Introducing AI in your company: a guide for SMEs and teams
Introducing AI in your company means systematically identifying repetitive processes, involving your team early on, and starting with a manageable pilot project that brings visible relief before you roll the solution out to other areas.
What does introducing AI into a company mean?
Introducing AI into a company does not mean buying software and hoping it works, but understanding processes, automating the right steps, and having the team on board from the very beginning. An AI rollout is not a traditional IT project that is installed once. It requires continuous adaptation, willingness to learn, and a clear idea of which problem needs to be solved.
The Fraunhofer IAO focuses on AI adoption as an organisational process. An article by the SIX Group on AI confidence shows: AI must be integrated into the company and into its processes, not merely rolled out. Leaders must actively help shape the process, not just sign off on it.
Areas where Swiss SMEs are already achieving relief through AI today:
Accounting and receipt processing
Customer service and email triage
Marketing and content production
Internal reporting and appointment scheduling
Which processes are best suited for AI first?
The best entry point for adopting AI is not the most impressive use case, but the one where your team loses the most time and requires the least amount of thinking. The first step is to define clear business goals and align them with the processes that are the most expensive, slowest, or most tedious. The key question is: Which process costs the most time? Not: Where does AI sound particularly impressive?
A three-criterion filter separates suitable processes from unsuitable ones:
Repetitive: The task runs the same way every week.
Rule-based: There are clear if-then rules without significant room for discretion.
Time-consuming: The manual effort clearly outweighs the benefit of handling it in person.
Processes that meet all three criteria are suitable initial candidates. Four examples from everyday SME life:
Receipt capture and invoice processing
Process: Manually recording, verifying, and filing receipts (as well as how AI makes accounting in Switzerland more efficient)
Why it is suitable: Always the same structure, clear fields, high time commitment with low decision-making depth.
Example tool: AI-native accounting software such as nextesy, document management with OCR recognition.
First step: Route all incoming receipts into a digital inbox and enable automatic recognition.
Customer service emails
Process: Screening incoming enquiries, categorising them, and formulating standard replies.
Why it is suitable: The majority of enquiries are recurring; response templates can be structured.
Example tool: ChatGPT with company-specific instructions or Microsoft Copilot in an Outlook environment.
First step: Define the five most common types of enquiries and store a sample response for each.
Internal knowledge search
Process: Employees manually search for information across folders, wikis, or old emails.
Why it is suitable: Search time is purely overhead work with no direct added value for clients.
Example tool: Document AI applied to existing repositories; NTT DATA uses similar approaches with GoodGPT for structured knowledge search across enterprises.
First step: Consolidate all relevant documents in one location before activating the tool.
Content drafts in marketing
Process: Creating initial drafts for newsletters, social media posts, or product descriptions.
Why it is suitable: The initial draft takes a lot of time, but will be edited anyway.
Example tool: ChatGPT or a specialised writing tool with brand briefing.
First step: Define a briefing template that sets out tone of voice, target audience, and length.
Before a suitable process can be automated, one fundamental question still needs to be answered: Is the relevant data even digitised and accessible in usable quality? Without this prerequisite, any pilot will fail, regardless of how good the chosen tool is. A quick data check before starting saves weeks of frustration later on.
What do you tell your team?
Wherever AI is introduced, questions arise that no tool can answer: That is why a good AI rollout does not begin with the software, but with a conversation. The most common concern within the team is: “Will I be replaced?” This question is legitimate and must be addressed openly, not explained away. AI should be positioned as a tool that frees up time for value-adding tasks, but does not replace people.
Studies indicate that many companies overestimate technical readiness and underestimate human readiness. A lack of AI skills is the biggest hurdle for many companies when adopting AI, not a lack of tools. The Fraunhofer IAO recommends involving employees right from the process selection stage, not just at go-live. A visible early win also helps: according to the Fraunhofer IAO, most SMEs start with a first flagship project, which often helps break down reservations about AI within the team.
Four concrete steps for the conversation with the team:
Communicate transparently: Explain the goal of the AI rollout openly before rumours arise. “We are automating repetitive tasks so you can focus on things where you are genuinely needed.”
Let the team help decide: Involve employees in deciding which tasks to automate first. Those who have a say are less likely to resist.
Involve early adopters: Identify team members who have an affinity for AI and give them a visible role as internal champions. Peer-to-peer learning works better than top-down training.
Establish a feedback channel: Set up an accessible channel where the team can report issues and observations. Mistakes are a source of learning, not a failure.
Leaders who visibly use AI themselves remove the biggest hurdle: uncertainty over whether the commitment from above is genuinely serious. Read more on why fear of AI in SMEs is usually unfounded.
Four steps to your first AI pilot project
An AI pilot is not an aimless experiment, but a clearly defined step with a process, a team, and a benchmark for success. The four phases—discovery, design, pilot, and scaling—can lead to a productive outcome within two to four months.
Discovery
Step: Identify bottlenecks, conduct a data check, assess resources.
Goal: Find a clearly defined process that meets all three criteria (repetitive, rule-based, time-consuming) and is immediately measurable.
Outcome: A prioritised list of two to three candidate processes along with an assessment of data availability.
Common mistake: Choosing the largest and most complex process first because it seems most impressive.
Design
Step: Define goals, clarify governance (who is allowed to do what, who is liable for errors), establish roles.
Goal: Create a written mini-plan with success criteria before activating the first tool.
Outcome: Clear responsibilities, defined KPIs, and a timeline for the pilot.
Common mistake: Postponing governance questions until “later” and then lacking a clear answer when an error occurs.
Pilot
Step: Test a small, clearly defined process with a real team and real data, not demo data.
Goal: Gain initial dependable experience and evaluate the process under real conditions.
Outcome: Concrete before-and-after data on time spent, error rates, and usage behaviour.
Common mistake: Testing only with the most enthusiastic person, who would use the tool anyway.
Scaling
Step: Document successes, expand to further areas, continuously optimise.
Goal: Turn a functioning pilot into an ongoing operational process.
Outcome: Role-specific training, internal champions, and a measurement routine that keeps progress visible.
Common mistake: Scaling before the pilot has delivered stable results.
The Fraunhofer IAO stresses in its KI-ULTRA project model that no phase should be skipped. The pressure to quickly show visible results often leads to rushing discovery and design, causing the pilot to start on shaky foundations.
What do you need to consider legally?
Introducing AI does not mean ignoring data protection, but asking three questions before activating a tool. In parallel with the pilot, Swiss SMEs need to keep two legal frameworks in mind: The Swiss Data Protection Act (FADP) regulates how personal data may be processed, and the EU AI Act is the European Union’s first comprehensive regulation specifically governing requirements for the deployment of AI systems. The EU AI Act can also be relevant for Swiss SMEs that work with clients or suppliers in the EU area. Companies must categorise their AI applications by risk: High-risk systems are subject to stricter requirements than simple automation tools. The SIX Group recommends a data strategy based on the “three Rs”: Data must be right, relevant, and representative before serving as the basis for AI decisions.
Three questions that should be answered before starting any AI tool:
What is being processed: Check whether personal data of employees or clients flows into the system, and ensure that this happens in compliance with the FADP.
Where is the data stored: For cloud solutions, prefer server locations in Switzerland or the EU; enquire explicitly if the provider does not give clear information.
What can be proven: Keep documentation of which data comes into contact with which AI system. This documentation requirement is not bureaucratic overhead, but protection in the event of a dispute.
How do you measure whether AI adoption is working?
If you don’t measure, you often only realise after months that while the tool is installed, it is barely used. Three manageable KPIs without a BI tool are enough to evaluate the pilot and justify scaling. ipt and practical experience in AI data strategy confirm that SMEs without a data science team can make valid assessments using simple before-and-after comparisons. The SIX Group warns against the productivity illusion: First define what “productivity” means in your own company, then measure. Share results regularly with the team so that the benefits become tangible for everyone.
The three KPIs that work without specialised tools:
Time savings per process: Clock the manual effort before the pilot and compare it four weeks after rollout. Even a rough estimate is better than no figure.
Error rate in the automated step: Count how often the output of the AI tool needs to be corrected. A declining error rate means the model is learning or the underlying data is improving.
Weekly usage rate in the team: How many people actually use the tool every week? A usage rate below 50% is a sign of a need for training or adoption issues.
A maximum of three to five KPIs, clearly visualised, with one responsible person: that is all it takes for an initial overview.
How nextesy natively integrates AI into everyday SME operations, without tool sprawl
For SMEs looking to launch a pilot without an IT department behind them, nextesy shows how AI capabilities like receipt capture, payment matching, and invoice processing are natively embedded into existing administrative workflows rather than sitting alongside as a fifth individual licence. The product was built from the ground up as an AI-native platform, advised by the ETH AI Center, the HSG (University of St. Gallen), and supported by Innosuisse, the Swiss Innovation Agency. The difference between a natively integrated and an add-on solution is the difference between immediately noticeable relief and another tool that nobody opens regularly.
What this means in concrete terms for getting started:
Automation: Receipt capture, payment matching, and invoice processing run directly in the system—no separate tool, no manual intermediate steps.
Data protection: Developed in line with Swiss FADP standards, no detours via foreign servers.
Price: From CHF 29/month for SMEs, from CHF 19/month for fiduciaries; test free for 14 days without a credit card.
Migration: Transitioning from existing solutions is actively supported, including a migration bridge.
