AI use cases · Departments · Industries
What AI can actually
take off your team's plate.
More than thirty tasks from customer service, sales, administration, engineering and management, each with its input, the AI step and the human check. Find the one that comes up most often in your business, and I will help you judge whether it makes a good pilot.
Discuss your projectFree initial conversation · Directly with Daniel Gläser
- 01Input
- 02AI draft
- 03Human review
- 04Approved action
Customer service
Customer service & communication
This is where most requests arrive as free text. The AI sorts, drafts and searches. Nothing goes out unless a person has approved it or the workflow has been explicitly agreed as automatic.
Triage a shared inbox
Input: emails to info@ or service@. The AI identifies the request, the customer and the urgency, assigns a category and routes it to the right person. Anything it cannot place with confidence goes to a review list that a person works through.
Workflow automationDraft replies to routine enquiries
Input: the customer's email plus your standard wording, price lists and delivery terms. The AI writes a draft in your house style and notes which document each detail came from. Your team reads it, corrects it and sends it themselves.
AI agentsWebsite chat for recurring questions
Input: visitors' questions about services, delivery times, procedures or opening hours. The chat answers only from approved content, is labelled as AI and hands over to your team as soon as a question falls outside what it knows. Conversations are spot-checked regularly.
AI chatbotSummarise phone calls
Input: call notes or, with everyone's consent, a recording. The AI summarises the request, any commitments and the next steps and files them against the customer record. Whoever took the call checks the summary before it turns into tasks.
Sales
Sales & quotations
In sales, response time matters. AI does the preparation so your team reaches the customer sooner. Prices and commitments stay with people.
Draft a quote from the enquiry
Input: the customer enquiry, your item master, price list and similar past quotes. The AI assembles line items, flags missing details and produces a draft in your template. A person signs off prices, discounts and delivery dates.
AI agentsPre-screen tenders and specifications
Input: tender documents or a requirements specification as a PDF. The AI pulls requirements, deadlines, required evidence and exclusion criteria into a checklist with page references. Sales decides whether to bid after reading the original passages.
Document processingQualify incoming enquiries
Input: a contact form, a trade fair lead or an email. The AI assigns the request to a product area, turns missing mandatory details into a follow-up question and creates a record in the CRM. Sales decides on priority and the right contact.
Workflow automationFollow up on open quotes
Input: open quotes from your ERP or CRM. A workflow spots quotes that are due, and the AI writes a short follow-up that refers to the actual content. It is sent after approval, or automatically only for cases you have explicitly defined.
Workflow automationAdministration
Accounting & administration
Receipts, contracts and forms recur constantly and are easy to check. That makes them particularly good candidates for a first pilot.
Capture incoming invoices
Input: invoices as PDFs, scans or photos. The AI reads supplier, amounts, VAT rates, line items and bank details and checks them against the purchase order and master data. Discrepancies and changed bank details go to accounts for review. Structured e-invoices in XRechnung or ZUGFeRD format are already machine-readable and need no AI for this step.
Document processingPrepare receipts for your tax adviser
Input: till receipts, credit card statements and travel expenses. The AI matches receipts to bank transactions, suggests a category and lists any that are missing. Posting and handover remain with accounts and your tax adviser.
Document processingKeep track of contract deadlines
Input: rental, maintenance, leasing and supplier contracts. The AI transfers terms, notice periods and price adjustment clauses into an overview that triggers reminders in good time. Interpreting individual clauses stays with management or your legal adviser.
Document processingPre-fill forms and questionnaires
Input: master data and a recurring form, such as supplier self-assessments, funding applications or customer questionnaires. The AI fills in known fields from earlier answers and documents and marks what is missing. Nothing is signed or submitted until someone has checked it.
Workflow automationHR
HR & organisation
HR has tighter limits. Under the EU AI Act, AI systems that filter applications, evaluate candidates or monitor employees' performance and behaviour count as high-risk. The GDPR also restricts decisions that are based solely on automated processing and significantly affect people. The cases below support organisation and communication without deciding about anyone.
Draft job adverts
Input: the tasks, requirements and conditions of the role. The AI drafts the advert and the internal job description and points out wording that is unclear or not neutral. HR is responsible for the content and for publishing it.
Answer staff questions
Input: questions about holidays, travel expenses, company cars or IT access. An internal assistant answers from policies and works agreements and cites the passage. If a rule is unclear or missing, it refers the person to whoever is responsible.
Knowledge assistantStructure onboarding
Input: checklists, handbooks and the access each role needs. A workflow creates tasks for IT, the line manager and HR, and an assistant answers new colleagues' questions from the documents. The onboarding itself remains a job for the people in the team.
Workflow automationPurchasing & stock
Purchasing, warehouse & logistics
Many orders, confirmations and delivery documents still arrive as PDFs or photos. The AI transfers them into your system, and people only handle the exceptions.
Match order confirmations
Input: suppliers' order confirmations by email. The AI reads items, quantities, prices and delivery dates and compares them with your purchase order. Only discrepancies reach purchasing, with both values side by side.
Document processingTurn emailed orders into sales orders
Input: customer orders that arrive as an email, PDF or spreadsheet rather than through a shop or EDI. The AI identifies the customer, item numbers and quantities and creates a draft order in the ERP. Inside sales checks any unclear item matches.
Document processingCapture delivery notes and freight papers
Input: photos or scans of delivery notes, consignment notes and packing lists. The AI links them to the purchase order and reports shortages or wrong items. Goods-in confirms the quantity actually counted.
Document processingStock questions in chat
Input: questions such as "How many units of item 4711 are left in the external warehouse?". An agent retrieves the figure from your inventory system through an interface instead of guessing it. It has read-only access; bookings stay in the system and with your staff.
AI agentsEngineering & production
Engineering, field service & production
The Chemnitz region is home to many machine builders and suppliers. Their knowledge often sits in service reports, manuals and test records that have grown over years. That is exactly the material AI can open up.
Maintenance reports from rough notes
Input: the technician's notes, voice memo and photos from the job. The AI turns them into a structured maintenance or service report with work done, parts used and open items. The technician checks and completes it before it goes to the customer and to invoicing.
Workflow automationSpot recurring faults in service reports
Input: years of service reports and fault notifications. The AI groups similar fault descriptions, even when they are worded differently, and shows which assemblies and machine types they cluster around. Service and design engineering judge whether there is a design cause behind it.
Local AISearch technical documentation
Input: operating manuals, maintenance schedules, spare parts lists and work instructions. An assistant answers questions such as "What is the maintenance interval for the type B spindle?" and names the document and page. Technicians always confirm safety-relevant values in the original source.
Knowledge assistantHandle test records and complaints
Input: test reports, measurements as a spreadsheet and customer complaints. The AI summarises results, drafts text for 8D reports or complaint responses and flags missing information. Root cause analysis and release stay with quality assurance.
Document processingKnowledge & management
Knowledge & management
An established business has plenty of knowledge, it is just scattered. AI makes it findable and summarises it. Judging and deciding remain management's job.
Make company knowledge searchable
Input: approved folders on a network drive or SharePoint, the wiki and project files. A knowledge assistant answers questions from these documents and cites the source. It only shows content the person asking is allowed to open anyway.
Knowledge assistantMonthly reports in plain language
Input: exports from accounting, inventory or time tracking. Fixed queries calculate the figures, and the AI describes changes and anomalies in a short report. The numbers come from your systems, not from the language model. Management does the interpreting.
Workflow automationAnalyse confidential documents in-house
Input: financial, HR or engineering documents that should not leave the building. A language model on your own server summarises them, compares versions and answers questions. We define access and logging before it goes live.
Local AIMinute meetings
Input: bullet points or, with every participant's consent, a recording. The AI produces minutes, decisions and a task list with owners. The chair checks them before they are sent. Why cases like this make an easy start, and what the research says, is covered in the guide to AI use cases for SMEs.
Guide with figures and a 90-day planBy industry
Typical starting points by industry
The same building blocks look different in every industry. Here is one case per industry that lends itself to a first pilot.
Trades
Input: a customer enquiry with photos, measurements and voice notes from site. The AI drafts a quote from your standard items and a site report from the notes. The master craftsman or owner checks quantities, prices and dates.
AI agentsRetail & e-commerce
Input: manufacturer data, data sheets and customer questions. The AI writes product copy from technical data, answers delivery status questions through the shop interface and categorises return reasons. New product copy is checked before it goes live.
Workflow automationLaw firms & tax practices
Input: client post, tax assessments, receipts and your own template letters. The AI files documents to the right matter, highlights deadlines in assessments and searches earlier submissions. Professional confidentiality makes a model on your own hardware particularly worth considering here. A qualified professional checks every substantive statement.
Local AIManufacturing
Input: work instructions, fault reports and the shift log. An assistant on the line answers questions from instructions and documented past faults, and shift handovers are summarised from the entries. The AI does not touch controls or machine parameters.
Knowledge assistantFitness & hospitality
Input: questions about memberships, class timetables, bookings or allergens. A clearly labelled chat answers routine questions from your information and hands cancellations, complaints and special requests to the team. Allergen information comes only from the list you maintain.
AI chatbotProperty management
Input: tenant reports by email or form, tenancy agreements and house rules. The AI assigns damage reports to the property and trade, drafts the contractor order and answers questions from the house rules and agreement. Management approves any order that costs money.
AI agentsFirst case
Which case should come first?
Four questions decide whether a use case makes a good pilot: How often does it come up? Is the data available digitally? What does a mistake cost? How long does it take to check a result?
A good first pilot
A case you can test and measure with real examples within a few weeks.
- Frequency: the task comes up daily or weekly, for example in the inbox, with invoices or with routine questions.
- Data: examples and documents exist in digital form, even if they are not uniformly structured.
- Cost of errors: a mistake is caught during the check and can be corrected before it reaches anyone outside.
- Review effort: a person can confirm or reject a result quickly.
Better left for later
A case that only makes sense once you have experience and solid foundations.
- Frequency: the task only occurs a few times a year, so building it hardly pays.
- Data: the knowledge lives in people's heads rather than documents, or sits in systems without an interface.
- Cost of errors: a mistake directly affects money, safety or people, such as payments or HR decisions.
- Review effort: checking a result takes almost as long as doing the task.
Working together
A clear path to everyday use.
Collect tasks
Together we list the tasks that regularly eat time and estimate how often they occur and what they cost today.
Choose the first case
Based on data, cost of errors and review effort, we pick a pilot with a measurable goal.
Build and measure the pilot
Using real, representative examples, we test quality, limits and running costs.
Adopt or drop
If the pilot convinces, I connect it to your systems. If not, we drop it and move on to the next case.
Before we begin
Good questions. Clear answers.
Which AI use case should a mid-sized business start with?
Usually a task that comes up every day and whose result is quick to check, such as triaging a shared inbox or capturing incoming invoices. Which case comes first for you is something we work out in the initial conversation, based on frequency, data, cost of errors and review effort.
Does every automation need AI?
No. Where fixed rules are enough, such as moving data from one system to another or sending a reminder on a date, conventional automation does the job. AI earns its place with free text, documents that vary in layout and questions in natural language. We often combine both in one workflow.
Do we need new software for AI applications?
Rarely. Most cases connect to systems you already have: the mailbox, a network drive or SharePoint, ERP, CRM and accounting. Where there is no interface, I look at exports, imports or a small add-on application.
Can we use AI in HR?
For drafts, internal questions and organisation, yes, under the usual data protection rules. Systems that filter applications, evaluate candidates or monitor employees' performance and behaviour are high-risk AI under the EU AI Act and come with extensive obligations. I generally advise small and mid-sized businesses against them. The guide to the EU AI Act has the details.
What does implementing an AI use case cost?
It depends on integrations, data volume, choice of model and review effort. A case that only touches one mailbox and one file store is smaller than one with an ERP interface and multi-step approval. We start with a bounded pilot whose effort is agreed up front, and I work out the running costs for model and hosting with you beforehand.
Does our data stay in the company?
If it needs to, yes: a language model can run on your own server, on your premises or in your data centre. For less sensitive cases, models in EU data centres are an option. Data type, performance needs and cost decide which route fits. Data flows and contracts are reviewed for the specific use.