Today
A simple question about customers, products or costs takes an export, a spreadsheet and a day. Whoever asks waits for someone with time to spare. So most questions never get asked.
An example
At Talwerk Antriebstechnik the head of sales asks which of the 50 largest customers ordered less in the last 6 months than in the same period a year earlier. The agent lists 9 customers with the change per product group, as a bar chart. The biggest drop is an agricultural machinery maker buying helical gearboxes: 260 units instead of 410.
Your data, wherever it lives
Revenue, orders and costs sit today in the ERP, customers and opportunities in the CRM, quantities and times in the MES, and the reports on top in databases and countless spreadsheets that someone exports and merges by hand. demi reads, understands and links them into one model of customers, products, orders and costs, with your rights. The AI answers your question in seconds with exact figures and the query behind them, instead of after a day of exports, and you decide what follows.
Where the data lives today
- ERP
- CRM
- MES
- Office files
- Databases
How demi solves it
Three layers in this order: connect the data first, then the rights, then the AI.
- 1 · ConnectData Layer
Data sources in this case: ERP, CRM, MES and spreadsheets. demi replaces nothing and builds from them one model of customers, products, orders and costs. Finance defines once what counts as revenue, order intake or contribution margin, and that holds for everyone.
- 2 · PermitPermission Layer
Rights follow the org chart: each person can only query data they are allowed to see, and salaries or margins stay with the roles that see them today. Every query is recorded.
- 3 · ActIntelligence Layer
The agent turns the question into an exact query, and the result comes from the data, not from the language model. The query sits next to the table or chart so you can check it.
How it runs
- 1
Clarify the data sources: ERP, CRM, MES and spreadsheets
- 2
Define key figures once, such as what counts as revenue
- 3
Ask your question in plain language
- 4
Get figures, a chart and the query behind them
- 5
Drill down, filter or save it as a report
What changes
Answers straight from your systems, no spreadsheet work
Every figure comes with its query
More decisions based on data
What does it bring you?
Check it with your own numbers
Every starting value is an assumption you can change. The calculation sits right below, step by step.
Your numbers
Result
+240 hfreed per year
Assumptions and calculation
demi answers with figures from your data by fixed rules and shows the calculation; you decide.
30 min → 18 min per case
How we calculate
100 data questions × 30 min × 40% × 1.0 (manual) × 12 months = 14,400 min = 240 h
Value per year, before demi costs
Freed per year+240 h
Assumption: labour cost of €49.50 per hour in German manufacturing in 2025 (Destatis), plus workplace, IT and overheads.
= Value per year, before demi costs
+€15,600
gained per year
240 h ÷ 1,760 h per role = 0.1 full-time roles
An estimate, not a promise. The starting values are assumptions for an industrial company with 150 to 500 people.
Combine with more cases and get the packageQuestions about this use case
How do I know the figures are right?
The figures come from a query demi actually runs on your data, not from a language model's memory. Key figures have fixed definitions, for example revenue as the invoiced net amount. If a question is ambiguous, the agent asks a follow-up question instead of guessing.
Can I see how a figure was computed?
Yes. Every answer shows the query, the key figures used and the data timestamp. You can drill down to individual orders or invoices. That way finance can check a figure before it goes into a presentation.
What can each person query?
Exactly what they are allowed to see today. Rights follow your org chart: sales sees its customers, plant management its plant, salaries stay with HR. A question about data outside someone's rights gets no answer containing that data.
What if the ERP data is messy?
demi merges duplicate customers and items into one record and makes gaps visible instead of hiding them. If, say, many orders lack a product group, the agent says so alongside the answer. Fixes happen in the source system, and the answer follows.
Do I need to learn a query language?
No. You ask the way you would ask a colleague. Anyone who wants to check the query still sees it in full. Questions you ask often can be saved as a report that keeps itself up to date.
