The AI Customer Journey describes a customer journey in which an AI assistant accompanies or takes over individual steps from defining needs through provider selection and evaluation to contact or transaction. For your SME, this means that services, prices, availability, trust signals, and limits of action must be so clearly defined that both humans and AI systems receive the same reliable answers.
This trend has already impacted consumer behavior. According to a 2025 consumer survey published by the Capgemini Research Institute, 58 percent of respondents stated that they prefer using generative AI over traditional search engines to access product and service recommendations. This figure doesn't prove that traditional search is disappearing. However, it does demonstrate that AI-powered product search and provider selection are becoming a relevant approach.
In my work with SMEs in South Tyrol, I've observed a crucial point: the new customer journey rarely fails due to a lack of AI. More often, the problem lies in unclear services, scattered data, contradictory information, a lack of responsibilities, and undefined approval processes. That's precisely where you should focus your efforts.
What distinguishes the AI Customer Journey from the classic customer journey
The classic customer journey typically leads people through search engines, recommendation portals, websites, forms, and personal conversations. The potential customer opens several pages, compares information, and makes the decision themselves.
In an agent-based customer journey, an AI assistant takes on more coordination. An AI agent goes a step further: Within predefined rules, it can retrieve data sources, use tools, and perform permitted actions. This shifts part of the AI-driven buying process from the visible website to a dialog-based or automated environment.
- AI Customer Journey: A customer journey where AI supports the search, evaluation, selection, or individual action steps.
- Agent Customer Journey: A customer journey in which an AI agent retrieves information, evaluates options, and prepares or executes actions within defined rules.
- Human approval: A binding control point at which a responsible person reviews, confirms, modifies, or rejects a proposed action.
- Transaction interface: A technically defined access point through which an authorized system can, for example, query availability, submit a request, or initiate a booking.
Stripe's documentation describes Agentic Commerce as the use of AI agents that find and evaluate offers and complete transactions. The Agentic Commerce Protocol allows merchants to provide standardized product catalogs and purchasing processes. However, this development does not mean that every SME must immediately enable autonomous ordering.
An agent-driven customer journey isn't good when AI makes as many decisions as possible. It's good when information is reliable, actions are transparent, and boundaries are clearly defined.
The AI Customer Journey in five phases
1. The question: What is actually needed?
In the classic search journey, the prospective customer formulates a search query, opens several results, and tries to match their problem to a suitable offer. The choice of words influences which providers are displayed.
In the agent-based customer journey, the prospect describes their goal in everyday language: for example, "I need a bilingual website for a small tourism business" or "Find an electrician in my area who has availability next week." The AI assistant breaks down the request into criteria and searches for relevant information.
Your website needs to list more than one service name. It should clearly answer the following questions:
- What specific problem does the service solve?
- For which target groups and company sizes is the offer suitable?
- In which field do you work?
- What requirements must be met?
- Which cases are explicitly not suitable for your business?
- In which languages are consultation and services available?
Human checkpoint: For complex or security-relevant requirements, an AI assistant can structure the needs. A qualified person must then verify the technical suitability and address any open questions.
2. Summary: What information does the system understand?
In the traditional search process, the prospective customer reads product pages, brochures, reviews, and frequently asked questions. Unclear or contradictory information must be identified by the individual.
During the agent journey, the AI summarizes services, prices, conditions, and differences. If clear information is missing, the system can supplement it with information from external sources or confuse similar offers.
Every important service therefore requires a concise, independent description. It should include the scope of services, the result, the process, the duration, exclusions, and the next step.
Prices do not necessarily have to be published as a fixed amount. A transparent pricing logic such as "per area," "per person," "per service module," or "after on-site inspection" is more informative than simply stating "price on request."
Structured data supplements visible content with machine-readable information. However, it does not replace clear language. If the visible content is incomplete or incorrect, technical markup does not make the information reliable.
Human control point: Those responsible within the company must determine which source is binding and who maintains prices, conditions, capacities and services.
3. Provider selection: Why should AI consider your business?
In traditional searches , rankings, advertisements, well-known brands, and personal recommendations influence the initial selection. People also react to design, language, images, and personal appeal.
AI-powered customer searches compare providers based on specific criteria. These can include location, specialization, availability, pricing, experience, references, accessibility, and cancellation policies. A business can be a perfect fit in terms of expertise but still be overlooked if crucial information is missing or cannot be clearly assigned.
Reliable signs of trust include:
- Full company name, location and contact details,
- clearly named responsible persons,
- verifiable qualifications and authorizations,
- specific references with task definition and result
- up-to-date ratings on verifiable platforms,
- transparent terms and conditions of business, booking and cancellation,
- Consistent information on websites, directories and profiles,
- Visible update data for time-critical information.
Positioning becomes more important as a result. An AI assistant can only make meaningful distinctions if it's clear what your business stands for and who your services are intended for. Technical implementation cannot compensate for unclear positioning.
Human checkpoint: A pre-selection process must not inadvertently narrow the available options. Selection criteria, paid placements, and identifiable conflicts of interest should be transparent.
4. The test: Is the recommendation reliable?
In the traditional customer journey , a person checks references, reviews, work samples, terms and conditions, and personal impressions. Follow-up questions are made by phone, email, or in person.
In the agent-based customer journey, an AI assistant can combine information from multiple sources, highlight inconsistencies, and ask targeted follow-up questions. Uncertainty must remain recognizable as such. "The company offers this service" is a different statement than "The company might be suitable based on similar references."
For a reliable assessment, your company needs clearly defined evidence. An anonymous logo in a reference collection is less meaningful than a documented project that includes the initial situation, services provided, timeframe, and results. Equally important are readily available contacts for questions that aren't answered by publicly available information.
Human checkpoint: Professional suitability, special requirements, sensitive data and risks must be checked by a responsible person before a binding commitment is made.
5. Contact or transaction: What is AI allowed to do?
In traditional travel, the prospective customer submits a form, makes a phone call, schedules an appointment, or completes a booking. The action step is visible and consciously initiated.
During the agent journey, a system can compile contact details, retrieve available appointments, pre-fill an inquiry form, or prepare a transaction via an interface. With appropriate authorization, an AI agent can also execute a booking or order.
An agent-ready website therefore requires clear technical permissions and operational rules. The interface must not only define what is technically possible , but also specify what the AI agent is allowed to do.
- What data is the AI assistant allowed to read?
- What information is the system allowed to transmit?
- Which action is merely a non-binding inquiry?
- When does a reservation or a contract come into effect?
- What Budget- or do value limits apply?
- How long is availability valid?
- Who is authorized to approve or revoke an action?
- How are the process, data source, and consent logged?
Human checkpoint: Explicit confirmation is required before any significant financial, contractual, data protection, or security-related decisions are made. Tacit consent is not a reliable approval rule.
Recommendation, prepared action, and autonomous transaction
In everyday language, various functions are referred to as "AI agent." For your business, the following distinction is more important than the technical term:
- Suggestion: The AI assistant suggests suitable providers or offers. The human then reviews and decides independently.
- Prepared action: The AI assistant gathers information, selects an appointment, or creates a shopping cart. The human reviews the summary and explicitly approves the next step.
- Autonomous transaction: The AI agent executes an order, booking, or other binding action within predefined rules.
For many small businesses, a prepared approach is the most sensible way to start. The business saves time because inquiries arrive complete and structured. The prospective client retains control over price, scope, and closing.
You shouldn't delegate these decisions without first checking them.
- Contracts concluded above defined Budget- or value limits,
- individual discounts and different payment terms,
- Processing or disclosure of sensitive personal data,
- Custom-made products with irreversible costs,
- professional decisions relating to safety or health,
- Commitments in cases of unclear availability or capacity,
- Conflict cases, complaints and legally relevant declarations.
The legal framework also depends on the specific application. Regulation (EU) 2024/1689 requires, in particular, a risk management system pursuant to Article 9 and human oversight pursuant to Article 14 for high-risk AI. Article 50 contains additional transparency obligations for certain AI systems.
Not every AI application used by an SME is automatically a high-risk system. The application, the data processed, the risk class, and the company's role must be assessed separately.
Three scenarios from the service, craft and tourism sectors
Scenario 1: Selecting a management consultancy or specialized service provider
An entrepreneur is seeking support with positioning, digitalization, and redesigning his website. He describes company size, market, languages, and more to an AI assistant. Budgetframe and desired result.
The AI assistant can only reliably compare suitable providers if their positioning is clearly defined. Information is needed on the scope of services, suitable clients, methodology, specialization, team, references, and opportunities for initial contact. A statement like "We offer customized solutions for all industries" provides hardly any usable selection criteria.
A sensible procedure looks like this:
- The AI assistant structures the needs and identifies open questions.
- The system narrows down providers by specialization, region, language, and project type.
- The interested party receives a reasoned pre-selection with sources and identifiable uncertainties.
- The AI assistant prepares a request including the initial situation, goals, and framework conditions.
- The interested party reviews the data and deliberately submits the request.
- In the initial meeting, both sides check whether working methods, values and expectations match.
If references, responsibilities, or a clear description of suitable clients are lacking, the system should not make a reliable recommendation. Especially with strategic services, personal fit remains a crucial part of the decision.
Scenario 2: Inquiry to a craft business
A homeowner is looking for a tradesperson to replace several windows. The AI assistant needs more than just the information that the company offers "window installation in South Tyrol".
For a usable inquiry, the system needs the service area, the trades offered, material options, approximate capacities, requirements for an on-site appointment, and information for preparing a quotation. Photos, measurements, and the property address can facilitate preparation but must be processed securely and for the intended purpose.
A realistic scenario:
- The AI assistant asks for the object type, location, number, dimensions and desired materials.
- The interface checks whether the location is within the catchment area and whether viewing appointments are available.
- The system submits a structured, explicitly authorized request.
- The company checks photos, dimensions, structural situation and capacity.
- An employee confirms the on-site appointment.
- The binding offer is only created after professional review and human approval.
Without information on the catchment area, capacity, and type of work, the AI assistant should not provide a binding delivery date or final price. For custom-made products, the order should not be based solely on estimated dimensions.
Scenario 3: Searching for and booking a tourism offer
A family is looking for accessible accommodation in South Tyrol. They require five nights, breakfast, parking, a pet-friendly room, and clear cancellation policies.
Up-to-date, multilingual data is particularly important in the tourism industry. The AI assistant must be able to differentiate between room categories, actual availability, included services, surcharges, accessibility, check-in times, and cancellation policies. Machine translation must not translate "partially accessible" into "fully accessible."
The process can be structured as follows:
- The AI assistant records the time period, persons, age of the children and special requirements.
- A booking interface provides current availability and valid prices.
- The system displays inclusive services, surcharges, and conditions separately.
- The guest confirms the room, total price, cancellation policy and personal data.
- Only then will the booking be binding and confirmed.
If reliable accessibility information is lacking or availability data is outdated, the AI should only prepare a request. For South Tyrolean businesses, a well-planned German-Italian website strategy is also crucial: content must be technically consistent in each language, not just linguistically similar.
Improving the AI Customer Journey in 30, 60 and 90 days
In the first 30 days: Gain clarity
- List your most important services and offers in full.
- Define target groups, areas of application, exclusions, and catchment area.
- Check the company name, address, contact details, and opening hours for consistency.
- Provide additional price information or a clear pricing logic.
- Match references, qualifications, conditions, and frequently asked questions to the appropriate services.
- For each piece of information, assign a responsible person and a reliable source.
The result after 30 days should not be a new AI system, but a reliable information base. This work is already improving traditional search, consulting, and internal processes.
By day 60: Structuring queries and data
- Revise service pages so that each page answers a specific user question.
- Add suitable structured data and check its consistency with the visible content.
- Build request forms according to actual operational requirements.
- Separate mandatory information from optional information.
- Define which data may be transmitted via email, form or secure interface.
- Check multilingual content for factual inconsistencies.
If your website is to be not just an information platform, but a reliable data source, then viewing it as a knowledge interface is helpful . Crucially, a well-maintained source is essential, rather than many conflicting data sets.
By day 90: Test interfaces and releases
- Define roles, rights, Budgetlimits and release levels.
- Start with a limited function, such as an appointment request or request preparation.
- Test normal cases, incomplete information, and deliberate false queries.
- Log data source, time, suggested action, and release.
- Define when a process is handed over to a human.
- Check for cancellation, correction, failure, and manual processing.
- Measure whether inquiries become more complete and whether time is actually saved.
Only automate further once the limited process is running reliably. At Berger+Team, work on AI and digitalization therefore begins with the operational bottleneck, the available data, and the responsibility – not with a specific tool.
Win-win-win instead of hidden influence
A well-designed agent-driven customer journey saves prospects time searching, delivers better leads to the business, and reduces unnecessary work. This benefit must not come from manipulating choices, obscuring uncertainty, or forcing a sale without clear consent.
The best AI-powered buying process isn't the fastest. The best process leads to a suitable and understandable decision.
The win-win-win principle is a benchmark for me. An AI assistant should empower people, not overwhelm them. Technology can improve orientation, but it must not make responsibility invisible.
Questions and answers about the agentic customer journey
Will I lose control over the purchasing process due to an AI agent?
Not if rights and permissions are clearly defined. Start with recommendations or pre-defined actions and only allow autonomous transactions within clear boundaries that you can test and modify at any time.
What data is an AI assistant allowed to process when responding to a query?
An AI assistant should only collect data that is necessary for the specific purpose. Personal or sensitive data requires a legal basis, transparent information, limited access, and appropriate deletion rules; you should have the specific implementation reviewed for compliance with data protection regulations.
Do I need a completely new website for that?
Not necessarily. A small website can already function well if services, target groups, location, terms and conditions, references, and contact methods are clearly defined. Technical interfaces can be added gradually later.
Is structured data already an interface for AI agents?
No. Structured data helps systems understand publicly visible information, while an interface enables targeted queries or actions. For availability, appointment reservations, or bookings, you usually need controlled technical access with defined permissions.
What does it cost to switch to an AI-powered customer journey?
The costs depend on the initial state, data quality, and the desired level of automation. Start with clear content and a structured query; this limited start is usually more economical than immediately building an autonomous transaction.
How can I prevent incorrect recommendations from AI systems?
Incorrect recommendations from external systems cannot be completely prevented. You can reduce the risk through clear service descriptions, consistent company data, reliable evidence, up-to-date terms and conditions, and regular test requests in the most important languages.
Who is liable in the event of an incorrect booking made by AI?
Responsibility depends, among other things, on the system used, the authorization, the contract terms, and the specific error. You should have legally binding booking procedures for Italy and the European Union reviewed and ensure that all consents are clearly documented.
How important is multilingualism for South Tyrolean SMEs?
Multilingualism is relevant when target groups search for and compare services in German, Italian, or English. The crucial factor is not the number of languages, but the consistency of pricing, availability, terms and conditions, and scope of services in each published language version.
Where should a small business sensibly begin?
Start with a service, a clearly defined target group, and a common inquiry process. If an AI assistant can prepare a complete request from this, and a human approves it, you have a controllable entry point with demonstrable benefits.
What role does human approval play in the long term?
Human approval remains important wherever context, responsibility, or significant consequences are involved. Good automation reduces unnecessary routine, not human judgment.
Your prioritized self-assessment
First, check whether an outsider can clearly understand from your website what you offer, who the service is suitable for, where you operate, and what the next steps are. Then, check whether your pricing or pricing logic, availability, references, terms and conditions, and responsibilities are up-to-date and consistent.
Next, decide which actions an AI assistant is allowed to prepare and which an AI agent is allowed to execute. If this boundary cannot be described in a clear sentence, the process is not yet ready for an autonomous transaction.
After over 20 years in branding, web development, and digitalization, my experience is clear: technology amplifies existing clarity, but also existing chaos. If you want to streamline your AI customer journey for both people and AI systems, we can identify the biggest bottleneck and an appropriate first step in a direct introductory meeting. You can reach Berger+Team at info@berger.team.
