Starting in October, the top management at my former employer became almost impossible for us employees to reach. They were completely absorbed in budgeting for the following year. The result reflected the balance of power more than actual needs.
In today’s Executive Corner, you’ll find tips for planning your 2027 budget—and making use of any remaining budget for 2026.
Regulatory Updates
From a regulatory perspective, it has been a quiet week, and I’d like to report on two planning activities.
a. EU
The relevant committees are currently revising the list of standards harmonized under the MDR. For many standards, the deadline for adoption is to be extended; a few are slated for removal, and some are to be added, including some in the dental sector.
I’ll provide more details once the list has been adopted.
b. USA/FDA
The FDA has published its plans for guidance documents. Taken together, the A-List and B-List contain 13 guidance documents scheduled for completion in draft or final form in 2027.
Six of these documents address high-tech topics such as software, cybersecurity, robotics, and AI. The others concern the regulatory system and specific products such as air-purifying respirators.
It makes sense that the FDA wants to keep pace with rapidly evolving technologies in its regulatory guidance, too.
If you are planning to develop products in these areas, you should keep a particularly close eye on the final guidance on AI-enabled Device Software Functions and the draft guidance on Post-Market Cybersecurity.
Budget planning for 2027 is underway. Don’t budget for more tokens—budget for better processes. Companies that fail to invest will fall behind. Those that invest wisely and proceed thoughtfully will need less money than they might expect and can achieve their first results as early as 2026.
Why You Should Invest in AI
Competitiveness: Development cycles need to shrink, while the number of product variants is expected to grow. Those who can achieve market authorization faster and at lower cost will win. Almost impossible without AI.
Regulatory requirements: The number of regulatory requirements is growing, and the pace at which they change is accelerating.
Without AI, this would require more staff—people you may struggle to find and may not be able or willing to afford.
Competence: AI competence doesn’t emerge just because someone makes a decision. It develops through practice. The sooner you build this competence—which is also required by the AI Act and ISO 13485—the sooner you can increase your company’s productivity.
A clear signal: An AI budget shows your team that you are serious. Combined with an AI policy and agreed objectives, it encourages everyone to finally get started.
What You Should Budget For
Let me start with this: I consider “token maximizing” a donation program for US tech giants.
How much you should budget for each area depends on where you currently stand. Here is my proposal:
Getting started: "We begin"
Building up: "We broaden"
Scaling: "We embed"
Total per employee per year
~1.000 €
~2.500 €
~5.000 €
Usage (tokens, agents, hardware if applicable)
20%
45%
65%
Enablement and external components
35%
20%
10%
Roles and working time
10%
15%
10%
Compliance
15%
10%
10%
Access (License)
20%
10%
5%
Notes on the table:
The figures are averaged across all employees, not just those who do—or are expected to—use AI.
Although the percentage allocated to enablement decreases as you progress, the absolute amount does not. €350 is enough for one in three employees to attend a one-day seminar costing €1,000.
Most compliance costs are one-off expenses, for example, validating tools, establishing an AI policy, or negotiating a company agreement.
Dedicated hardware is currently worthwhile only if utilization is high or if you do not trust providers’ contractual assurances regarding IP protection. Otherwise, I would wait. Models are becoming more efficient and computers more powerful, which should make local deployment more affordable in the future.
How to Manage Your Budget Planning
Choose Your Approach Deliberately
There are several ways to proceed:
Top-down: This requires a clear AI strategy, which is often lacking.
Bottom-up: This would be ideal. But if there were enough grassroots initiatives, more companies would already have progressed beyond their initial experiments.
Delegate with accountability: Top management sets the framework—high-level objectives, policy, and budget. Each department, such as RA, QM, or Development, must test use cases. These activities then become part of the objectives agreed with department heads and their teams.
A personal note: I would prefer curiosity and interest in improving one’s own work to be motivation enough. But bonuses help.
Measure Processes, Not Tokens
It is difficult for me to estimate the opportunity costs your AI experiments may be generating. So keep an eye on your process KPIs. These should improve—and, of course, they should quantify a genuine contribution to your company’s value. Examples include:
Time from the end of development to approval of the Technical Documentation
Number of review cycles
Number and type of request for additional information from the Notified Body
CAPA turnaround times
Increase the Budget in Stages
Budgets are often spent faster than targets are achieved. I would therefore release only around 40% of the planned monthly expenditure shown in the table above during the first quarter. However, increase the budget as soon as the success criteria have been met. And set aside a reserve of 25–30%.
Manage Budgeting Like Any Other Process
You know the PDCA cycle: plan, do, check, act. It applies to your budgeting process, too: plan in 2026, implement and measure in 2027, and improve in 2028. If you would rather not wait that long, use your remaining 2026 budget to get some practice in.
If You Still Have Budget Left for 2026
Choose a process and a KPI that you want to uptimize using AI. Using ready-made components rather than developing everything from scratch will get you there faster. That is how participants in our "Zero to Hero" program use our proven intellectual property, including prompts and formalized review criteria.
Contact us if you’d like to accelerate your progress this year using our IP.
Perhaps you know the feeling: I use AI to help me write texts, only to find myself thinking, “I would never write like that.” Sentence fragments such as “The review carries,”“One cause runs through everything,” or “The deviation is caught.” Shudder!
So should I just go back to writing everything myself? Often, yes. But you can avoid the worst excesses by setting clear rules in a system prompt or a skill. For example:
“Every sentence must describe an action or an observation: who does what to which object, or what the reader can see as a result.”
"Prefer active, concrete, and specific verbs."
“No exaggerated enthusiasm or marketing speak.”
“No abstract metaphors, clichés, or empty phrases.”
“Keep content short, concrete, and specific. When in doubt, leave it out.”
It helps a lot to add your examples to each rule.
b. Deciding Whether to Switch to a New Model
I find it challenging to keep up with every new model version and type. And we’re only at the beginning: updates will become more frequent, and the range of specialized models will continue to expand.
As a result, we will increasingly have to decide whether and when it makes sense to switch models. After all, every change comes with costs, such as revalidation.
It may sound obvious, but a switch is only justified if you can demonstrate a benefit or if the existing model is no longer available.
The following steps can help you assess whether switching is worthwhile:
Define the intended purpose of your AI system (for example, minimizing the risk of a Notified Body significantly delaying market authorization) and estimate the value of achieving that benefit.
Derive quantitative criteria from the intended purpose that the AI system must meet. For example, for sensitivity: “At least x% of all nonconformities must be detected.”
Specify test cases, create acceptance criteria, and retain the test data. This will save you most of the effort the next time you switch models.
Run the tests to determine the performance values for both the old and the new model.
Based on the results, decide whether to switch to the new model and document your decision.
With kind regards,
Christian
Founder Johner Institute
Upcoming Seminars
Certified Professional for Medical Software (CPMS)
October 26-29, 2026 (4 days!)
Get prepared to meet the regulatory requirements for your medical software!
The seminar covers the relevant legal requirements and standards for medical software. Practical exercises and valuable tips will help you quickly and easily implement your new knowledge in your company.
Our top experts will provide you with this expertise directly. They have helped countless software manufacturers to meet the requirements, and are also experienced auditors.
IEC 60601-1 Safety of Medical Electrical Equipment
November 10-11, 2026
Learn how to successfully implement the requirements of the IEC 60601 family of standards in practice and ensure standard-compliant documentation.
In this seminar, you will get all the knowledge to speak to investigators and auditors on an equal footing. You will develop a comprehensive understanding of the basic concepts of the IEC 60601 family of standards, and learn how to successfully implement the relevant requirements in practice using various exercises and how to create standard-compliant documentation.
