Draft—for design review | August 2026

Commercial forecasting | Example Client AG

AI-enhanced
commercial forecasting

A practical first step for one business unit, with reusable AI capabilities across the group: product vision, delivery path and technical feasibility.

For

Person Six—EVP Business Unit A | Person Seven—CFO

Person Four—Investor | Person Three—VP Group IT

With

Person Five—SVP Sales

Product owner

From

Person Two | Person One

Example Sender GmbH

47.4060° N | 9.7453° E – CAMPUSVÄRE, DORNBIRN, AUSTRIA

01 | 18

Agenda

This hour, four blocks

Building on the discovery and your feedback from July—the business need stands; today is about the how.

01

Product vision

What is being built—and how it looks for its users

02

Approach and pilot

Monthly increments and the shape of the start

03

Data, operations and IT

Where the data lives—and who plays which role

04

Investment and approval

Commercial model and the next steps

Example Client AG—Business Unit A | August 2026

02 | 18

Where we are

From inspiration to a validated use case

Inspiration

AI keynote at the sales meeting

Moonshot cases from the team—forecasting emerged as the obvious first use case

Use case

Defined by the business, not bought off the shelf

Commercial demand forecasting—the known pain of the Excel-based process

Concept

Illustrative demo built and reviewed

AI baseline, prioritised review, reason codes and dashboards—shown to both business units

Sponsorship

Aligned with EVP and CFO

Person Six and Person Seven support the approach in principle

Exchange

The second unit sees value in the AI capabilities

On top of their own tools and initiatives—not as a shared platform

Now

Decision to start

Green light for the pilot—data collection plus the AI intelligence layer

Example Client AG—Business Unit A | August 2026

03 | 18

Starting point

The process stands—the Excel slows it down

Since January 2026 the monthly rolling demand forecast runs over 12 months. Finance takes the numbers—but accuracy varies, and planning in one shared Excel file is the pain.

Production sites

11

Account managers

~50

Product groups

7

Months horizon

12

What already works

  • Monthly cycle established and lived
  • Finance takes the numbers for the financial forecast
  • Fewer ad-hoc requests and analyses

What hurts every month

  • One shared Excel—slow, rows overwritten, all typed by hand
  • Accuracy varies—heavy review and correction effort
  • No automation, no AI, no visualisation

Example Client AG—Business Unit A | August 2026

04 | 18

Target picture

From typing to an AI-supported forecast

Roughly 70 per cent of all customer, product group and site combinations barely move—yet today each one is typed in by hand every month. The assistant generates them automatically and directs attention to the cases that actually need a decision.

01

The AI builds the 12-month forecast automatically—from history, seasonality, budget, YTD and orders on hand.

02

Stable combinations are confirmed instead of typed—and the model learns from every correction.

03

Critical cases are found by anomaly detection—presented with context and reasoning.

Example Client AG—Business Unit A | August 2026

05 | 18

The AI layer

Three layers—only one is reporting

01

Automated data layer

No collecting, no copying, no retyping: actuals, budget, orders on hand and master data flow in automatically and versioned—every month, without anyone’s effort.

02

AI forecast engine

A trained model generates the forecast per combination, detects anomalies on its own and learns from every planner correction. AI-enhanced forecasting—not a formula in a cell.

03

Sales copilot

“Why is Group B in Plant A 12 % below budget?”—the copilot answers in natural language, names the drivers and proposes an adjustment. Instead of pivot tables.

What it is not

An Excel moving into a dashboard. Dashboards are the surface—automation and AI are the product.

Example Client AG—Business Unit A | August 2026

06 | 18

Product vision

Draft—all data illustrative

The account manager’s cockpit

Forecast cockpit—cycle September 2026

  • Account manager
  • RSM
  • SVP Sales

42

Combinations

29

AI forecast accepted

13

For review

69 %

Completion

CustomerProduct groupSiteAI forecast SepΔ Budget
Customer AGroup APlant A840 k SQM+2 %Accepted
Customer BGroup BPlant B410 k SQM±0 %Accept
Customer CGroup BPlant A660 k SQM+18 %Review →
Customer DGroup APlant C290 k SQM−12 %Review →
Customer EGroup CPlant D220 k SQM+6 %Review →

Sales copilot

Why is Customer C / Group B in Plant A 18 % above my forecast?

Orders on hand for Q4 are at 1.180 k SQM—18 % above your last forecast. Driver is a tender win from October; the model flagged the jump as an anomaly.

Suggestion: accept +18 %, reason code Tender.

  • → Accept
  • Adjust
  • Each role its own view—AM, RSM, SVP
  • Confirm instead of type, reason codes instead of free text
  • Export as one Excel—any time

Example Client AG—Business Unit A | August 2026

07 | 18

Product vision

The view for management and finance

Volume, deviations, accuracy and bias per site, region and planner—and, for the first time, one consolidated orders-on-hand view across all eleven sites.

Forecast FY—m SQM

74,6

Δ Budget

+2,1 %

Accuracy—MAPE

11,8 %

Bias

+1,4 %

Two views

  • Volume per site and quarter—Δ budget, Δ prior year, full year
  • After completion: volume, revenue and CM1 vs. budget and prior year

Drill-down and export

  • Business segment | customer group | ship-to | product group
  • Export as one Excel—any time. Today: no consolidated view exists

All figures illustrative

08 | 18

Architecture

Separate applications, shared AI capability

Business Unit A workflow

Dedicated demand-forecasting assistant

A new working process for account managers, built for the unit’s granularity and planning rhythm.

Protected AI intelligence layer

Proprietary methodology and IP

Internals stay protected. Outputs stay transparent: it flags an issue, shows the evidence, suggests an action—users confirm, adjust or explain.

Other workflows

AI enhancements on existing tools

Forecasting in the second unit today, a future S&OP environment tomorrow—existing tools stay, AI capabilities plug in via AI-readable interfaces.

Inputs may include

Approved historical data | existing forecasts | actuals | orders on hand | pricing | contextual business information

Outputs may include

Anomaly detection | quality checks | suggested baselines | prioritised review | explanations | natural-language interaction | AI-readable interfaces

Conceptual—architecture and data sources not yet agreed

09 | 18

AI Check

The AI Check can sit on top of existing systems

An independent, low-friction capability that analyses existing forecasts and inputs—no new forecasting platform required.

AI Check | possible functions

  • Missing-input and deadline alerts
  • Anomaly and bias detection
  • Challenge implausible forecasts
  • Compare with history, actuals and context
  • Surface outliers, risks and opportunities
  • Prioritised review queue
  • Existing forecasting tools
  • Existing planning processes
  • Approved data access

It can support other systems and workflows where approved data access exists—the second unit’s current tools as well as a future S&OP environment—without requiring anyone to adopt this forecasting process.

Example Client AG—Business Unit A | August 2026

10 | 18

Success

What we measure success by

The levers are time and usability—and after the last cycle, explicitly: improving accuracy.

Primary—time

Hours per cycle

Baseline in the first cycle, then a clear downward trend—the most important number.

Primary—relief

~70 % automatic

Share of combinations that arrive pre-filled and only need confirming.

Primary—adoption

Cockpit over Excel

A short user survey per cycle, active use of the pre-filled forecast.

Accuracy—measurably better

MAPE and bias per site, region and planner | overrides → 0 | completion per cycle | reason codes on critical cases. Measurement is part of every acceptance—a success dashboard shows each cycle whether the investment carries itself.

Example Client AG—Business Unit A | August 2026

11 | 18

Delivery

A usable increment every month

No big bang: every stage goes end-to-end and delivers real value inside the live forecast cycle. You set the order—and you can stop after any cycle.

M1

The Excel replaced

Today’s process as a real application in the live cycle—scoped views, versioned. First global orders-on-hand view.

M2

Pre-fill and alerts

Explainable rules pre-fill the forecast. Contradictions are flagged automatically.

M3

Review per click

One-click accept for stable cases. Reason codes and completion display.

M4

Dashboards and accuracy

Views for region and management. MAPE and bias per site and planner.

M5

The assistant leads

Guidance through the critical cases. Explanations and summaries per cycle.

Explicitly excluded for now

A full S&OP platform—the running S&OP initiative stays untouched | supply and capacity planning | forced group-wide standardisation | a big-bang replacement programme

Built to fit the IT landscape

Connects to existing systems and approved data access | site-level access rights, EU data residency where required | monthly stop-or-go keeps the risk small

Example Client AG—Business Unit A | August 2026

12 | 18

Pilot

One plant in depth, all plants in view

P&L and data live at plant level—but an isolated plant pilot would create two parallel process worlds. This cut avoids exactly that.

In depth—one pilot plant

  • Account managers work the real cycle in the cockpit
  • End-to-end: data capture, forecast, review
  • Along the plant’s P&L and data structure—Plant A is the proposal

In breadth—all 11 sites from day one

  • Central consolidation and dashboards across all plants
  • Excel deliveries simply keep flowing in at first
  • Orders on hand visible across plants | automatic reminders

No duplicate process emerges: the very first stage removes work from everyone involved—instead of adding it.

Example Client AG—Business Unit A | August 2026

13 | 18

Data and operations

The data stays in your system world

Sources—at the client

Forecast Excel

Controlling exports—PY, budget, actuals

Orders on hand from the plants

Build and pilot—EU

EU data centre

Only necessary exports—under NDA and DPA

No training on your data

Production—your environment

Your own cloud and dashboard landscape

Data sovereignty fully with the client

Handover to IT—or continued operation

Variant Sovereign—EU-sovereign

Open or non-US models in the EU data centre. No US access, maximum data sovereignty.

Variant Frontier—US models, EU region

Leading models via a hyperscaler, processing in the EU. Maximum model quality.

The choice is yours—both variants without training on your data

Example Client AG—Business Unit A | August 2026

14 | 18

Internal IT

Internal IT as a partner

Involved early instead of confronted late—with a clear split of roles and a dedicated technical session as the next step.

The external team delivers

Product: forecast logic, cockpit, dashboards

Support through the live cycles

Operation of the pilot

Together

Architecture and data model

Security and governance requirements

The path to production

Internal IT

System access and exports—controlling, plants

Constraints and technical sign-off

Target infrastructure, optionally later operation

Why external? The client is not buying a software product, but:

Speed | AI engineering and prototyping expertise | a low-risk pilot—stoppable monthly, ready for handover into your environment, the IT roadmap stays untouched

Example Client AG—Business Unit A | August 2026

15 | 18

Investment

Transparent, modular, stoppable at any time

Entry—module 1, first cycle

24.000 €

Around 80 hours: today’s forecast as a real, role-based application—usable in the next cycle. Then you decide.

Complete pilot—corridor

90–115 T€

Roughly 350–450 hours across four modules, time and material—no fixed fee, you steer scope and pace.

Hourly rates—net

Standard 300 € | from 100 h 260 € | from 250 h 240 €—marginally staggered above each threshold.

Billing model negotiable

T&M is our proposal, not the condition. Also possible: 70 % on commissioning, rest on acceptance—or a base plus a fee per connected plant.

For context

Classic projects of comparable depth run three to five times higher—with the first usable result only after months.

Only hours worked | stop after each cycle | prior work credited on commissioning

Example Client AG—Business Unit A | August 2026

16 | 18

Team

Senior delivery, embedded in a network

Partnership, strategy and contract

Person Two

Digital and AI strategist, innovation studio—work for international consumer brands. Sparked this initiative with the keynote at the sales meeting.

Offer, contract and invoicing: Example Sender GmbH

Delivery and responsibility

Person One

Co-founder and CTO of a data company, previously CTO of a social network. Systems in production at several industrial groups.

AI-assisted engineering—weeks instead of months

Project steering and continuity

Person Eight

Senior project manager—around 20 years of automated data solutions.

Forecasting and controlling methodology

Person Nine

Finance business partner, controlling and auditing.

No single-person risk—cover and depth of expertise are secured

Example Client AG—Business Unit A | August 2026

17 | 18

Decision

Start now.
Keep the AI layer open for reuse.

Action A

Green light for the start.

Begin structured data collection—needed regardless of any tool decision—and integrate the AI intelligence layer. Monthly increments, stop-or-go after each cycle.

Action B

Keep the reuse path open.

Explore where the AI layer adds value elsewhere—the second unit’s forecasting, the S&OP initiative—through a short, structured discovery. Not a prerequisite.

This week: technical session with IT | confirm the pilot cut | prepare NDA and sample data

Let’s Talk

47.4060° N | 9.7453° E – DORNBIRN, CAMPUSVÄRE

18 | 18