Fractz

What we do

Predictive Analytics

Forecast sales and supply chain with custom ML models.

Talk to an engineer

What we do

Forecasting models built on your own history, delivered into the system where the decision actually gets made. We start by building the dumbest possible forecast and only ship a model that beats it — a surprising number of projects fail that test.

  • Benchmarked against a naive forecast
  • Drift monitoring from day one
  • Delivered into the tool you already use

You probably need this if

What we build

01

A data audit, first

We tell you whether your data can support a useful model before you spend on building one. Sometimes the honest answer is not yet, and that answer is cheaper now.

02

A benchmark you have to beat

Last period plus trend is the baseline. A model that cannot beat it is not worth operating, and knowing that early saves the budget.

03

A production model with monitoring

Accuracy tracked against actuals over time, with alerting on drift rather than an annual rediscovery that it stopped working.

04

Delivery into the decision point

Forecasts land in the ERP, planning sheet or dashboard your team already opens. A model behind a separate login does not change decisions.

Typical stack

  • Python
  • scikit-learn
  • LightGBM
  • XGBoost
  • dbt
  • Postgres
  • BigQuery
  • your ERP / BI tool

How an engagement runs

  1. Weeks 1–2

    Data audit and baseline

    What you have, what it can support, and the naive benchmark to beat. Ends with a go or no-go.

  2. Weeks 3–7

    Model and validation

    Feature work, model selection and backtesting against held-out periods.

  3. Week 8+

    Integration and monitoring

    Deployment into the decision tool, accuracy tracking and a retraining schedule.

Process

  1. 01

    Discovery & Audit

    Goal analysis, process mapping, and KPIs.

  2. 02

    Tailored build

    Rapid development with biweekly sprints.

  3. 03

    Deploy & Scale

    Monitoring, training, and continuous optimization.

Engagements

  1. 01

    Fashion manufacturing

    Product operations for a leather goods maker

    Pellemoda S.r.l.

    A production platform the marketing team runs without engineering support

    System
    Digital product platform + operations automation
    Duration
    Multi-phase
    Read the engagement
  2. 02

    Education

    Marketing automation for an international college

    H-Farm College

    Manual marketing tasks automated; team time moved back to campaigns

    System
    Marketing automation + reporting
    Duration
    Multi-phase
    Read the engagement

Common questions

How much history do we need?
Usually two to three full seasonal cycles for anything seasonal. We assess this in the data audit and tell you before you commit to a build.
What accuracy can you promise?
None, before seeing the data — and be careful with anyone who does. We commit to beating a documented naive benchmark, and we report the number honestly either way.
What if accuracy degrades?
It will. Drift monitoring and a retraining schedule are part of the build, not a later upgrade.
Where does it run?
EU-resident infrastructure by default, with GDPR obligations designed into the pipeline rather than reviewed after it.

Services