SMD – Position on Artificial Intelligence

AI supports.
Expertise decides.

How SMD Group uses artificial intelligence in search and watching – selectively, under control, and with human experts as the final authority.

AI-assisted Expert-led Quality-controlled
Keep reading
01
The Market Tension

The market is divided.

In conversations with clients, we encounter two contrasting expectations: some explicitly want AI to be used, while others prefer to avoid AI tools. Both expectations are understandable – and SMD can accommodate both.

“Use AI.”

The expectation: faster processes, lower costs, and more efficient screenings.

Faster processes
Lower costs
Efficient screenings
Early-stage ideas · many variants

Key challenges in using AI

According to 300 executives at large enterprises:

40%

name compliance with AI regulation as one of their biggest challenges.

47%

name explainability & transparency as one of their biggest challenges.

Source: “Great Expectations” study by Gowling WLG / FT Longitude, 2026. Survey of 300 senior executives from companies with annual revenues of at least GBP 500 million, conducted from late January to mid-February 2026.

SMD’s answer addresses both expectations: use AI purposefully, keep data under control, safeguard quality.

02
The Core Service

At SMD, word similarity is calculated by an algorithm.
Not by generative AI.

Assessing the similarity of word marks – both in searching and watching – is our core competence. We determine the degree of similarity using our proprietary SMD algorithm – developed over decades and continously refined.

Fixed criteria

Similarity is determined using fixed, traceable criteria – not by a generative model whose output can vary even when the input is identical.

Score AI

The degree of similarity between two terms is calculated algorithmically — not estimated by AI.

People review

For ordered searches and watches, our legally trained staff select the most relevant hits and compile the report.

03
In Practice

Where SMD already uses AI today.

Concrete and hands-on – in client-facing services as well as internal processes.

Tap a card for details →

Searching & assessment
Data quality
Proven algorithms · no generative AI
Internal processes & development
No AI · proven
Word mark similarity
reproducible algorithms
Proven method

Deliberately NO generative AI: established algorithms always return the same result for the same input and same data set.

Human-controlled

AI can provide supplementary information to help customers prioritize search results. The search logic itself remains under human control and is reproducible.

Image recognition
Logo AI Knockout
Pre-Search
AI supports

Analyses visual structures within seconds – ideal for screenings, preliminary searches, and large numbers of logo variants or countries.

Humans decide

For in-depth searches, conventional device mark searches remain the quality benchmark. Final hit selection and reporting: qualified staff.

Decision support
AI Goods & Services Score
Faster Comparison of goods & services
AI supports

The tool returns the proximity of the goods and services of two trademarks as a percentage, plus a brief reasoning of the analysis.

Humans decide

The output supports the review process, while the interpretation of the result and any conclusions remain with the client.

Opposition management
AI Conflict Potential
optional add-on in TMZOOM
AI supports

An AI trained by trademark attorneys automatically assesses alert marks found by our algorithm – and delivers a recommended action based on a transparent and structured analysis on similarity, overlap of goods and services, distinctiveness, and likelihood of confusion.

Humans decide

The SMD client weighs out the arguments and decides – the final evaluation remains his professional responsibility.

Data quality
Better data basis
Vienna Classification
AI supports

A model trained on SMD data adds missing Vienna classes – device marks are classified more completely and become easier to find.

Humans decide

Curated trademark data and professional review in the background safeguard data quality.

Client service
Watch order review
e.g. after a new Nice edition (NCL 13-2026)
AI supports

Compares large lists, surfaces changes faster, and prepares possible adjustments in a structured way.

Humans decide

Whether and how coverage needs to be adjusted is up to the client – in consultation with SMD.

Development
Agentic Coding
Software development
AI supports

Supports implementing, reviewing, and evolving code – more speed and efficiency in the development process.

Humans decide

Professional oversight, security review, and responsibility stay with experienced developers.

Internal & creative
Internal processes
Reports, drafts, concepts
AI supports

Provides groundwork, structure, or ideas – for reports, first drafts, copy editing, or brainstorming.

Humans decide

Assessment, selection, professional review, and final execution are done by people.

04
Quality Control

Same input. Same result?

Run the same search several times. A generative AI tool may answer differently each time. SMD’s algorithms are reproducible.

Search termRIMATEP
Generative AI tool Not necessarily reproducible

Result per run

SMD algorithm Reproducible

Every run – identical

Result: always the same
REMITAP Remitop RINODEP RumiTop Rematinib
Same input → same result.
A real-world case

In a specific test, an AI search tool assessed the same trademark differently across repeated runs. In one run, the mark was not taken into account at all and was therefore not available to the attorney responsible for the matter. For decisions that need to be reliable, this demonstrates a real risk. SMD relies on algorithms and human expertise for the best quality results.

Conclusion

The more commonplace AI becomes, the greater the risk that healthy skepticism toward automatically generated results will fade.

If erroneous AI output goes uncorrected, the incentive to apply due care may also diminish over time.

SMD therefore favors a controlled and responsible use of AI.
05
The Right Depth

Greater scope requires more control

Drag the slider: the more that depends on a search, the more depth and control it needs.

Low-riskDecision-critical
Screening Shortlist Filing
Initial screening
Early name ideas · many variants · limited budget

A lean approach combining automated methods with targeted AI support may be sufficient here. Fast, cost-efficient, and suitable as a pre-selection.

SMD aligns its choice of method with purpose and risk: efficient when a first orientation is enough – in-depth and quality-controlled where decisions depend on the result.

06
Data Protection & Confidentiality

What the AI sees –
and what it never will.

Example: hit relevance support in TMZOOM. Only the bare minimum is sent to the AI. Confidential data stays on SMD servers in Germany.

SMD servers · Germany

This information stays confidential – is NOT transmitted

Trademark names
Owners
File numbers
Logos & client context
API boundary
G&S specifications →
← AI output
AI system · API

ONLY this data is received – data-minimised & purpose-bound

Goods & services specification of the ordered mark
Goods & services specification of the found mark
The AI returns
Similarity of the specifications78 %
Confidence of the assessment91 %
+ a brief reasoning

The final assessment remains a matter of professional responsibility.

Servers in Germany

Searching & watching data remain on SMD servers.

Temporary, then deleted

Data transmitted via the API is deleted after processing and is not used to train AI models.

Confidentiality preserved

No confidential client or context data are sent to external AI systems.

07
Our Stance

Use AI purposefully. Keep data under control. Safeguard quality.

AI is a tool

It supports – it is not automatically a promise of quality.

AI algorithm

Word similarity searching rests on proven, reproducible methods.

Confidentiality first

Only the data required for the task – no trademark names, owners, or context.

SMD stays expert-led

AI supports. Assessment, control, and responsibility stay with people.

AI can bring speed, structure, and new possibilities.

Reliable results require reproducible methods, controlled data, and professional expertise.

Not humans or AI.
The right combination matters.

AI supports. Expertise decides.