AI use cases for mid-sized companies
Ten processes from sales, purchasing, accounting, engineering and administration in which AI reliably takes over work today. For each one, we describe how it works, how the benefit is measured and what a person still decides.
Sales · Purchasing · Accounting · Engineering · Administration
How to recognise a good use case
An AI use case is worthwhile when a task occurs several times a week and one person is responsible for the process. The necessary information has to be in emails, PDFs, the ERP or document management. Errors must also be detectable and correctable before they take effect. Then you can measure beforehand how long a case takes and check afterwards whether the AI saves time or avoids errors.
Less suitable are processes with few cases a year and knowledge that only exists in individual people’s heads. The same applies to decisions that cannot be corrected afterwards.
Sales and customer service
Here the benefit usually comes from shorter response times. Prices, dates and commitments remain with your employees.
From customer enquiry to draft quote
Enquiries arrive by email or as a PDF. The AI extracts items, quantities and dates, retrieves prices and availability from the ERP or CRM, flags missing information and creates a draft quote.
We measure the benefit by minutes per enquiry, time to draft and the share of drafts without rework. Prices, delivery dates and sending the quote are decided by a person. Item master and prices must be maintained in the ERP.
Sorting and answering customer requests in service
The AI assigns incoming requests to a topic, retrieves the order or delivery status, suggests a reply and creates the ticket.
We measure the benefit by time to first response, time to resolution and the share of escalations. Binding commitments, goodwill and anything that costs money are decided by your staff. It needs a ticketing or CRM system with an interface.
Purchasing and accounting
Documents and quotes are easy to structure. The AI reads, compares and presents discrepancies, and bookings and orders only happen after approval.
Reading delivery notes, order confirmations and invoices
The system recognises the document type, reads the fields and matches them against the order and master data. If everything matches, the document goes to the ERP or document management, and discrepancies go to a person.
We measure the benefit by errors per field, processing time per document and the share without manual rework. Booking, payment and resolving discrepancies stay with a person. Structured e-invoices are processed directly; text recognition is only needed for paper or scans.
Comparing supplier quotes
The AI places prices, terms and delivery times from several quotes side by side, prepares follow-up questions and monitors promised delivery dates.
We measure the benefit by the time for the comparison, the completeness of the information and overlooked discrepancies. The order itself and any change to suppliers’ bank details are decided by a person. Quotes must be available digitally.
Engineering, service and quality
Technical knowledge sits in manuals, service reports and people’s heads. AI makes it searchable, but only with source references.
Technical knowledge with source references
Manuals, operating instructions and service reports become searchable. The AI answers a question with a reference to where the information was found. The search only shows what the person asking is allowed to open.
We measure the benefit by search time, the share of cases solved without follow-up questions and the correctness of answers. Whether an answer is right for the specific case is judged by a person. If there is no supporting source, the AI says so. Documents must be stored in a repository with clear access rights.
Capturing maintenance and service reports
Technicians dictate or briefly note down the job. The AI turns this into a structured report with the asset, fault pattern, action taken and materials, and passes it to the system.
Metrics are the time for the report and the completeness of mandatory fields. The report and invoicing are approved by a person. It requires a fixed report format.
Assigning complaints
Incoming complaints are sorted by product, type of fault and customer. The AI looks for similar earlier cases and suggests the responsible team.
We measure the benefit by lead time per complaint and the share of correctly assigned cases. Cause, action and goodwill are decided by a person. Complaints must already be recorded digitally.
Administration, controlling and IT
A lot of time goes into reports, tickets and tenders. The AI prepares, and the calculations happen in the specialist systems.
Structuring tenders
The AI reads tender documents, extracts requirements, deadlines and required evidence and creates a checklist for the bid team.
We measure the benefit by the time to the go/no-go decision and by overlooked requirements. Participation, price and the content of the bid are decided by the bid team.
Draft reports in controlling
KPIs come from the ERP or BI system and are calculated there. The AI flags variances and drafts a first commentary for the monthly report.
We measure the effort for the report and check whether the figures match the system. Assessing the variances and approving the report stay with a person. KPIs must be clearly defined in the BI system.
Preparing IT service tickets
Tickets are structured, missing information is requested and suitable instructions are found in the internal documentation.
We measure the benefit by processing time and the share of tickets resolved at first contact. Every change to permissions and systems is approved by a person. Critical actions run through separate, tightly limited tools.
Where we advise against AI
We do not use AI as a starting point for selecting job applicants, for medical decisions or for controlling safety-critical machinery. The legal requirements are high, and the EU AI Act classifies AI used in recruitment, for example, as a high-risk system. And an error often cannot be undone.
Processes with few cases a year are also rarely worthwhile. There, setting it up costs more than it saves. We point this out in the initial consultation.
How much time a use case saves
That depends on the volume and the process, which is why we do not give blanket percentages. A worked example: with 1,000 cases a month and four minutes saved per case, including review and rework, around 67 hours a month are freed up.
Time freed up is not yet a saving. Whether it translates into money depends on what your team uses it for. In the AI Potential Check, we calculate this with your own volumes and times.
Frequently asked questions about AI use cases
Which AI use cases are most worthwhile for SMEs?
Processes with many similar cases and digital data: customer enquiries and quotes, incoming documents, technical knowledge and service reports. There the benefit can be measured, and errors are noticed before they take effect.
What is a good first use case?
One that occurs frequently, has a responsible person and whose result a person checks before it takes effect. Often this is reading incoming documents or creating a draft quote from a customer enquiry.
Do we need our own data or our own model for AI?
You almost never need your own model. The AI works with your existing data from ERP, CRM and document management, which is queried at runtime. What matters is that this data is maintained and accessible.
Does AI replace employees?
Not in the processes described. The AI takes over reading, sorting and preparation. Checking, deciding and taking responsibility remain with your employees, who then have more time for it.
How do we measure whether a use case pays off?
Before the start, we measure processing time and error rate in the current process. In the pilot, we measure the same values with AI, including review and rework. Only then is the decision made on regular operation.
Which use cases are not suitable for AI?
Processes with few cases a year, decisions about people such as selecting applicants, medical decisions and controlling safety-critical machinery.
Finding the right AI use case
Briefly describe the process that costs your team the most time. In the free initial consultation we tell you whether AI makes sense there and how the benefit can be measured.
