DATA SCIENCE / 03

Forecasting & optimisation

Put assumptions to the test.

We assess whether your historical data can provide a useful planning advantage. We compare simple methods with suitable models, test beyond the training data and translate results into a bounded use case. A well-supported decision not to proceed is also a useful outcome.

IN PRACTICE

A question from your working day.

For teams planning demand, inventory or capacity that want to assess a model’s value and the cost of errors before production use.

Plan demand for a defined area

We examine an agreed product group and forecast horizon. A seasonal or other simple baseline shows whether a more complex method provides measurable additional value.

Allocate capacity within real constraints

For a bounded planning workflow, we formulate the objective, available resources and hard constraints. Scenarios reveal trade-offs; your team judges whether a proposed plan makes business sense.

CONCRETE DELIVERABLES

What we agree to deliver.

We specify the extent of these deliverables in your proposal, together with verifiable acceptance criteria.

Objective & evaluation plan
A defined horizon, baseline, error measures and business costs of incorrect decisions. Optimisation work also specifies the objective, constraints and feasible solutions.
Reproducible model comparison
Versioned preparation and model runs. Training, validation and held-out test data are separated appropriately; time-series work respects chronological order and checks for data leakage.
Evaluation & decision brief
A baseline comparison, error and relevant subgroup analysis, uncertainty and applicability limits. With a recommendation to continue, collect more data or stop.
BEFORE WE START

A scope both sides can work with.

What we need from you

  • A specific decision, a planning horizon and a business assessment of errors and constraints.
  • Historical data you are authorised to use, including target values, known changes and relevant influencing factors.
  • Sufficient coverage and a realistic test period; suitability is assessed in the initial step.

What needs a separate agreement

  • No guaranteed accuracy, cost reduction or improvement over simple methods.
  • The prototype is not automatically approved for production. Real-time integration, user interfaces and model operations are scoped separately.
  • Compute budgets, external data, licences and the number of experiments have agreed limits. Retraining is not unlimited.
STEP BY STEP

Reviewable progress. Clear decisions.

  1. Assess value & feasibility

    We clarify the objective, data history, error costs and baseline. If the data is insufficient, we document the missing prerequisites.

  2. Set up comparison & validation

    Data splits are documented. Models are developed using training and validation data; the held-out test data is reserved for the final evaluation.

  3. Review results critically

    We evaluate the agreed measures on the test set, examine failure cases and compare against the baseline. Limits and uncertainty are stated explicitly.

  4. Decide whether to proceed

    Your team receives the code, evaluation and a walkthrough. Any further use has a separate agreement for monitoring, intervention and reassessment criteria.

YOUR TEAM TAKES OVER

The handover is part of the project.

We agree ownership, access and usage rights at the start, including the conditions of third-party components.

  • Source code, environment details and documented data versions for reproducible model runs.
  • An evaluation report covering the baseline, data splits, error analysis and usage limits.
  • Documented model assumptions and a business walkthrough, including potential decision errors.
A SCOPED PROJECT / PRICE ON REQUEST

Start with one concrete question.

Describe the decision, your available sources and who will use the result. We clarify feasibility and propose deliverables, acceptance criteria, schedule and price. Each further stage is agreed explicitly.

A short description is enough for the first enquiry. Please do not send confidential datasets or access credentials through the contact form.

Request a project proposalView shop scope (German)

Project work and ongoing costs are shown separately. Hosting, licences, compute and support are included only when explicitly agreed.

A good place to start.

neurofunken knowledge assistant

Hello! What would you like to improve? I can explain our services and help you find a useful next step.

Replies from our editorial knowledge base.

Please do not enter confidential information. Privacy