Guides / GUIDE-001

Finding and prioritising AI use cases in your company

Run a short discovery process with the people who do the work, score candidates on value, feasibility, data readiness and risk, and start with two or three quick wins.

Most SMBs do not lack AI tools. They lack a shortlist of concrete, worthwhile problems. This guide takes you from "we should do something with AI" to a scored backlog and two or three quick wins in about four to six weeks. It covers the unglamorous work too: small automations and data preparation, which often bring the first real benefit.

Why it matters

  • Adoption is rising, but most companies are still at the start. In Bitkom's 2026 survey (603 companies, 20+ staff), 57% use AI, up from 36% a year earlier. Nine in ten say they are only at the beginning [1]. Destatis counted 26% of companies with 10+ employees using AI in 2025, and only 23% of small firms (10–49) [2].
  • Missing use cases and missing data are named barriers. Among companies without AI, 30% say they have not identified any use cases and 40% say they lack the necessary data. Among companies already using AI, 42% name lack of employee acceptance as a problem [1].
  • Picking the wrong first project is expensive. A flashy chatbot on poor data damages trust, and that hurts every project after it. A boring but working automation, such as reading incoming invoices or sorting the shared inbox, builds trust and budget.
  • Bottom-up beats top-down. The people who handle the process every day know where time is lost. If they help choose the use cases, they are far more likely to accept them later (see GUIDE-008).

When you need it

  • Before buying any AI licence beyond a general assistant, or before choosing a vendor (GUIDE-002, GUIDE-010).
  • When individual departments already use AI tools on their own ("shadow AI") and you need to steer that activity (GUIDE-007).
  • Once a year, or after a major system change (new ERP or DMS), to refresh the backlog.
  • Before applying for funding or free support: most programmes expect a concrete project idea (GUIDE-012).

Who is involved

Role Part in the process
Management Sets goals and budget, takes the final decision on priorities
Department leads (process owners) Bring the problems, own the business case, nominate participants
End users Describe the real workflow, pain points and workarounds; test quick wins
IT Checks systems, interfaces, licences you already own (e.g. M365, DATEV, ERP), and effort
Data owners Say where the data lives, how clean it is and who may access it
Works council (if elected) Involve early: §90(1) no. 3 BetrVG requires timely information when work processes are planned with AI; many candidates will trigger §87(1) no. 6 (GUIDE-004)
DPO / IT security Quick screening for personal data, DPIA need and data-leak risks (GUIDE-005, GUIDE-013)

In public bodies the staff council (Personalrat) takes the place of the works council under the BPersVG or the state laws.

Step by step

1. Set the frame (half a day, management + 1–2 leads). Name two or three business goals, for example "cut order processing time", "relieve the back office despite staff shortage" or "faster quotes". Fix a rough budget and a time box. Decide who coordinates. Inform the works council now, not after the pilot.

2. Collect pain points, not AI ideas (1–2 weeks). Ask each department one question: "Which recurring tasks cost you time, cause errors or annoy you?" Use a simple form or a 30-minute round per team. Look specifically for low-level work that nobody calls "AI": - copying data from e-mails or PDFs into the ERP (orders, delivery notes, invoices) - sorting and routing the shared inbox - reformatting Excel exports, merging lists, cleaning customer and article master data - writing standard letters, meeting minutes and summaries - searching for documents and answers across drives

These usually become the quick wins. Also note data preparation tasks (deduplicating master data, scanning paper files, structuring PDFs, tidying file shares). They are often the precondition for every larger project and deserve their own backlog entries.

3. Run a discovery workshop (half a day to one day per area). Participants: 4–8 people from the process (including end users), one person from IT, the coordinator, and a works council member if there is a council. Agenda: 1. 20 min: what current AI can and cannot do, with live demos on your own sample documents 2. 60 min: map one or two processes step by step (sticky notes or whiteboard); mark the time sinks 3. 60 min: turn time sinks into use-case candidates; match them against this catalogue to see typical solutions, effort and stakeholders 4. 30 min: first rough rating (step 4) and open questions

Keep the IHK workshop target in mind: leave with at least one prioritised use case [3].

4. Score every candidate (1–2 hours per batch). Use a 1–5 scale and a simple weighted sum. Keep it in a shared spreadsheet:

Criterion Question Weight
Value Hours saved per month, error reduction, revenue or service effect 30%
Feasibility Can it run with tools you have or a standard product? Integration effort? 20%
Data readiness Is the data available, digital, reasonably clean and accessible? (GUIDE-003) 20%
Risk & compliance Personal data, AI Act risk class, works council relevance, customer impact 15% (scored inversely)
Acceptance Do the affected people want it? Is there a champion in the team? 15%

Write down a rough number for value: "4 people × 3 h per week on order entry" makes discussions concrete. For the risk check, the EU AI Act Service Desk with its Compliance Checker is a useful first filter. It does not replace legal advice [4]. Remember that Annex III high-risk duties (for example AI in hiring) apply from 2 December 2027, and Art. 50 transparency duties have applied since 2 August 2026 (GUIDE-014).

5. Plot and pick (one meeting). Put candidates on a 2×2 matrix of value against effort (feasibility + data readiness). - Quick wins (high value, low effort): start now, usually with existing tools or a small low-code flow (n8n, Power Automate). - Strategic projects (high value, high effort): plan a pilot (GUIDE-009). Often you need a data preparation project first. - Enablers: data clean-up, scanning and DMS structure. Score them by how many other use cases they unlock. - Parked: write down why, and review in six months.

6. Choose the implementation path deliberately. Do not default to buying a SaaS product for everything. For each shortlisted use case, ask: does a feature in software we already own cover it? Is it a small automation we can build ourselves with low-code and an LLM API? Does it need a specialised product? Or does data sensitivity require self-hosting? A healthy portfolio usually mixes all of these (GUIDE-002).

7. Commit and communicate (one week). For every quick win, name an owner, a success metric (e.g. minutes per order), a 4–8 week time box and the people who will test it. Share the full list, including what was parked and why, with all participants and the works council. Then people can see their input was used.

Realistic total effort for a 50–250 employee company: 8–15 person-days spread over 4–6 weeks, plus the workshop time of participants.

Checklist

  • [ ] Business goals and budget frame agreed by management
  • [ ] Works council / staff council informed at the start (if one exists)
  • [ ] Pain points collected from every relevant department, including end users
  • [ ] Low-level automations and data preparation tasks listed explicitly
  • [ ] At least one discovery workshop per core process held
  • [ ] Each candidate scored on value, feasibility, data readiness, risk and acceptance
  • [ ] DPO and IT security screened the shortlist
  • [ ] Implementation path (existing tool / build / buy / self-host) checked per use case
  • [ ] 2–3 quick wins with owner, metric and time box
  • [ ] Results shared with all participants, including parked ideas

Templates & tools

  • This catalogue: filter by complexity: low, time_to_value: days/weeks and data_prep_effort to find quick-win candidates and their typical stakeholders.
  • Mittelstand-Digital Zentren: free, vendor-neutral support, including AI trainers. The current centres end step by step by December 2026. The BMWE announced on 20 October 2026 a new network of six regional centres from 1 January 2027 and eleven specialised centres from May 2027 [5]. Checked 2026-10-31.
  • IHK: free overviews of AI support offers, an AI literacy self-check, and links to the DIHK use-case collection [6].
  • Fraunhofer KI-ULTRA guides (BMAS-funded): problem-driven and technology-driven use-case search with guiding questions, plus a human-centred implementation guide [7].
  • EU AI Act Service Desk / Compliance Checker: first risk screening per use case [4].
  • Spreadsheet scoring template: the table in step 4 is enough. Do not buy a tool for this.

Common pitfalls

  • Starting with the technology. "We need a chatbot" is not a use case. Start with the problem and the hours lost.
  • Only asking managers. They know the strategy, but not the workarounds. Bring in end users from the first workshop.
  • Ignoring data preparation. Many promising ideas fail because the data is on paper, scattered or duplicated. Make data clean-up visible as its own project and give it credit.
  • Overlooking small automations. A 20-step low-code flow that saves an hour a day is not "real AI", but it often pays back faster than any flagship project.
  • Defaulting to buy. Check first what existing licences and simple self-built automations already cover.
  • Telling the works council late. If it hears about a system that logs user activity only at rollout, expect delays. The BAG treats objective suitability for monitoring as enough to trigger co-determination.
  • No metric. Without a baseline (e.g. time per invoice before the pilot) you cannot prove a quick win, and the next budget gets harder to obtain.

Sources

  1. Bitkom: "Erstmals nutzt die Mehrheit der Unternehmen KI", press release, 14 Sep 2026. https://www.bitkom.org/Presse/Presseinformation/Erstmals-nutzt-die-Mehrheit-der-Unternehmen-KI
  2. Destatis: Enterprises using AI technologies, by employment size class (ICT survey 2025). https://www.destatis.de/EN/Themes/Economic-Sectors-Enterprises/Enterprises/ICT-Enterprises-ICT-Sector/Tables/icte-new-1-enterprises-artifical-intelligence.html and https://www.destatis.de/DE/Themen/Branchen-Unternehmen/Unternehmen/IKT-in-Unternehmen-IKT-Branche/_inhalt.html
  3. IHK Weiterbildung: AI use-case workshop for SMEs (workshop goal: at least one prioritised AI use case). https://ihk-weiterbildung.de/anmeldeformular/2038767/
  4. European Commission: AI Act Service Desk and Single Information Platform (Compliance Checker). https://ai-act-service-desk.ec.europa.eu/en and https://digital-strategy.ec.europa.eu/de/news/commission-launches-ai-act-service-desk-and-single-information-platform-support-ai-act
  5. DATEV news on BMWE press release: "BMWE startet Netzwerk Mittelstand-Digital Zentren", 20 Oct 2026. https://www.datev.de/web/de/berufsgruppenuebergreifend/nachrichten/ndsr/wirtschaft/Digitalisierung-Mittelstand-BMWE-startet-Netzwerk-Mittelstand-Digital-Zentren
  6. IHK (ihk.de/ihklw): "Kostenfreie Unterstützung bei der Einführung von KI". https://www.ihk.de/ihklw/produkte/beratung-und-service/ecommerce-und-internet/kuenstliche-intelligenz/kostenfreie-unterstuetzung-bei-der-einfuehrung-von-ki-6396292
  7. Fraunhofer: KI-ULTRA guides on strategy, change and AI project implementation (BMAS-funded). https://websites.fraunhofer.de/ergebnisse-ki-ultra/ki-ultra-leitfaden-auswahl

Accessed 2026-10.