Today
The forecast is usually last year plus a feeling. Customer forecasts come through portals and as spreadsheets by e-mail, promotions sit in sales notes, open quotes and new projects in the CRM, and the planner knows the season from experience. It never comes together in one figure that planning and purchasing trust. So too much is ordered, tying up stock and cash, or too little, and the goods are missing exactly when the customer calls them off.
An example
For its 250 g potato salad, demi expects 186,000 tubs at Tarvina for June instead of 142,000 the year before, and says why: a retail customer has posted a barbecue promotion for weeks 24 and 25 in its portal, a new listing with a second retail chain is in the CRM and counted at half, and over the past three years June was on average 9 % above May. Through the recipe this becomes 30 tonnes of potatoes, 8.7 tonnes of rapeseed oil and 186,000 tubs with lids. The planner checks the reasons and releases the figure, and purchasing orders oil and tubs accordingly.
Your data, wherever it lives
Consumption, orders and stock are in the ERP today, customer forecasts in customer portals and as spreadsheets in the mailbox, promotions and open quotes in the CRM and in sales e-mails, seasonal patterns in old planning spreadsheets. demi reads, understands and links all of it, with your rights. The AI calculates a reasoned forecast for every item every month in minutes, and planning decides which figure counts.
Where the data lives today
- ERP
- CRM
- Office files
- E-mails
- Web & portals
How demi solves it
Three layers in this order: connect the data first, then the rights, then the AI.
- 1 · ConnectData Layer
The case needs data from the ERP, the CRM, e-mail, spreadsheets and customer portals. demi replaces nothing and links, per item, the consumption of recent years, open orders, customer forecasts, planned promotions, open quotes and new projects with bills of materials or recipes. Every forecast then also shows which materials, parts and packaging it needs.
- 2 · PermitPermission Layer
Customer forecasts, promotions and quotes are visible to sales, planning and purchasing; individual customers' terms stay with sales. The forecast changes nothing in the ERP: planning releases which figure counts, and only then does it become an order proposal. Who changed which figure and why is logged.
- 3 · ActIntelligence Layer
The figure itself comes from a statistical method on consumption, season and trend; customer forecasts, promotions and quotes, weighted by their likelihood, are added as separate, visible building blocks. The language model only writes the reasons, such as which promotion or which customer forecast lifts the figure, each with its source. If a customer's forecast and call-offs have drifted apart for months, the agent points it out.
How it runs
- 1
demi reads consumption, orders and stock per item from the ERP, plus the season of recent years.
- 2
Customer forecasts from portals and e-mails, promotions and open quotes from the CRM are added.
- 3
The agent calculates the forecast per item and month and notes what lifts or lowers it, with its source.
- 4
Bills of materials and recipes turn it into demand for materials, parts and packaging.
- 5
Planning checks the outliers and releases the figures, and purchasing orders accordingly.
What changes
Forecasts that planning and purchasing trust
Fewer shortages and less excess stock
Every change in the figure with a reason and a source
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
+180 hfreed per year
Assumptions and calculation
demi calculates a reasoned forecast per item from every source; planning checks and releases it.
10 min → 7 min per case
How we calculate
300 item forecasts × 10 min × 30% × 1.0 (manual) × 12 months = 10,800 min = 180 h
Value per year, before demi costs
Freed per year+180 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
+€11,700
gained per year
180 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
Does a language model calculate the forecast?
No. The figure comes from a statistical method on your consumption; customer forecasts, promotions and quotes are added as separate, visible building blocks. The language model only writes the reasons and names the source of every building block.
What does demi do with customer forecasts that are never right?
demi compares every customer forecast with what the customer actually called off afterwards. If a customer has been well above or below for months, the agent shows it and proposes weighting that forecast accordingly. Planning decides whether to adopt it.
How do new items without consumption get into the forecast?
Through similar items that you define or that demi proposes from bill of materials or recipe, pack size and customer, and through the quantities the customer named in the quote or listing. The forecast is then marked as a start-up forecast and gets more precise with every call-off.
Does this replace material planning in the ERP?
No. Material planning in the ERP still calculates order proposals and safety stock. demi gives it a better, reasoned demand figure, and planning releases which figure counts.
