Sheet P-01 / Data products: datasets, reports and models

From trusted data to daily decisions.

Decisions are only as good as the numbers in front of you. We build datasets, reports and machine learning models as products: owned, used, and made to work together. The result is one version of every number, answers that arrive faster each time, and nothing that depends on one person.

FIG. 01 One column of the Maison Carrée, the module: photographed left of the section line, drawn right, with its neighbours at the spacing the building uses. M is the diameter at the foot; Vitruvius set out the whole front from that one measure, so every part fits every other. Photograph Conceptuel, Wikimedia Commons, CC0 (public domain).
Sec. 01 / Where decisions stall

Less time on which number is right, more on what to do about it.

Most data is a by-product. A system wrote it down and nobody owned it after that, so every report is a one-off request with its own definition of margin, and the forecast runs on a copy of a copy. Three teams bring three numbers to the same meeting and the meeting goes to which one is right. Treating data as a product changes that. Each dataset, report and model has an owner, a user, and definitions the next product reuses. The first thing you notice is that the numbers agree. The second is that each new answer comes faster and costs less than the last.

One of these parts will be the module, and the thickness of the columns will be equal to one module.
Vitruvius, De architectura, Book III
AFTER VITRUVIUS, BOOK IIIM2¼ M3 M18 MONE MODULE, REPEATEDFig. 02 / The module
Sec. 02 / What we build

Three kinds of product that work as one.

Datasets

The facts of your business, defined once and kept current: customers, orders, contracts, assets. Documented, owned, available to every team and tool. Finance, sales and operations start from the same facts, and nobody rebuilds the customer list for their own report.

Reports and metrics

Management and operational reporting on one agreed set of definitions. Revenue is revenue in every report and the board pack reconciles with the floor. The meeting is about the decision, not the number.

Machine learning products

Forecasts, scores and risk flags trained on the same datasets your reports use, delivered inside the systems where people act: the planning tool, the CRM, the ordering screen. Measured on the outcome they were built to improve, so you know what they deliver.

The standard

Shared identifiers, agreed definitions and common formats, written down and enforced, so every product works with every other without a new project. This is where the return compounds: the second product costs less than the first, and the tenth is cheap.

Parts list, sheet P-01.
Sec. 03 / How we work

Every product ends in a decision.

01 / Base

Define

Which decision, who makes it, who owns the product. We write that down before any tool is opened and agree the definitions with the people who will act on the answer.

02 / Shaft

Build

Dataset, report or model, built on your platform to the shared standard, tested against real outcomes, documented, and reviewed with its users before it goes live.

03 / Capital

Embed

Into the meeting, the workflow or the system where the decision is made, with monitoring and a named owner. Your team is trained to run it, and we step back.

Fig. 04 / Elevation
Sec. 04 / Why datalyft

What this buys you.

Numbers you can govern by.

One owner and one definition per product, so the board, finance and operations read the same figure and the discussion moves to what to do about it.

Returns that compound.

Products built to one standard are reused, not rebuilt. Each new report, dataset or model draws on the ones before it, so the cost per answer falls as the portfolio grows.

Less risk, less dependence.

Ownership, documentation and service levels mean no product lives on one laptop or in one supplier's head, datalyft included. Your team runs it after hand-over.

Power BI · Microsoft Fabric · dbt · Python · scikit-learn · MLflow

Fig. 05 / Setting out

North found with a gnomon, after Vitruvius, Book I: the morning shadow at B, the afternoon shadow at C, arcs from both crossing at D, and the meridian through D and the centre. Weesp, 52.31° N · 5.04° E.

Sec. 05 / Start

Which number would you most like to stop arguing about?

A 30-minute conversation about the reports, datasets and models you have, which of them should become products, and which one to start with. No obligation.