How to recognise a good AI agency
As the market is diverse, there are a few reliable indicators to help with the selection process. A key criterion is specific references: a reputable AI agency can describe a comparable project from the recent past, detailing the problem, scope, results and even setbacks, rather than simply displaying a wall of logos. A second indicator is the pragmatic approach: rather than a multi-year transformation programme, a good agency starts with a clearly defined, measurable first step – often a proof of concept that demonstrates within a few weeks whether the AI solves the specific problem.
Thirdly, knowledge transfer forms part of the value proposition: documentation, handover and training ensure that the know-how remains within the company and that no long-term dependency arises. Fourthly, compliance is not an afterthought, but is taken into account from the outset – this includes the General Data Protection Regulation (GDPR), data processing agreements and the requirements of the EU AI Act. Finally, a professional AI agency is characterised by measurability: before the project begins, it specifies the key performance indicators against which success can be gauged – such as time saved, response times or conversion rates.
Red flags to watch out for when selecting a provider
- Blanket promises of a miracle cure (“AI is revolutionising everything”) without a single concrete use case for your own business.
- No indication of price until a paid workshop lasting several weeks has been booked.
- Advertising using “proprietary AI”, which is essentially just a purchased standard language model with the agency’s own logo.
- Guarantees of success with precise accuracy promises, without knowing the client’s data set.
What an AI agency costs
Price ranges are wide and depend heavily on the scope of work. In line with industry standards, a one-day strategy workshop typically costs in the low four-figure range, a proof of concept in the region of around 5,000 to 15,000 euros, and a complete initial project comprising a potential analysis and the first productive use case typically in the mid five-figure range. On top of this come running costs: the usage-based token costs of the language models used (such as OpenAI GPT, Anthropic Claude or Google Gemini), as well as operation and monitoring as a separate item. The key factor in evaluating a quote is not so much the absolute figure as the transparency of the pricing model: Hidden costs for data preparation, dashboards or ‘model maintenance’ are a red flag.
AI agencies and AI visibility (AI SEO)
A growing sub-segment is AI SEO agencies, which specialise in Generative Engine Optimisation (GEO) – in other words, ensuring that a brand appears in the responses from AI systems such as ChatGPT, Perplexity or Google AI Overviews. The strategies differ from traditional SEO: structured data, citable content, technical crawlability for AI bots and server-side rendering. An AI agency with GEO expertise can measure whether and how a website is crawled and cited by AI crawlers. Market rates for ongoing GEO will be noticeably higher than for traditional SEO by 2026; for medium-sized clients, monthly fees in the four- to low five-figure range are standard. Anyone commissioning an AI SEO agency should ask specific questions about how AI visibility is measured – for example, via citation tracking, brand mentions in language models, or a technical check of the status code an AI bot actually receives. If the answer remains vague and focuses solely on keyword density and backlinks, the agency’s GEO expertise is questionable.
The selection process: asking the right questions
The suitability of an AI agency can be assessed effectively during the initial consultation if the right questions are asked. A concise list of questions that distinguishes substance from mere sales rhetoric has proven its worth:
- What comparable project has been implemented in the last twelve months – and what went wrong?
- What is the smallest possible first step, and how much does it cost?
- How will you ensure that our own team can continue working independently once the project has ended?
- How will the GDPR and the EU AI Act be handled in this specific case, including model hosting and data flows?
- Which key performance indicators will be used jointly to assess whether the project was worthwhile?
An AI agency that answers these questions without evasion is worth a closer look. Vague or evasive answers are a clear warning sign. Another important consideration is ownership of the solution: code, prompts, configurations and data should belong to the client, and the architecture should be deliberately model-agnostic to avoid vendor lock-in. This makes it possible to switch language model providers at a later date without having to rebuild the entire solution from scratch.
Organisational fit also plays a role. Medium-sized companies with between 50 and 5,000 employees have different requirements to large corporations: they need rapid, tangible results, dedicated points of contact and a provider who understands the specific characteristics of established system landscapes and limited internal AI resources. A specialised AI agency with experience in the SME sector is generally the better choice here than a provider geared towards large corporations, whose methods and pricing models are not tailored to this target group.
A real-world example
A medium-sized online retailer wants to ease the burden on its customer service team. A full-service AI agency would not immediately sell a large-scale system, but would first analyse the most frequent service enquiries. On this basis, a proof of concept is developed: a chatbot that uses retrieval-augmented generation to access the shop’s actual product and FAQ content and demonstrates, within two to four weeks, whether it can reliably answer a significant proportion of enquiries. Only once the defined key performance indicators – such as a measurable ticket deflection rate – have been achieved will the next steps follow: expansion, deeper integration into the shop and ticketing systems, and training for the customer service team. This step-by-step approach is characteristic of a reputable AI agency and reduces the client’s investment risk.
Legal framework
AI projects in Germany and the EU are subject to several legal frameworks. The GDPR regulates the handling of personal data and requires, amongst other things, data processing agreements and – depending on the data category – EU hosting or pseudonymisation. The EU AI Act supplements this framework with risk-based obligations and is coming into force in stages. A competent AI agency takes these requirements into account from the outset and integrates them into the architecture and operational concept, rather than addressing them retrospectively.
Frequently asked questions about AI agencies
What exactly does an AI agency do?
An AI agency analyses business processes, develops tailored AI use cases, implements them technically, integrates them into existing systems, trains the team and measures the benefits. The focus is on productive implementation, not purely on consultancy.
What distinguishes an AI agency from an AI consultancy?
An AI consultancy primarily provides strategy and a roadmap. A full-service AI agency also handles the technical implementation and operation – it doesn’t stop at the concept stage.
How much does it cost to work with an AI agency?
A strategy workshop typically costs in the low four-figure range, a proof of concept around 5,000 to 15,000 euros, and an initial production project typically in the mid five-figure range. On top of this come ongoing token, operational and monitoring costs.
How can I recognise a reputable AI agency?
By looking for specific references, a pragmatic, low-risk starting point (proof of concept), built-in knowledge transfer, compliance (GDPR, EU AI Act) factored in from the outset, and clear, pre-agreed key performance indicators.
Do I even need an AI agency, or is traditional automation sufficient?
Not every process requires a language model. For clearly rule-based workflows, deterministic automation is often more cost-effective and requires less maintenance. AI is particularly worthwhile for unstructured language, documents, classification and high variability. A good AI agency will be honest about this.
