Tool price ≠ total cost
The licence is only the tip of the iceberg.
insights · ai strategy for mid-sized companies
When does AI pay off for mid-sized companies?
Use case discovery meets TCO model: AI can support business processes and save money doing so. Beyond programming, every company has countless fields where AI can add real value. What is often missing is a way to show clearly whether an introduction pays off — and a technology for implementation.
This page answers five questions: the biggest hurdles to adoption, what it takes to make it worthwhile, which costs really arise, how regulation affects costs — and what a reliable ROI view looks like.
practical example
A (simple) digital business process
The complaints assistant receives customer enquiries and steers the first internal activities — an end-to-end process in eight steps:
- 01 · MS Outlook
Cyclical polling of the mailbox
- 02 · AI analysis
Analysis of the incoming customer email
- 03 · Database check
Stakeholders? Has this complaint occurred before? Same component? Same customer?
- 04 · Calendar lookup
Check stakeholder availability
- 05 · MS Teams
AI-supported creation of a briefing
- 06 · Calendar entry
Appointment for stakeholders with briefing
- 07 · MS Outlook
AI-based draft of the outgoing email
- Result
First response and internal coordination run without manual intervention
2–3 h
manual
32 s
automated, AI-supported
4.074
Tokens per run, 1–2 / week
Sovereignly hosted
AI model, GDPR-compliant
the three main hurdles
The biggest hurdles
The question is no longer whether, but when AI pays off. The bottleneck is rarely the technology — it is selecting the right use cases and enough of them, embedded in the company-specific framework.
Regulatory requirements
EU AI Act, GDPR and specific frameworks shape the overall picture.
Measurable revenue gains are missing
Initiatives rarely show a measurable increase in revenue.
criteria
Four criteria for the “does it pay off” question
Rule of thumb: only when all four criteria come together does an idea become a viable business case.
(Many) use cases
Is the use case clear, the underlying processes described and modelled in BPMN 2.0 and DMN? Where do the processes sit on the map — management, core, support?
Structured data
Is usable data available and accessible? Which data is processed — and what requirements does that place on the AI model?
Volume & frequency
High numbers and repetition create economies of scale. Token volume matters too.
Economics
Does the benefit exceed the TCO over five years — where is the ROI?
approach
From use case to automation forecast
Every activity is assessed: rule-based → RPA · knowledge-based → AI agent · decision-critical → human approval. The result is a forecast instead of a gut feeling — AI-supported and experience-based, with resistance reduced.
- 01
Find use cases
Use Case Discovery describes the process in detail
- 02
BPMN 2.0 modelling
Model the process cleanly to the ISO standard incl. frequency and duration
- 03
Detect automation potential
What can be automated (RPA) and supported by AI
- 04
Forecast token volume
Determine the degree of automation from volume and potential
- 05
Transfer into the TCO model
Decision based on numbers
economics: tco instead of licence price
The tool price is only the tip of the iceberg
Viable use cases, an intelligent tech stack, targeted data preparation, interfaces, training, securing ongoing operation and governance decide economic success — not the tool. The cheapest entry is rarely the most economical solution.
Licence / tool price — visible
- Data preparation & access
- Interfaces & integration
- Training & adoption
- Operation — model calls, quality gates
- Governance & security
The TCO model: making acquisition and operating costs transparent
Rank use cases, compare deployment options — on premise, private cloud, cloud, hybrid, co-pilot, GPUaaS, LLMaaS — lease, buy or already available. Monetary parameters: model size, number of seats, PUE, electricity price, hardware lifecycle, inflation, staff costs. More on the page about the Sellium Method.
risk & compliance
Regulation as a cost driver
Regulations largely determine how demanding AI use becomes in ongoing operation. They are not a one-off item — they cause ongoing effort and feed directly into the TCO forecast.
Risk & compliance
EU AI Act
Classify use cases, document technically, prove conformity where required. More on the EU AI Act
Data protection
GDPR
Clarify impact assessment, legal bases and data processing agreements.
Data protection
Cloud Act
Conflicts with the GDPR. Legal jurisdiction over the provider decides.
Industry frameworks
e.g. TISAX
OEMs often demand EU jurisdiction; prototype protection demands seamless access control. ISO/IEC 42001 also plays a role.
the path to roi
Five steps to a reliable ROI view
Step 1: Use cases
Cleanly described, in large numbers.
Step 2: AI readiness & IT check-up
A detailed picture of the company.
Step 3: Token volume forecast
Determine the degree of automation from volume and potential.
Step 4: TCO model
Processes the input data and calculates ongoing costs.
Step 5: ROI
Decision based on numbers.
the result — an example from practice
845.000 €
Savings p.a.
1.185.000 €
Delivery effort for all use cases
1.5 years
Payback
70 %
of the benefit comes from 5 specific use cases
Preference in this case: a hybrid approach.
take-aways
The key take-aways
- 01
Use cases should be available in large numbers — cleanly documented.
- 02
The more use cases run, the greater the leverage.
- 03
Token volume and the required model quality matter too.
- 04
(AI-supported) business process automation does not need large infrastructure.
- 05
Self-hosted AI often beats co-pilot on cost and compliance.
Further reading: The Sellium Method · EU AI Act · Technical embedding of AI · Funding options
contact
Let's talk
Happy to help with anything AI. We will get back to you promptly.
Get in touch
Thank you for your message.
We will get back to you shortly. If it is urgent, you can reach us at +49 371 524 99 140 or contact@sellium.ai.