The Builds — data & machine learning

Your data is sitting idle. We put it to work.

Putting a company’s data to work starts simple: reliable dashboards, forecasting, anomaly detection. Machine learning only comes in where rules no longer suffice.

What we build with your data

Sales, purchasing, production, tickets: you already own years of data. Four concrete uses, in increasing order of complexity.

Reliable dashboards

Figures everyone accepts because they’re traceable back to the source. Often the most profitable build — and the least often sold.

Forecasting: sales, stock, workload

Anticipate demand, size inventory, plan team workload from your own history. With an honest evaluation protocol: we measure the error before we promise anything.

Anomaly detection

Unusual invoices, consumption drift, production deviations: spotting what a human no longer sees in the volume.

Compliant RAG on your documents

Query your internal documents in natural language, with answers traced back to their sources — hosted under your jurisdiction, not in a public LLM.

Proof before promise

House rule: machine learning only where rules no longer suffice. When an explicit business rule does the job, we deliver the rule — cheaper, more legible, easier to audit.

A typical data register and dashboard

An excerpt from a typical data register — where each data point originates, its quality, who consumes it, what it would enable — and the resulting dashboard, every figure traceable back to the source.

Data register: sources, quality, possible uses
Documented dashboard, traceable figures
Predictive model with an evaluation protocol and error measurement
GDPR processing record (legal bases, minimisation)

Illustrative example — not a client document

What a typical engagement looks like

1. Data assessment

An inventory of what you own: sources, quality, available history. This is the data workstream of The Baseline — no reliable data, no reliable model.

2. Prioritised use case

We pick the case where data pays off fastest, with its AI Act classification and GDPR framework set before a single line of code is written.

3. Build and honest evaluation

Rules first, ML if needed. Every model is evaluated against reality, with an error measurement written down in black and white.

4. Deployment and monitoring

A model degrades over time: drift supervision, scoped retraining. In-house, or through The Engine Room, under SLA.

And all of it passes the audit

GDPR by design

Legal bases identified, data minimised, data subject rights tooled. The GDPR processing file is a deliverable, not an after-the-fact fix.

Compliant RAG, a reference architecture

Your documents stay under your jurisdiction: controlled ingestion, source traceability in every answer, no silent traffic to a public LLM.

AI Act classification of the use case

Every use case is classified before it’s built. High-risk system obligations apply from 2 August 2026; AI literacy (Article 4) since 2 February 2025.

Frequently asked questions

Do we have enough data?

Often, yes, without knowing it: a few years of invoicing or sales data is enough for a first useful forecast. The data assessment answers that question precisely — and if your data isn’t enough, we tell you in writing rather than build a shaky model.

Isn’t ChatGPT enough?

For drafting text, often yes. For working with your internal documents and data, a public LLM raises confidentiality and jurisdiction questions your auditor will ask too. A compliant RAG, hosted under your jurisdiction, gives answers traced back to your sources — and passes the audit.

Do we need machine learning everywhere?

No. Our rule: ML only where rules no longer suffice. An explicit business rule is cheaper, more legible and easier to audit than a model. We’ll tell you honestly which case you’re in.

How long before a first result?

A reliable dashboard ships within a few weeks. A predictive model depends on the state of your data — which is exactly what The Baseline measures. Either way: a fixed-scope engagement, quote within 48 hours.

Your data is worth more than it currently earns

Start with a free Express Scan: 90 minutes with an engineer to spot where your data pays off — and where it won’t.

Request an Express Scan See the method