AI and automation in medical device marketing: what to automate now, and what comes next

AI is about to reshape how medical device companies build and run their marketing, but almost nobody has worked out which parts to automate, in what order, or how to prove any of it is working.

If you're a medical device marketer, you're under pressure to move faster, do more with the same team, and show a return, all while the tools change underneath you every few weeks. This article is our honest view of where marketing automation in medical devices actually is right now, what you can sensibly automate today, and what should stay human for a long time yet. We work downstream, so that's where we'll focus: awareness, education, demand capture, and conversion.

Here's what we'll cover:

What we actually mean by a marketing tech stack
Where things stand now: heavy investment, thin proof
What probably won't be handed to AI any time soon
What you can automate, from on-brand assets to instant follow-up
Why your SOPs, not your software, become the bottleneck
The limiters that will decide who wins: attribution, security, compliance, cost, and people
How to get ahead without losing trust

A quick note on transparency: Podymos is a HubSpot partner agency, so HubSpot comes up a few times below. We've kept this even-handed and we mention other tools throughout, but we wanted you to know where we stand.

What is a marketing tech stack?

A marketing tech stack is the set of software tools you use to run your marketing, plus the connections that let those tools work together.

That usually includes a CRM to hold your contacts and pipeline, a content management system for your website, email and social tools, analytics and reporting, creative tools for producing materials. Some of these can be separate tools or combined into one. As AI gets better at pulling disparate data together, you might one day build a stronger stack by deliberately splitting tools rather than combining them. That's a thought to prove out in practice rather than a rule today, but it's the kind of new option automation may open up.

The stack isn't just the tools, though. It's also the wiring between them: the automations and data flows that mean an action in one place can trigger something in another.

This matters because AI doesn't replace your stack. It sits on top of it and runs across it. The question for the next few years isn't whether to use AI, it's which parts of your stack you connect it to, and how safely you do that. Done well, AI takes out repetition and human error, speeds up the whole process, and ideally gives you a clear picture of the ROI for each of your campaigns, so you can target your budget smarter and faster.

Where we are now: heavy investment, thin proof

Most companies are investing heavily in AI and can show very little for it that they can actually prove.

MIT's 2025 report, The GenAI Divide: State of AI in Business 2025, found that despite $30 to $40 billion invested, around 95% of enterprise AI initiatives delivered no measurable impact on profit, while only about 5% achieved real revenue acceleration. A more recent global CMO survey from Comviva found that while roughly 90% of organisations are increasing their AI marketing investment, only about 16% of marketing leaders feel confident defending that spend with clear business evidence, and 58% blame the difficulty of revenue attribution, because AI touches so many points in the journey that its contribution is hard to isolate.

This is the thread that runs through everything below. The problem usually isn't whether AI can do something. It's whether you can prove it worked, and whether it's wired into the rest of your business properly. MIT also found that buying from specialist vendors and building partnerships succeeded far more often than building everything in-house, which is worth remembering before you commission a bespoke system.

Messaging and brand won't be handed to AI anytime soon

The two things least likely to be automated any time soon are also the two that matter most: your messaging and your brand.

Get your messaging wrong and everything downstream costs more, takes longer, and is less effective. It's like trying to sail a ship with the anchor still down. Every ad, every campaign, every sales conversation works harder than it needs to, and that's where marketing costs quietly escalate. Messaging carries a deeply human element: the empathy to understand what a surgeon, a procurement lead, or a patient actually feels, and the judgement to know what to say to them. That's something machine learning doesn't have in the same way.

The same is true of brand. AI can help you produce variations and move faster, but the creative core, the read on your audience, and the decisions about what your brand stands for stay human, and that's where a lot of the value will concentrate.

Here's the useful part, though. Once your messaging and brand guidelines are clear, you've created exactly the inputs that AI needs to do good work downstream.

On-brand content and assets at scale

Once your messaging and brand are set, AI will increasingly be able to produce a consistent suite of materials from them: ads, social posts, email, exhibition graphics, and even a first draft of a website.

The prize here is consistency, speed to market, and the ability to change things quickly. From one set of brand guidelines, you could generate copy, images, and graphics that hold together across every channel.

This is the direction Adobe is moving in, and it's worth watching. Adobe GenStudio for Performance Marketing lets teams upload brand guidelines, templates, and pre-approved assets, then generate on-brand content across email, paid media, and social, with brand governance and approval workflows built in. Two of its features matter especially for a regulated field. You can lock disclaimers and citations into every variant word-for-word while the rest of the content stays flexible. And you can sign assets with content credentials, so each one carries a traceable history. In April 2026 Adobe went further and announced Adobe Brand Intelligence, which it describes as a continuously learning engine that moves teams beyond static brand guidelines into a system that learns from review feedback, approvals, and rejections.

The bottleneck shifts from creation to approval

When AI can draft materials in minutes, the slowest part of your process becomes approval, not creation, and that's where the real advantage will be won.

In medical devices, content moves through regulatory, medical, and legal review in weeks or months. If you can suddenly create in minutes but still approve in months, you've gained very little. The companies that pull ahead will be the ones that rewrite their standard operating procedures, the SOPs that govern how materials get reviewed and signed off, so that compliant materials can move through in days.

That's not just a rules change, it's a systems change. It means presenting materials to your reviewers in a consistent, structured way every single time, with the relevant claims and references attached, so review becomes faster and more predictable rather than a fresh negotiation on every asset. Get this right and you unlock something bigger: the ability to publish a steady stream of compliant content, which is exactly what the next stage demands.

Awareness and the shift to search everywhere optimisation

Being found is no longer about ranking on Google. It's about being present and cited everywhere your buyers, and the AI tools they use, are looking.

The industry term for this is search everywhere optimisation. It pulls together traditional SEO with two newer disciplines: answer engine optimisation (AEO), which is about getting your content extracted and cited in AI answers, and generative engine optimisation (GEO), which is about how often and how accurately AI systems represent your brand. By some measures, around 93% of AI-driven searches now end without a click, so visibility increasingly means being included in the answer, not just ranking in a list of links.

AI tools learn about you from across the web, drawing heavily on sources like Reddit, YouTube, review sites, industry publications, and clinical literature. Reddit's prominence in AI answers has been a high-profile example of this recently. So credibility now has to be broad and authentic: original content on your own site, yes, but also video, social, a presence in the places your audience gathers, and new clinical publications where you can support them. You're also writing for three readers at once now: the human, the search engine, and the large language model, plus the crawlers that need to move across your site cleanly. Each weighs different signals, and each of the major AI tools behaves slightly differently.

This is where GEO becomes a discipline of its own, and it's going to be a huge area. If AI tools are describing your brand to buyers, you need a way to listen to how you're being represented, and a plan for what to do when it's wrong. That's something the industry will have to develop quickly, with close ties to the AI models themselves. There's a sharp regulatory edge to it, too. As AI tools travel across every channel, including places like Reddit, anything your team says in public can be picked up, repeated, or surfaced in an answer. That makes it more important than ever that everyone is clear on what they can and can't say about your products, because a loose claim in a forum could resurface inside an AI answer, and that's a real regulatory risk.

This is not optional, because the decision is being made before anyone contacts you. The 6sense 2025 Buyer Experience Report found that 95% of the time the winning vendor is already on the buyer's Day One shortlist, that the pre-contact favourite goes on to win around 80% of deals, and that buyers have defined 83% of their requirements before they ever speak to your sales team. Gartner found that 67% of B2B buyers now prefer a buying experience with no sales contact at all. Waiting for someone to call your sales team is no longer a strategy.

Capturing and converting demand: instant follow-up

When someone fills in a form or downloads a guide, the speed and quality of your follow-up is about to become almost entirely automated.

This is where a lot of the near-term automation will land, and it depends on having a capable CRM underneath everything, something like HubSpot or Salesforce. If you're weighing up which to build on, we've covered this in detail in our guide to choosing the right CRM for your medical device company.

Here's a realistic example. Someone views two of your product pages and downloads a guide. That can automatically trigger a personalised email and a LinkedIn message, or a voice agent that calls to understand how you might help. A voice agent is an AI system that can hold a spoken phone conversation in real time, understanding what the person says and responding naturally, rather than reading from a fixed script or menu. The effect is that by the time a lead reaches a human, it's already qualified, which means your sales team spends its time on the conversations most likely to close. You can use the same systems to keep educating that buyer afterwards, sending the next piece of information they need without anyone lifting a finger.

There's a genuine dilemma forming here, though, and it's worth keeping an eye on. Buyers are increasingly ignoring gated content; one analysis found that up to 92% of gated offers now go ignored. If people won't hand over their details, and they expect to find everything through AI anyway, how do you capture contact information at all? There's no settled answer yet. The likely shape of it is that more of your discovery becomes anonymous and AI-mediated, and capture leans on fewer, higher-value moments. The practical advice for now is to watch closely how your audience engages, where they are, and what they're actually willing to do.

The next frontier: cold outreach, voice agents, and hyper-personalisation

Cold outreach could become fully automated once AI voice agents are genuinely good at it, and the companies that win will be the ones that stay highly specific.

The pattern looks like this. AI does the background research to work out whether a company is a real fit, in other words, whether you actually solve a problem they have, with that research carefully curated against your messaging and what you sell. It then briefs a voice agent that can make the call. Picture a product launch, or a meaningful update to an existing device. AI could pick out the specific accounts that have the exact problem your launch solves, and have a voice agent reach out to them. But this only works if your messaging is genuinely perfect and properly tested, and if the call clearly adds value. Get it wrong and you'll annoy the very people you most want to reach, and burn them early. Done in a sensible, measured way, it can be highly effective.

The risk is obvious: this will be overused, and generic, irrelevant automation will be ignored and resented, especially given how many buyers already want to avoid sales contact. That raises an interesting question. If buyers are trying to avoid sales contact, will they see an AI voice agent as exactly that, a salesperson without the emotional read, and bristle at it? Or will they accept it more easily, treating it more like a voice advertisement or a voice note than a person trying to sell to them? Nobody knows yet, and the answer will shape how far this goes.

Specificity is the whole game. If you're genuinely speaking to a real problem for that exact buyer, it will likely work very well. If you're not, it's unwanted noise. Push this further and you reach hyper-personalisation: reading daily signals and triggers to work out where each buyer is on their journey and what they need next. That's data-hungry, and it's likely to live in custom-built apps running quietly in the background, which is where tokens start to matter. Tokens are the units of text that AI models process, and they're what you're billed on. Pull enough data through enough models often enough and token consumption climbs fast, which brings us to the harder part of all this.

The limiters that will decide who wins

The drivers of all this are speed, consistency, and personalisation. The limiters are harder, and they're what will separate the companies that win from the ones that simply spend money.

Proving the return: revenue attribution

This is the big one, and it's why so much AI investment looks unconvincing. As we saw earlier, most organisations can't yet prove their AI marketing is generating returns, largely because AI influences so many touchpoints that isolating its effect is genuinely difficult. A strong CRM sitting underneath your AI helps, because it keeps your data in one place, but reporting may need to sit alongside or above it rather than inside it.

Security and the connectors you choose

For any of this to work, AI has to connect to your stack, and every connection widens both what the AI can do and what could go wrong. There are three common routes. MCP, or Model Context Protocol, is an open standard for connecting AI to tools and data, and it's quickly becoming the default. Beyond it sit direct API connections and command line (CLI) access, which can reach further into your systems. The further the reach, the bigger the risk: over-broad permissions, exposed credentials, data leaving your environment, or an agent taking an action you never intended. There's also a practical question of whether a platform will even allow the connection, as LinkedIn and others restrict what can plug into them. The principle to hold onto is least privilege: give each connection the narrowest access it needs, and require human approval for anything irreversible.

Regulatory compliance at speed

Speed is only useful if what you produce is compliant. Any tool you use has to understand the rules, or they're worse than useless, churning out irrelevant or unusable content. For us, that means EU MDR, FDA requirements, and the specific claims approved for each product.

There's now another layer to keep an eye on: the EU AI Act, the first comprehensive law of its kind. It came into force in August 2024 and is being phased in. The dates have been shifting, so treat the following as the current direction rather than the final word.

The headlines for marketers are reasonably simple. From 2 August 2026, you'll need to tell people when they're interacting with AI, such as a chatbot. The labelling of AI-generated content is expected from 2 December 2026, later this year, with a short grace period for systems already in use before then.

The cost of tokens

As personalisation deepens and more data flows through your models, token costs climb, and they can scale out of control if nobody is watching. Rationalising and optimising token use, deciding what genuinely needs a model and what doesn't, becomes a discipline in its own right rather than an afterthought. This is also where the right help pays for itself: suppliers and consultants who specialise in this area can add real value, because using tokens efficiently is a skill in its own right.

Training your team to use it all

You can build the perfect tech stack and still get nothing from it if your team can't use it well. This is one of the biggest limiters of all, and most people know it's there. Before you roll anything out, benchmark where your team actually is, so you understand what they can handle today and where the real gaps sit. Then move them forward at a pace that matches that starting point, because pushing too fast is how things break and how security slips.

Training has to be ongoing, because there's so much to learn in AI and it keeps changing. The model that works best is a simple one: learn it, try it, optimise it, then deal with any issues that come up, before you take on the next challenge. Step by step, your team builds real proficiency. Pair all of this with the human-in-the-loop checks below, so people are building confidence on workflows that are still being supervised.

How to get ahead without losing trust

The way to get ahead is to decide your stack deliberately, connect it securely, and keep a human in the loop until you've earned the confidence to step back a little.

Keeping a human in the loop means a person reviews and approves what the AI produces or does before it goes out. Start with humans checking at many points. As you build genuine confidence in a given workflow, you can reduce the number of checkpoints, but in a regulated field you'll likely never be able to remove them completely. Do this in a way that you're comfortable with, and accept that the human in the loop is going to matter for a long time, arguably forever.

The bigger strategic task is defining your stack: which software you'll use, how those tools connect, and what rules govern the way they pass work between them to deliver the outcome you want. All of that has to happen on a secure network so that none of your customer information leaks. Think hard about where your data is stored, which matters more in the EU, and about the companies you partner with. Are they certified to the right standards? Look for ISO 27001 for information security, the newer ISO 42001 for AI management systems, published in 2023, and SOC 2.

The opportunities here are real and the pace is genuinely exciting. The companies that fall down will be the ones that rush to do everything at once, or that don't think deeply enough about what they're actually trying to achieve. Doing it deliberately, and in a way that builds trust, is the whole point.

Useful resources

A few related reads from our Learning Centre that go deeper on the topics above:

The bottom line

The winners won't be the companies that automate the most. They'll be the ones that automate the right things, in the right order, with proof and trust built in.

Keep your messaging and brand human, because that's where your advantage lives. Automate the things that genuinely benefit from speed and consistency: on-brand asset creation and instant, intelligent follow-up. Rewrite your SOPs so compliant content can move in days rather than months. Build a broad, credible presence so AI tools cite you, because the decision is being made before anyone calls. And treat attribution, security, compliance, token cost, and your team's readiness to use it all as the real limiters, not afterthoughts. Get those right, keep a human in the loop, and you'll be ahead of almost everyone.

If you'd like to talk through what a sensible, secure automation roadmap looks like for your device, book a call with our team. We'll give you a clear, honest view of where to start.