Prompt Library

Finds 8–12 problems in everyday work and assigns fixed IDs (P1–P12).

You are an experienced process and efficiency consultant. Help me find the real PROBLEMS in my company – the concrete pain points in everyday work that AI could later address.

My company in 2 sentences:
[Briefly describe: what do you do, who do you sell to, how big are you?]

Our most important systems and data sources (optional):
[e.g. CRM (Salesforce), ERP (SAP), Outlook, shared network drive, accounting software, ticketing system. Name only what you actually use – the solution proposals later build on it. If you are not sure, leave this field empty.]

Problems we are already aware of (optional):
[List problems that already come to mind – one per line. If you know how often something occurs and how long it takes, add it (e.g. “CRM data entry, daily approx. 30 min”). The AI will pick these up, sharpen them and add more. Leave this field empty if nothing comes to mind yet.]

Find 8 to 12 concrete problems we struggle with daily or weekly – taking the problems I listed above into account as well. Real, specific problems – not abstract goals. Use these 3 lenses:

1. Recurring process steps/activities that regularly cost time.
   Example: entering customer data into the CRM.
2. Cross-departmental, rule-based processes and activities involving many documents or a lot of research and searching.
   Example: approval of all expenses >€1,000 by purchasing & accounting; checking contracts for standard clauses.
3. Large volumes of unstructured data that are currently used only partially or manually.
   Example: historical customer emails.

While searching, also think broadly about these kinds of AI potential – as a search space, NOT as the structure of your answer:
Searching and finding knowledge · summarising and structuring information · drafting texts, quotes, reports, emails · checking documents, data or entries · classifying, routing or preparing cases · spotting patterns in feedback, tickets, emails or reports · preparing decisions through upfront analysis.

But always name only the PROBLEM, never the capability – otherwise the solution is already sitting in the problem statement.
Wrong: “lack of automatic classification of incoming tickets”.
Right: “incoming tickets are assigned to the right department by hand”.

Output as a clean list with IDs – one short, precise label per problem (max. 12 words, e.g. “manual effort of CRM data entry”), no solutions:

P1. [problem one]
P2. [problem two — daily, approx. 30 min]   ← time figures only if I stated them myself above
P3. ...

IMPORTANT:
- Give every problem an ID in the format P1, P2, P3 ... (P1 to P12). The ID is fixed – we reuse it in all later steps.
- Carry over frequency and duration ONLY if I stated them above – then unchanged. Never estimate yourself how often something occurs or how long it takes. Without input from me, the time figure is simply left out.
- NO market, HR or leadership topics (skills shortage, price pressure, competition, motivation). Only activities that someone here actually carries out day to day.
- Do NOT propose any AI use cases or solutions yet. Only the problems.
- Entire list max. ~200 words.

First step in an AI tool of your choice (ChatGPT, Claude, Gemini …) — ideally run all four prompts one after another in the same chat. Collect problems only here, no solutions yet. The IDs (P1, P2 …) carry through every step.

Finds up to 8 opportunities using generative and agentic AI, grouped by area (O1–O8).

You are an innovation strategist focused on generative AND agentic AI in our industry. Think opportunity- and potential-driven.

You already know our company from step #1 – do not ask about it again.

Additional information, only if it was not stated there:
[Markets/regions and what we are particularly strong at. Leave empty if it was already covered in step #1. Only if you are working in a new chat: paste the company description from step #1 here.]

For a company like ours, answer:

Which concrete AI use cases have the greatest leverage in our industry – separated into production, sales, service and administration?
Take into account what creates real competitive advantage and what competitors or comparable industries are already doing – but do not list that separately.

Think in TWO stages and exhaust both:
- Generative: the AI produces or processes a single thing on request – a text, a summary, an analysis, an assessment. One step, triggered by a human every time.
  Example: drafting a reply to a complaint email.
- Agentic: the AI handles a multi-step process itself – it is set off by an event, gathers the information it needs from several sources, decides between paths, executes steps in our systems and presents the result to a human for approval.
  Example: responds to incoming complaints on its own – matches them to the order, pulls delivery and inspection data, assesses whether the claim is justified, drafts the reply including a goodwill proposal and submits it to service for approval.

At least 3 of the opportunities must be agentic. For agentic opportunities always name the trigger, the steps and the point at which a human approves – “an agent” on its own is not an opportunity.

Output:
- Group the use cases by the four areas (production / sales / service / administration).
- MAXIMUM 2 opportunities per area – so at most 8 in total (O1–O8). Better 1 strong than 2 weak ones. Take only those with the greatest leverage.
- Give every opportunity an ID in the format O1, O2, O3 ... (we reuse these in step #3).
- Phrase every opportunity directly as a concrete AI solution idea (what the AI would do) – it will later be carried over without an associated problem.
- Mark the top opportunity with the greatest leverage in each area with ⭐.
- 1–2 sentences per use case (up to 3 for agentic ones, so that trigger, steps and approval fit in), one of them containing a concrete example in brackets: which data, which system, which trigger (e.g. “reads incoming complaint emails and proposes a reply including a goodwill amount based on the last 200 resolved cases”).
- NO general trend or technology lists. Every opportunity must connect to a concrete work situation in a company like ours – to an activity someone there actually performs.
- You do not know our day-to-day. If you derive an opportunity from the industry, state the assumption in the sentence (“If complaints are still assigned by hand at your company: …”) – and then still be just as concrete. Better an openly stated assumption than a claim that is not true.
- No generic phrasing such as “AI-supported optimisation of X”. Always name WHAT the AI reads and WHAT it produces.
- No introduction, no summary, no further lists.

Second step in the same chat: look ahead — where do generative AND agentic AI offer the greatest leverage? At least 3 opportunities must be agentic. The IDs (O1, O2 …) are reused in step 3.

Merges steps 1 and 2 into one copyable table, max. 20 rows.

You now have two inputs. Bring them into ONE combined table with three columns: # | Problem | AI solution idea.

Input A – Our problems (from step #1):
[Paste the numbered problem list from step #1 here]

Input B – Our opportunities (from step #2):
[Paste the opportunities O1, O2, O3 ... from step #2 here]

Input C – Our systems and data sources (optional):
[The same systems as in step #1, e.g. CRM (Salesforce), ERP (SAP), Outlook, shared network drive. In the examples below use ONLY systems from this list. If it is empty, phrase the examples system-neutrally (“from the incoming enquiry emails”) and do not invent system names.]

STEP 1 – Merging (do this FIRST, before you build the table):
Compare every opportunity from input B with every problem from input A. Two entries BELONG TOGETHER if they concern the same activity, the same process step or the same document – even if they are worded differently.
→ They then become ONE row: problem from A, solution idea from B, ID combined as “P3+O4”. Never two rows for the same thing.
Then also compare the opportunities with each other and merge those that mean the same thing (“O2+O5”).

STEP 2 – Add the rest:
- Remaining problems: enter the problem (ID P1, P2 ...) and formulate a fitting AI solution under “AI solution idea”.
- Remaining opportunities: enter the opportunity under “AI solution idea” (ID O1, O2 ...), leave the “Problem” column EMPTY.

STEP 3 – Trimming (only if necessary):
Count the rows AFTER the merging in step 1. If there are 20 or fewer, the table is finished – there is nothing left to do here.
If there are more than 20, delete the surplus rows outright. Do not merge any further – merging was step 1, here things are only left out.
Delete first what adds the least: rare one-off cases, rows with no recognisable time saving, and rows for which we do not hold the necessary data at all.

QUALITY OF THE AI SOLUTION IDEA – the most important part:
Every solution idea has two parts:
(a) What the AI concretely does: input → processing → output. Not just the topic. Max. 25 words – for multi-step, agentic solutions max. 35, but then including trigger and approval point.
(b) A concrete example in brackets: which data source, which document, which system, which trigger. If you have a good, specific idea – name it. Better concrete and open to challenge than general and indisputably right.

Weak: “AI supports quote creation.”
Good: “Generates a draft quote for approval from the enquiry email, the price list and the last 3 quotes sent to the same customer (e.g. Outlook enquiry → draft in the CRM, the engineer only checks the line items).”

FORBIDDEN, because too shallow: “AI-supported automation of X”, “chatbot for X”, “AI agent for X”, “autonomous agent takes over X”, “AI supports X”, “optimisation through AI”, “data analysis with AI”, “intelligent processing of X”.
If a row sounds like that, rewrite it: WHAT does the AI read, WHAT does it do with it, WHAT comes out at the end?

If you cannot think of anything more concrete for a row than the obvious standard solution, mark it with ❓ instead of dressing it up.

Output the result as a clean Markdown table:

| # | Problem | AI solution idea |
|---|---------|------------------|
| P1 | manual effort of CRM data entry | [what the AI reads, does, produces] (e.g. ...) |
| P3+O4 | manual quote creation | [merged row from problem P3 and opportunity O4] |
| O1 |  | [opportunity from #2 as an AI solution idea] (e.g. ...) |

RULES:
- Every row is unique in substance. If while writing you notice that two rows mean the same thing, you missed it in STEP 1 – merge them there retroactively instead of leaving both in.
- “Problem” column: only real problems as a short, precise label (e.g. “manual effort of CRM data entry”) – no solutions.
- If a problem from input A carries a time figure (e.g. “daily, approx. 30 min”), carry it over unchanged. Do not invent frequencies or durations yourself and do not add any where there are none.
- Opportunity rows with no associated problem: leave the “Problem” column EMPTY.
- Output ONLY the table, with no further text before or after it.

Third step: paste in the results from steps 1 and 2. The result is a table (# | Problem | AI solution idea) with at most 20 rows.

Strips anything confidential from the table — ready to submit for the joint evaluation.

Take the finished table from the previous step (# | Problem | AI solution idea) and make it ready for submission to a joint evaluation together with other companies.

FIRST CHECK EVERY CELL FOR CONFIDENTIAL INFORMATION and remove or replace it:
- Company names (ours and other people’s) → replace with the industry, e.g. “a precision machinery manufacturer”, “a major customer from the automotive industry”.
- Names of people, teams, department heads, customers, suppliers → delete or replace with the role (“the sales assistant”).
- Specific figures that allow conclusions to be drawn: revenues, prices, margins, unit numbers, headcount, contract values → delete or round to a rough order of magnitude.
- Product names, project names, internal code names, locations, customer numbers, contract numbers, file paths, email addresses, links.
- Anything recognisably secret, security-relevant or legally sensitive (ongoing legal cases, complaints naming individuals, HR data).
- System names such as CRM, ERP, Outlook, SAP may stay – they are not confidential.

Do NOT change the substance: the problem and the AI solution idea must remain understandable and concrete for outsiders. Anonymising means replacing, not leaving out. “Quotes for Müller GmbH take too long” becomes “quotes for existing customers take too long” – not “quotes take too long”.

Then output the cleaned table in exactly the same structure as before:

| # | Problem | AI solution idea |

No other text – no introduction, no summary, no list of the changes you made.

Last step before submitting: remove everything confidential from the table. Proofread the result yourself — anonymising means replacing, not omitting. Then submit the cleaned table with your workshop code.