On-premise · Mittelstand · Your data stays in-house

First the route,
then the software.

We solve concrete operational decision and process problems — not with a standard product you have to adapt to, and not with a greenfield build from scratch. But with a proven platform as the foundation, tailored to your processes, data and systems in eight to twelve weeks. Together with your business teams, running on your own infrastructure.

Request a call →Live demo
How we workSee Wayfinder

On-premise

First production-ready version in 8–12 weeks

Your data and logic stay yours

Pipelines◦Processes◦Documents◦Control◦Assistant◦Operations◦On-premise◦Pipelines◦Processes◦Documents◦Control◦Assistant◦Operations◦On-premise◦

N 01 — The approach

We close the gap between standard systems and daily operations.

Standard software covers most of it. The rest — precisely the part where your operation differs from every other — ends up in spreadsheets, one-off queries and a few people's heads. That is where we start: at the concrete operational decision, close to the process, connected to your existing systems, running on your hardware.

Where we stand

Not

Standard software you have to adapt your processes to.

Not a finished product that merely gets installed and configured on site. The process defines the requirement — not the licence model.

Not

Custom development on a greenfield site.

Not a project starting from zero — with the budget, the runtime and the risk that experience says come with it.

But

A proven platform as the foundation for your solution.

An existing platform and reusable technical building blocks — adapted to your processes, data and requirements in eight to twelve weeks. Iteratively, in short cycles, together with your business teams.

We connect what is already there — read-only, without touching your core systems. Only several sources together give a reliable heading:

ERP / Inventory

Orders, sales, prices, customers, stock

Production

Capacity, shifts, scrap, bottlenecks

Sales

Promotions, customer behaviour, pipeline, churn signals

External signals

Season, holidays, weather, events, market data

The full product — problem, solution, modules, roles, security →

N 02 — The method

One repeatable method — five steps, eight to twelve weeks.

Every project follows the same pattern. What differs is the business problem — not the path to solving it. First value comes from existing data, not after a system overhaul.

  1. 01Week 1–2

    Problem & target

    Name the concrete operational decision that should get better — and how we will measure it at the end

  2. 02Week 2–4

    Connect data & systems

    Connect ERP, inventory management, production and Microsoft 365 read-only, make data quality visible from day one

  3. 03Week 4–7

    First working version

    No clickable mock-up, no concept paper: a running application on your real data

  4. 04Week 7–11

    Sharpen it together

    Short cycles with the users in the department — whatever does not hold up in daily operations goes back out

  5. 05From week 8–12

    Handover & operations

    Go-live on your infrastructure, monitoring, alerting, maintenance, data processing agreement and SLA — for the long run

What you get is not a concept paper but a system that runs in daily operations and that the department actually trusts. Whatever comes next takes the same path.

N 03 — The platform

Wayfinder — Demand Intelligence

The most fully developed case on our platform: daily forecasts per item, customer and site. Accuracy stated openly instead of a black box. And derived from it, what actually needs doing tomorrow morning — reorder points, safety stock, priorities, alerts.

Wayfinder AI Cockpit

Week 18 · Demand Intelligence Node #01 · Live On-Premise

Total Demand (30D)#01
€ 2.48 M
+8.4% vs. last year
Forecast Accuracy#02
96.4%
MAPE 7.2% · WAPE 4.8%
Service Level (OTIF)#03
98.3%
+1.6 pts vs. last month
Active Forecast Nodes#04
894 SKUs
12 cold-start · 0 outliers

Demand History vs. AI Forecast

Historical demand & 14-day future horizon

AI Model:
Actual Demand AI Forecast (Ensemble (LightGBM + TFT)) 80% Confidence Band (P10-P90)
MAPE: 7.2%
TODAY1.Mai4.Mai8.Mai12.Mai16.Mai20.Mai24.Mai28.Mai+4T+8T+12T+14T

ABC/XYZ Demand Matrix (894 SKUs)

Click any cell for real-time segment filtering

ABC = Revenue share (A: High, B: Medium, C: Low)
XYZ = Demand stability (X: Steady, Y: Variable, Z: Erratic)
ABC \ XYZ
X (Stabil)
Y (Schwankend)
Z (Chaotisch)
A (Umsatz)
B (Umsatz)
C (Umsatz)
01

Forecast per article, customer & location

Hierarchically reconciled (MinT), refreshed daily — not once a quarter.

02

Transparent accuracy

Accuracy shown openly (MAPE/WAPE), with an 80% confidence band and SHAP drivers. Not a black box.

03

Straight to action

The forecast becomes reorder points, safety stock, priorities and alerts for daily operations.

Open the full demo →The product in detail
Cockpit / Control TowerForecast & confidenceSegmentationInventory & orderingCustomer intelligencePurchasingProductionScenariosAI assistant

Every other solution on this page is built on that same foundation — connectivity, data quality, permissions, auditability and operations already exist. That is why a customer-specific solution takes eight to twelve weeks and not two years.

Use case · Manufacturing

Replanning takes as long as planning.

A breakdown, a sick note, a rush order — and the week's planning starts over. We recalculate the plan in seconds, with changeover times, material and due dates, and show what it costs beforehand.

01

Apply the disruption

Line 2 down for eight hours, two operators missing, 400 units by Thursday — in plain words, not as a formula.

02

Recalculate

Backwards from the due date, against changeover time, material availability, shift and qualification.

03

See the price

Which order runs late, what revenue is at risk, what no longer fits at all — before anyone commits.

See dynamic production scheduling →

N 04 — Where we apply it

Same foundation,
different question.

The common denominator is always the same: the gap between what the standard system delivers and what the operation actually needs. Sometimes it is called forecasting, sometimes it is a spreadsheet nobody has dared to touch in six years. Every line here is something we have already built.

Pipelines

Automated data pipelines

We collect your data where it lives, validate it and deliver it processed every night — read-only, without touching your core systems. With traceable origin for every figure, automatic detection of changed source structures, and an alert the moment a run fails. Ready-made connectors:

SAP R/3SAP S/4HANADynamics 365SQLRESTSFTPE-commerceCSV / Excel

in production: multiple sites, nightly run

Processes

Replacing spreadsheets and Power Apps

Requests, approvals, lists and maintenance plans become real web applications — with Microsoft 365 sign-in, an approval chain resolved automatically, reminders until a decision is made, and a complete audit trail.

in production

Documents

Processing paperwork automatically

Orders, delivery notes, promotion sheets and invoices into the system via text recognition and a language model — including ongoing measurement of how accurately the recognition actually performs.

in production

Control

Cockpits for daily operations

Production scheduling, traceability, supplier scoring, open items, complaints — right through to the display on the shop floor screen.

in production

Assistant

An AI assistant on your own documents

Questions to manuals, master data, contracts and reports — answered by a language model running on your server. Nothing leaves the building.

Operations

Operations and accountability

Installation on your hardware, monitoring, alerting on incidents, maintenance windows, data processing agreement and SLA — plus a documented exit, written before anyone needs it.

Business case

Tailored to your company per use case

The value is operationally, financially and organisationally measurable — figures shown are typical industry benchmarks, dependent on project and data.

10–25 %

less capital tied up

through better quantity planning in inventory

+2–5 pts

service level

fewer shortages and ad-hoc escalations

Margin

higher margin

fewer rush orders and scrap, better steering

Days → hrs

faster planning

less spreadsheet reconciliation and crisis meetings

Decisive for success: data access, clear business owners, consistent use of the suggestions in daily operations and transparent accuracy measurement.

N 05 — Principles

What we don't do

Some of this would sell rather well — and would hollow out the very promise you hire us for. Which is why it sits here and not in the small print.

Turn your numbers into an industry benchmark.

→We never pool customer data. What is created in your building stays there — anonymised or not, and however tempting the comparison would be.

Move you into our cloud.

→Everything runs on your infrastructure. If you want the cloud, that is your decision — not our precondition.

Put up artificial licence hurdles.

→Whoever needs the solution in daily operations gets access. No staircase where every second question costs a step more.

Tie you to us.

→Your data, models, configuration and business logic belong to you. We write the handover down before anyone needs it.

Use AI because it says AI on the box.

→Where a rule suffices, we build a rule. A model goes only where it is demonstrably better — and the measurement stays visible.

Take on more than we can operate.

→We are a small team. We say no when a project does not fit what we can stand behind for the long run.

N 06 — From live operations

What is already running

Not pilot projects in a drawer, but systems somebody relies on every single day.

Food processing

From demand forecast into the cutting floor

Mid-sized processor · several production and distribution sites · organically grown ERP and production landscape

Starting point
Demand planning, purchasing and production planning ran on scattered spreadsheets and one-off queries — with no shared data basis and no reliable view of the coming weeks.
Scope
More than 50 functional areas in one application: forecasting, purchasing, slaughter and production planning, traceability, returns, supplier performance.
Delivery
A first production-ready version on real data, then step-by-step expansion in short cycles with the departments.
On-premiseForecast ensembleOrchestrated pipelineCI/CD

Administration

Requests and approvals without Power Apps

Internal administrative processes · Microsoft 365 landscape · multi-level approval chains

Starting point
A Power Apps flow grown over years was neither maintainable nor auditable: approvals stalled and responsibilities were kept by hand.
Scope
A web application with Microsoft 365 sign-in: managers are resolved automatically from the company directory, approval happens by email link or in the app, and every open step sends a daily reminder until it is decided.
Entra IDMicrosoft GraphAutomatic rollback

Platform

The foundation underneath

The technical foundation every project inherits

Starting point
For eight to twelve weeks to be enough for a customer-specific solution, the groundwork cannot be rebuilt every time.
Scope
A multi-tenant base with a dedicated connector per customer, data quality checks, reporting, alerting and permission management — delivered as a self-contained package on customer hardware, including documented offboarding.
Container deliveryMulti-tenantGDPR procedures

These references are anonymised at our customers' request. We name companies, logos and figures in conversation as soon as each release has been granted.

N 07 — Next step

Who needs what — and why?

The goal: decide with confidence whether a usable first version is possible in eight to twelve weeks.

  1. 1Which decisions should get better?
  2. 2Which data sources exist?
  3. 3Who uses the results in daily operations?
  4. 4How do we measure success after 12 weeks?

Talk to us directly

Two sentences are enough: which decision should get better? We'll get back to you personally — no sales machinery.

Write an email →support@wayfinder-ai.de
▸ Or see the live demo first

Personal reply — no newsletter, no mailing list.