, no outcome can be the test. What is left to examine is what you did, how regularly, and whether you can produce the evidence.
The same article now also obliges the Commission and the member states to support those efforts, including by publishing practical examples of what compliance looks like. Those examples become the reference point. They will also move.
Two things most organisations have not registered. Article 4 applies at any risk level, so it reaches the general-purpose assistants your teams already have open. And when the same Omnibus deferred the high-risk obligations to 2027 and 2028, it did not defer Article 4.
The Act reaches organisations established outside the EU wherever the output of their AI systems is used inside it. A submission that goes to the EMA. Safety data covering EU patients. Material used by an EU affiliate.
Where the work is done does not decide it.
General information, not legal advice. No ongoing capability programme, including this one, makes an organisation compliant with Article 4. Section 3.3 of the white paper sets out the position in full, with the regulatory note.
Most organisations made one of the two moves and treated it as the destination. Going from episodic to maintained solves availability. It does not, on its own, change what anybody does on a Tuesday.
The evidence behind the ninety-day figure. What Article 4 now requires of a regulated business, in full. The three costs that never appear on a purchase order. The arithmetic for a fifteen-person biotech and a five-hundred-person function. And the eight requirements to hold any AI capability provider to, including this one.
No sequence and no list swap. The paper, and an occasional note when something material changes in the regulations. Unsubscribe anytime.
The shared curriculum for regulated work: GxP validation and audit trails, data readiness, security and confidentiality, shadow AI and prompt poisoning, who owns AI-generated output, bias and hallucination management, how to build a value case your finance function will accept, and how to judge a vendor — including whether to build or buy at all, building compliance agents for your function, and a lot more. Common because the regulatory reality is common, not because it is introductory.
Deciding whether an AI output is fit for regulated use is a skill, and it is taught as one: what to check, in what order, against what source, and the point at which the answer is that no amount of checking makes this output usable. Every team gets the same standard, with the checklists to apply it, because a claim reviewed to one standard in Medical Affairs and a looser one in Commercial is not a standard.
AI applied to the tasks your function actually performs, with prompt guides your team can use the same afternoon. Medical Affairs learns AI for medical affairs. Regulatory learns AI for regulatory. Routing is by function, so nobody has to sit anything to be placed. Leadership and Strategy is a track in its own right; some organisations start there, because the people setting AI policy are often the ones furthest from the tools.
A regulatory writer is working on whether a generated summary is faithful to its source — the ways fidelity fails silently, how to catch each one, and what has to be in the audit trail once a model has touched a document inside the quality system.
A brand manager is working on claim concepts that survive MLR, the substantiation trail that gets them through it, and the point at which a model drifts quietly into territory the label does not support.
Neither of those is taught by a course about AI. Both sit on the same Foundations, and nothing above them is shared.
Questions submitted in advance and answered anonymously, which is what lets teams from competing companies sit in the same room and ask what they would never ask on a webinar with their name attached. Each session works a real problem end to end. The first session of each quarter covers what materially changed in the preceding ninety days.
Every module is recorded and taken when it suits the person. The live session is optional and recorded like everything else. There is no schedule to protect and no day to block. And, if you want to know how to do a specific thing, AI will surface exactly where that is, in which video. More below.
Not a keyword search across module titles. Your team asks in their own words — how do I check a generated summary against its source, what has to go in the audit trail, has anyone asked about this tool — and it returns the module, the point inside it and the timestamp, including from the recorded live sessions where the question has already been put to Dr Bates or in one of the training videos. Not a hallucinated answer. Not a keyword search. A real answer from a pharma AI expert who has worked with AI across all pharma functions. The AI does the finding of the answer, but it never does the answering.
They watch ninety seconds and go back to work. This matters most for the workflows that come round quarterly, which are the ones a team has always half-forgotten by the time the next one lands.
Guidance optimised for a previous generation of model produces worse results on the current one, and it fails silently — so someone has to be watching for it. Every track carries the compliance layer: GxP-aware workflows, EU AI Act and FDA updates, disclosure and documentation practice.
Per-person completion and progress, visible to the functional sponsor for their own team rather than only in aggregate. Quizzes and certification are being added, which moves the record from what was watched to what was understood. This is the difference between believing your team is trained and being able to show it.
Private office hours for your team alone, private live add on training sessions, built around your workflows, your tools and your governance context.
Many teams start with a live launch session: it blocks out the time, creates the momentum, and turns the library from something available into something being used.
Delivery into your own LMS, and SSO. Each quoted on request.
Pharma does not run pharmacovigilance as an annual workshop, or validate a system once and walk away. Every capability carrying real consequence is built, applied, maintained and governed continuously. The Institute applies pharma’s own logic to AI.
Every module is recorded and taken when it suits the person. The live session is optional and recorded like everything else. There is no schedule to protect, no day to block and nothing anybody has to sit.
Hit a challenge or realise something is taking you a lot of unnecessary time, then ask the library a question, and AI takes you to the real answer recorded by a pharma AI expert. No AI answer is written for you. One is located for you. It finds the one Dr Bates already gave. AI does the search. You are taken to a real answer, on video, given by a named person.
Your team asks in their own words — how do I check a generated summary against its source, what has to go in the audit trail, has anyone asked about this vendor tool — and it returns the module, the point inside it and the timestamp. Including from the recorded live sessions, so an answer Dr Bates gave a regulatory team member one month is findable by another regulatory team member in another month.
They watch a few minutes, get their answer, and go back to work.
Which inverts how training normally gets used. Not watch this in the week we tell you to.
Hit a problem on a Tuesday, ask, get the ninety seconds that answer it. It matters most for the workflows that come round quarterly — the payer dossier, the periodic safety update, the submission twice a year — which are exactly the ones a team has half-forgotten by the time the next one lands, and exactly where AI saves the most time.
Two outcomes from Eularis programmes, both measured on a named workflow rather than on a feedback form.
A regulatory affairs team took a reporting cycle that had been running to three months and brought it down to two weeks.
An R&D group took a documentation process that had taken six weeks and completed it in under a day — checking included, using the verification checklist from Foundations, because the checking time is part of the work and we count it. Other time savings meant they could focus on increasing their pipeline without increasing headcount.
Both came out of live programmes with those teams, which is what the Institute was built from.
Neither is a guarantee and neither happens to a team that does not apply what it learns.
What they establish is the size of the prize on the workflows where AI is genuinely strong in each function: not a marginal efficiency on everything, but a step change on the long, document-heavy cycles that swallow whole quarters.
Across the people we train, the floor we see is around a day a week returned per person — roughly nine working weeks a year, each.
That is the floor rather than the ceiling, and it is the number to run against your own loaded rate.
You do not have to accept our arithmetic. Use your own headcount and your own loaded rate: the bar a licence has to clear is lower than most leaders expect, and it falls as the unit gets larger.
Since 2003 she has done one thing: applied AI to real pharmaceutical problems, inside real pharmaceutical and biotech companies. As founder of Eularis she has led hundreds of AI implementations across the pharmaceutical value chain — discovery, clinical, regulatory, market access and commercial — on six continents, for the majority of the world’s twenty largest pharmaceutical companies, across every major function.
She lectures at INSEAD, Fordham, Henley and Imperial, and numerous other postgraduate university programmes in AI in life sciences, and assesses AI startup grants for Innovate UK.
More than forty senior pharma executives have publicly endorsed her work on LinkedIn — not with “great session,” but with what changed in how their teams work.
Every module is built or directed by her. When the tools shift or the regulations move, she is the one rebuilding the curriculum, because she is the one pharma leaders ask when it happens.
See all 40+ endorsements on LinkedIn
In most pharmaceutical companies a purchase like this does not stall at the function head. It stalls at third-party risk, a month later, over questions nobody thought to ask early. So the answers are on the page.
The programme requires no access to your systems and no upload of your documents. There is no integration to approve and no connection to your environment unless you choose integration with your LMS. No module asks a team to put a company file, a dataset or a patient record into anything. Worked examples use our own material. Your team applies what it learns inside your own environment, under your own controls.
We do not hold your people. Your sponsor enrols their own team and holds the roster. Names and contact details are not needed to deliver the programme and are not visible to us.
Nothing your team submits trains a model. Not questions, not session content, not usage data.
The monthly session, and its limits. Questions are submitted in advance and stripped of company and personal identifiers before they are answered. Sessions are recorded into the library. Teams are asked to keep unpublished trial data, safety data, personal data and anything commercially confidential out of the shared room — the value is in the problem, and the problem travels perfectly well without them. Where a question genuinely needs your own material in the room, that is what private office hours are for, under your own agreements.
Documentation before you sign, not after. Our data-processing position, DPIA support and full procurement pack are sent when you ask for them.
One price for the population covered, never per seat. Three licences: Foundations on its own, a single function’s track with Foundations inside it, or every function at once. Annual term against a purchase order, no automatic renewal, ninety days’ notice before the date. Full procurement documentation the day you ask for it.
A single purchase order carries as many units as you want. Once several functions want their own track, the all-function licence covers every function rather than only the ones who asked, and often costs less than the tracks separately. Ask for that figure before adding tracks up.
Start with Foundations across the whole company and add a function track when one function’s work justifies it. Foundations credits in full against that track whenever you add it, so starting narrow costs you nothing later. A forty-person biotech and a business unit of forty pay the same, because the price is set by the population covered.
Foundations can be bought on its own for any population up to an entire company.
Rates are quoted on a call. Qualified enquiries receive a written terms sheet within twenty-four hours covering the rate, the term options and everything procurement will ask for, in a form you can forward to finance without editing.
Not a deadline. A number. Twenty per cent below standard rates for as long as the licence stays active — not an introductory year that resets. It follows you onto tracks added later and onto an all-function upgrade, so it holds as the programme spreads across your organisation. Founding partners also shape the build order, so your function’s track is written in conversation with your team rather than handed to them finished.
Section 5 of the white paper runs the three-year comparison, and section 7.3 covers the budget line most organisations are booking this against incorrectly. Ask for the one-page business case — written to be forwarded without editing.
Not one of the five?
Two-year and three-year terms sit five and ten per cent below standard, held for the term. Price protection rather than a discount — the rate you sign is the rate you keep. It does not stack with the founding partner rate; whichever is better applies.
Not a satisfaction score and not a percentage. A workflow, a before, an after, and whether the checking time was counted in the after. Most providers cannot answer it, because most measure the session rather than the work. Ours is further up this page.
If the answer is another session, you are being sold the expensive model in economical packaging. Look for a structure: a refresh cadence, applied practice on live work, a library organised by role, a governance checkpoint in the rhythm.
A provider with genuine role-specificity answers immediately, and differently, for each. A generic catalogue describes the same content twice and calls the difference a learning path. Ask for both curricula side by side. The gap between them is the whole answer, and it takes about a minute to see.
The first question a generic platform passes comfortably. The second and third are the ones it cannot.
Ours goes to you in writing after the call: the full curriculum for every function you are weighing, side by side, so you can do exactly what this section tells you to do to everybody else. It is not published here for the same reason yours is not published — it took two decades to build and our competitors read sales pages too.
Every selection counts toward the build order. The more colleagues who select a function, the sooner it is fully live.
No, and be sceptical of anyone who says otherwise. Compliance depends on your own systems, jurisdictions and circumstances, and is a matter for your legal and compliance functions. What a maintained, role-specific programme with per-person records gives you is a set of measures, taken continuously, and the evidence of having taken them. That is what an obligation of effort is discharged with. It is not a guarantee, and nobody can sell you one.
It is not an event, it is maintenance — and it is not built by a training company. A workshop starts eroding the day the room empties, partly in memory and partly because the tools and the rules move on. Eularis has spent twenty-three years doing AI strategy and builds inside pharma functions, and the Institute is what came out of being asked the same questions often enough. If you have already trained your team, the Institute is what protects that investment.
Then you are ahead of most of your peers, and you should protect it. Ask them two questions privately. How much of this are they doing on top of their actual job? And when they run a Q&A, what happens to the hardest questions? Not because they lack ability, but because they are enthusiasts rather than full-time specialists, and those questions are reliably the ones carrying the risk. The Institute sits underneath your network rather than replacing it.
Entirely. Modules are built around real workflows — payer dossiers, medical inquiry response, submission drafting, MLR-compatible content development, launch preparation — with prompt guides written for those exact tasks, some running to ninety pages for a single function. The full curriculum for your function is sent in writing after a call.
As much or as little as each person needs. Everything is recorded and taken on demand, including the monthly session, and nothing is mandatory or scheduled. The library also answers questions directly: someone asks in their own words and is taken to the module, the point inside it and the timestamp, including from past live sessions. So the realistic answer is that most of the time comes back later, in minutes rather than hours, at the moment somebody hits a real problem rather than in the week you told them to watch something.
It is the founding premise. New content every month as the landscape moves and as teams request it; the first live session of each quarter addresses what changed in the preceding ninety days; the curriculum is directed by Dr Bates personally.
That is the intention at both ends. The price is set by the population covered and quoted in bands, so a small team pays a small-team price rather than a fraction of an enterprise rate. There are no per-seat charges, so nobody has to decide who is worth including. The reason it is built this way is that Article 4 does not scale with headcount: a twenty-person biotech filing in Europe carries the same obligation as a global pharma, usually with nobody whose job it is to think about it. At the other end the same ladder simply keeps going, and adding functions or regions happens under the existing agreement rather than through a second procurement exercise.
That is the normal route and the pricing is built for it. Foundations credits in full against a function licence, and a function licence credits in full against an all-function licence, at any point in the term — so proving it on one function first costs you nothing when you widen it. One direction only, and it applies to the subscription rather than to the add-ons.
Headcount is set at the contract date and the price is held for the term, so a unit that grows across a band mid-term stays at the agreed rate and we adjust at renewal. Headcount is self-declared: the band follows the population you tell us you are covering.
Yes, and that is the model rather than a favour. When one top-20 company’s medical affairs team specified twelve modules they needed, all twelve were built and delivered inside a month. Requests from licensed teams go to the front of the build queue, and what one function needs usually turns out to be needed by others — which is how the library grows in the direction the industry is actually moving rather than the direction a content plan predicted.
Yes. There is an unedited Foundations lesson on this page, and Dr Bates will take your team through the platform live before any commitment — the curriculum for the functions you are considering, a full prompt library, and the ask-a-question search, on a call. What we do not do is open the library for self-service evaluation, for the same reason you would not send a competitor your SOPs.
Price follows the number of people the licence covers, in bands running from a single small team up to twenty thousand and beyond. A small biotech and a global function are on the same ladder, at different points on it. What matters is the population you want covered, not whether the company is large enough to qualify.
No separate regional rate – one price list worldwide, in US dollars, whichever country the licence is held in. . A function operating globally is simply a larger population on the same ladder, added under the same agreement without a second procurement exercise. Note that a US-based team whose work reaches an EMA submission, an EU affiliate or EU patients is in scope of the Act.
At the level of the problem, yes, and that is exactly what submitting questions in advance and anonymously is for. Not at the level of your data. Teams are asked to keep unpublished trial data, safety data, personal data and commercially confidential material out of the shared room. Where a question genuinely needs your own material, private office hours are the right setting and are available as an add-on.
The monthly session is deliberately cross-company and anonymous — that is where the industry-wide signal is. Teams wanting sessions of their own can add private office hours built around their specific workflows, quoted on request.
Yes. LMS delivery and SSO are available as add-ons, quoted on request.
No, and that is a real gap in most AI training. Generative tools are what a medical writer or a brand manager touches daily, and they are covered thoroughly. But many discovery and R&D and other functions run on methods that have nothing to do with language models and in several cases predate them by a decade, and a curriculum written by someone whose own AI experience starts with LLMs does not know they exist. Ask any provider when their AI work began. Check their Linkedin profile and see if they have only been at this a few years and you will have your answer.
Quizzes and certification, so the record shows understanding rather than completion. And independent vendor evaluations, planned for a later phase, covering the tools in use in the functions enrolled. When they arrive, Eularis will take no referral fees, revenue share or promotional arrangements from any vendor it evaluates. That independence is the entire basis of the value.
Pharma already knows how to build capability that lasts. It does it for pharmacovigilance, for regulatory affairs, for clinical operations — for every function where the cost of getting it wrong is consequential. The logic is identical for AI. The only thing missing has been the structure.
Write you name and email and enquiry and we will get right back to you as soon as we can.