EY-Parthenon just published something that should make every medtech executive uncomfortable. Their analysis of AI across the biopharma value chain found that 40-50% of task hours in operations could be automated with currently available AI technologies. Not future AI. Not theoretical AI. The stuff that exists right now.
The uncomfortable part isn’t the number. It’s the implication: most of that work is still being done by people who should be doing something else.
In biopharma, the conversation has moved past “should we use AI?” to “are we using it on the right things?” EY’s core argument is that efficiency gains from AI — cost savings, faster processing, fewer errors — are table stakes. The real value comes when those gains get redirected toward mission-critical objectives: better drugs, faster approvals, broader patient access. They call it moving from “do better” to “do differently.”
Medtech is having the wrong version of this conversation. Go to any medtech conference right now and count the “AI” demos. Most of them are chatbots for commercial teams — AI-powered CRM assistants, rep-facing copilots, smarter lead scoring. Fine products, all of them. But they’re optimizing the 38% of your workforce that’s already revenue-generating while the other 62% drowns in data entry, invoice reconciliation, and manual order processing. That’s like hiring a personal trainer for the marathon runner while the guy carrying 200 pounds of unnecessary gear just keeps walking.
Meanwhile, biopharma companies are pouring $4 billion a year into AI investments and venture capitalists put $3.2 billion into 135 AI-driven deals in the last year alone. The gap between where medtech is and where it could be isn’t a technology problem. It’s a prioritization problem.
Here’s a framework for closing it.
The fat and the muscle
EY-Parthenon introduces a useful mental model: target the fat, empower the muscle.
In life sciences companies, G&A and operations functions account for roughly 60% of total headcount. These are the departments doing data entry, report generation, invoice processing, compliance documentation, procurement workflows — the repetitive, rule-based work that AI handles well today. That’s the fat. Not because the people are expendable, but because the work is.
The muscle is your commercial team and your R&D function. These are the revenue generators, the relationship builders, the people making clinical and strategic decisions that drive the business forward. AI’s role here isn’t to automate — it’s to amplify. Help a sales team analyze data faster. Give a regulatory team better tools for submission preparation. Surface patterns a product development team would take months to find manually.
The distinction matters because most medtech companies are deploying AI backward. The industry’s AI energy is almost entirely focused on the commercial side — chatbots for reps, AI-assisted selling, conversational CRM. And look, a chatbot that helps a rep prep for a surgeon meeting is nice. But it’s not going to move SG&A from 34% of revenue to 28%. Roland Berger’s Global MedTech analysis put hard numbers on what that gap is worth: medtech winners average 28.6% SG&A versus 34.4% for underperformers, and roughly twice the market cap.[5] That spread is just under six points of revenue, and it lives almost entirely in the back office, not in the field.
There’s a deeper problem with the chatbot fixation, and Nora Kako at Fractional AI nailed it recently: the form factor itself is wrong. A chatbot makes a hard task conversational. But the goal isn’t conversation — it’s completion. When your ops team gets a PDF purchase order in their inbox, they don’t need to chat with an AI about it. They need the AI to read the PDF, extract the line items, match them against the contract, and push the data into the ERP. No back-and-forth. No typing prompts. The work just gets done.
Fractional’s case studies show this across industries — the best AI implementations are invisible. They pre-populate the form. They do the lookup. They surface the exceptions. The user shows up and the thinking has already happened. That’s the design principle medtech should be applying to its operations: not “how do we give our team an AI assistant to talk to?” but “how do we make the manual work disappear before anyone sits down?”
The EY data shows why this matters at scale: across the life sciences industry, operations and G&A have both the highest density of automatable tasks and the largest share of headcount. That’s where AI creates immediate, measurable ROI. A chatbot for your sales team is a nice-to-have. Automated order processing for your ops team is a P&L lever.
Where the hours actually go in medtech
The EY data breaks out medtech specifically by domain and FTE distribution, and the numbers are worth sitting with:
Commercial functions (sales, marketing, customer engagement) represent 38% of FTE headcount in medtech — the highest of any life sciences segment. That’s not surprising. Medical device companies are sales-driven businesses. Field reps, territory managers, clinical specialists — they are the commercial engine.
Operations (supply chain, manufacturing, quality) account for 22% of headcount. G&A takes another 30%. Research and development is 11%.
Now overlay the AI opportunity. EY found that in biopharma operations, 48% of tasks can be automated (AI replaces the work entirely) and another 34% can be augmented (AI helps the person do it better). For G&A, automation potential is even higher: 54% of tasks replaceable, 30% augmentable. Only 17-18% of work in these domains falls outside AI’s current reach.
Compare that to commercial functions, where 33% of tasks are automatable and 43% are augmentable. Or research, where 88% of AI use is augmentation — helping scientists think better — and only 8% is full automation.
The back office is where you automate. The front office is where you augment. And the place most medtech companies haven’t started is the back office.
The medtech-specific gap
This is where EY’s biopharma analysis needs translation for medical devices, because the operational reality is different.
Biopharma back offices process clinical trial data, regulatory filings, and manufacturing batch records. Medtech back offices process purchase orders, charge sheets, inventory transfers, and contract pricing. The data formats are different. The workflows are different. But the fundamental problem is the same: people spending their time converting unstructured information into structured data that systems can act on.
BCG’s Medtech Companies Must Move Faster on GenAI (2023) makes the structural case: GenAI is “uniquely suited to tasks that incorporate a high volume of unstructured data inputs” and to relieving “the burden of tasks that are highly repetitive but not fully automated because of workflow constraints.”[2] That is medtech back-office work to the letter — PDF purchase orders, photographed charge sheets, text-message inventory requests.
In medical device operations, the data problem is acute. Field reps send charge sheets as photos. Hospitals email PDF purchase orders. Inventory moves by text message. The information exists — it’s just trapped in formats no system can read without a human intermediary.
Salesforce’s 2026 State of Sales report (cross-industry, not medtech-specific) put a number on what that costs: reps spend 60% of their time on non-selling work, including 11% manually entering data.[3] The medtech version of that 60% lands in two places — on field reps after each case, and on the customer-service team typing what the reps send into the ERP. Every hour of it is an hour pulled away from work that compounds.
The math is straightforward. Your ops team types data into ERPs. Your reps chase missing information instead of selling. Your finance team reconciles pricing across 300 different customer contracts. None of this is muscle work. All of it is fat.
Automation vs. augmentation: a practical map
EY’s framework maps cleanly onto four operational domains in medtech. I’d apply it like this:
Billing and order processing — automate. PO intake, charge sheet extraction, invoice generation, pricing validation, three-way matching. These are high-volume, rule-based tasks where AI doesn’t need human judgment for the large majority of transactions. The exceptions — pricing discrepancies, missing lot numbers, ambiguous surgeon preferences — still need your team. But they’re handling exceptions, not processing every order from scratch. Deviceflow customers have cut billing error rates by 60% and pulled 15-plus days out of their collection cycle by closing the gap between charge sheets and invoices before a human touches them.[6]
Field inventory management — automate the tracking, augment the decisions. I’ve written before about why legacy tools fail on field inventory. Field teams that track consignment and trunk stock manually lose hours every week to phone calls and spreadsheets. As the same Tegus interview rep put it, the value of getting visibility is not having to “spend an hour calling or texting people and waiting for a response” when you need to locate inventory across a territory.[4] The bigger prize is what happens with the data once you have it: demand forecasting, consignment optimization, expiration management.
Supply chain coordination — automate the routine, augment the planning. The automation opportunity is in order processing, inventory replenishment, and demand signaling — the high-volume routine work. The augmentation opportunity is in scenario planning, risk management, and supplier diversification — where human judgment compounds. Most medtech companies haven’t separated the two.
Commercial operations — augment, don’t automate. This is where EY’s “empower the muscle” principle applies. Your reps’ relationships, clinical knowledge, and institutional trust aren’t automatable — and shouldn’t be. But the Salesforce 2026 data referenced above shows reps already spend 60% of their time on non-selling work. The right play is removing the administrative drag so reps can spend that time on what they do well: being in the OR, building surgeon relationships, winning competitive cases. More reps isn’t the answer. More selling time per rep is. A 20-year ortho rep — someone who’d carried J&J, Stryker, and SI-BONE lines — put the diagnosis bluntly in a 2024 Tegus expert interview: “The Achilles heel of it is you’re purely reliant on the rep to do that, to input that.”[4] When a commercial system depends on a busy rep manually entering data between cases, the data doesn’t show up and the rep learns to dread the system.
The organizational design question
EY makes a point that most AI discussions skip: deploying AI isn’t just a technology project. It’s an organizational design project.
Their recommendation is to redesign the organization for AI from the start — not bolt AI onto existing structures. For large companies, that means identifying duplicative functions and high-volume routine tasks across departments, then mapping which processes AI can handle and redeploying people to more strategic work. For smaller companies, it means structuring for the future from day one: building AI into core workflows rather than layering it on later.
This resonates in medtech specifically because of how operations teams are built. A typical mid-size device company has 4-8 people in customer service, processing emails and entering orders. They have 2-3 people in billing, reconciling POs and generating invoices. They have inventory coordinators tracking spreadsheets. They have compliance staff managing documentation.
These aren’t individual automation projects. They’re one interconnected workflow: an order comes in (unstructured), gets processed (manually), triggers inventory movement (tracked in spreadsheets), generates an invoice (manually), and gets reconciled against a contract (manually). Automating just one step creates a bottleneck at the next one.
The organizational implication: you need to automate the workflow, not the task. And that requires rethinking how teams are structured around that workflow.
What EY gets right — and what they miss
EY’s analysis is strong on the macro framework. The “do better vs. do differently” distinction is useful — maybe the most useful framing I’ve seen for this problem. The fat-vs-muscle model is the right way to think about AI allocation. And their seven strategic recommendations — align AI with mission, redesign org structures, adopt an “AI-augmented everything” mindset, invest in reskilling, embrace ecosystem collaboration, champion accountability, execute change management — are sound.
What they miss is the implementation reality for companies below the Fortune 500 line.
Their recommendation to “benchmark against peers and competitors to identify gaps” assumes you have a peer group deploying AI at scale. In medtech, that peer group barely exists. Their recommendation to “invest in large-scale training programs in data science, AI ethics, or digital literacy” assumes you have the headcount and budget for a training program. Most device companies I talk to have a team of 6-15 people running all of operations.
The practical translation for medtech:
Start with the workflow that costs you the most hours per week — probably order intake and processing. Automate the data extraction and entry. Measure the hours recovered. Redeploy those hours to the exceptions and investigations that actually need human judgment. Then expand to the next workflow.
Don’t build an AI strategy deck. Build an AI-automated workflow.
The compounding problem
The thing that worries me most about medtech’s slow AI adoption: the gap compounds.
Companies that automate operations early build data infrastructure as a byproduct. Every automated order creates structured data. Every tracked inventory movement creates a signal. Every processed charge sheet creates a record. Over time, that data becomes the foundation for the augmentation layer — the forecasting, the pattern recognition, the strategic insights that EY frames as “do differently.”
Companies that wait don’t just fall behind on efficiency. They fall behind on the data that makes future AI possible. If you’re not capturing structured data today, the augmentation tools you buy tomorrow will be running on the same chaotic substrate — and underperforming accordingly.
Biopharma figured this out earlier than medtech did. The investment gap between the two industries on AI deployment is widening, not closing.
What this means for the next 3-5 years
EY’s vision of the “AI-enhanced organization” is directionally correct: smaller companies use AI to scale without scaling headcount, and larger companies use AI to absorb workload growth without proportional hiring. Both achieve an optimized operating model where AI handles the routine and people handle the strategic.
For medtech specifically, I’d frame the trajectory differently.
In the next 12-18 months, the companies that automate their commercial operations back office — order processing, billing, inventory tracking — will see immediate ROI in reduced headcount costs, faster cash collection, and fewer errors. These are two-way doors: you can always go back to manual processing if the automation doesn’t work. Low risk, high signal.
In 18-36 months, those early movers will have enough structured data to deploy augmentation tools for their commercial teams — demand forecasting, territory optimization, competitive intelligence. The companies that waited will be starting where the early movers were two years ago.
By 2029-2030, the gap will be structural. Companies with AI-automated operations will operate at fundamentally different cost structures. EY’s data suggests G&A functions could shrink significantly while operations teams shift from processing to oversight. The companies still running manual processes will face a simple math problem: their cost-per-case will be 2-3x higher than their automated competitors.
Supply chain complexity isn’t going away — every commercial leader I talk to puts it near the top of their operational pain list. The only question is whether you manage it with people or with systems.
The uncomfortable question
EY frames AI adoption as moving from efficiency to mission. In biopharma, the mission is drug discovery and patient access. In medtech, the mission is getting the right device to the right patient at the right time.
The uncomfortable question for medtech leaders: how much of your team’s time is actually spent on that mission versus the administrative work that surrounds it?
If a surgeon needs a specific implant configuration for Tuesday’s case, the value chain involves a rep confirming the order, an ops team processing it, a warehouse shipping it, a finance team invoicing it, and a compliance team documenting it. The mission-critical part takes minutes. The surrounding administrative work takes days.
That’s the fat. And until medtech gets serious about targeting it, we’re paying six-figure salaries for data entry — and wondering why we can’t grow faster.
Sources
- EY-Parthenon, How AI in Biopharma Can Drive Mission-Focused Growth (2026). Analysis of AI utilization across the life sciences value chain — fat-vs-muscle framing and the 40–50% operations automation potential. Citation taken from the published EY analysis; direct PDF verification pending.
- Boston Consulting Group, Medtech Companies Must Move Faster on GenAI (2023), p. 3 — GenAI is “uniquely suited to tasks that incorporate a high volume of unstructured data inputs” and to relieving “the burden of tasks that are highly repetitive but not fully automated because of workflow constraints.”
- Salesforce, State of Sales, 7th Edition (2026), p. 8 — Reps spend 60% of time on non-selling work, 11% on manually entering data. Cross-industry survey, not medtech-specific.
- Tegus expert interview, Territory Account Manager, 2024, pp. 5–7.
- Roland Berger, Global MedTech: How to Succeed in Uncertain Times (2022), p. 2 (figures echoed in Roland Berger’s 2024 Future of MedTech: From Growth to Profit) — Winners average 28.6% SG&A versus 34.4% for underperformers; winners average roughly twice the market capitalization.
- Deviceflow customer outcomes — Medical Ventures case study and internal customer metrics (2025–2026).