MagicMakersBook an audit
Machine learning consulting and data science

Machine Learning Consulting and Data Science That Reaches Decisions

We turn your data into analysis people can act on: pipelines, statistical modelling and machine learning, delivered inside the product or dashboard where the decision is made.

Book a data scoping callSee what we have shipped

Where machine learning earns its place

Plenty of “machine learning” problems are really data problems. If the numbers in your reports disagree, a model will not fix that. We start by getting one trusted calculation and a pipeline that feeds it.

Then we add the analysis that moves decisions: trend detection, anomaly flags, forecasting and classification, with the statistics to show the result is not noise.

We say plainly when a simpler method will do.

What we build

From the pipeline to the model to the screen where it is used.

Data pipelines and ETL

Scheduled pipelines that pull from your systems into one place, replacing manual exports and hand-built reports.

Analytics dashboards

Dashboards where every number reads from one stored calculation, so reports and screens agree.

Trend and anomaly detection

Statistical tests that separate real change from noise, such as trend tests and anomaly scores.

Forecasting and predictive analytics

Models that estimate demand, cost or risk, with an honest account of how wrong they tend to be.

Model evaluation

Baselines, hold-out testing and error analysis, so you can compare a model against doing nothing or a simple rule.

Models in production

Serving, scheduling and monitoring, so a model keeps working after the notebook is closed.

Analysis you can defend in the room

A number that cannot be explained or reproduced gets ignored. These habits keep ours useful.

  1. 01

    One calculation, one place

    Each metric is computed once, in one stored definition, and every dashboard and report reads from it.

  2. 02

    Statistics, not screenshots

    Trends and anomalies come with the tests that support them, so a change is real or it is flagged as noise.

  3. 03

    A baseline first

    Every model is compared against a simple rule or last period’s number. If it does not beat that, we say so.

  4. 04

    Drift is watched

    Inputs and accuracy are monitored in production, so a model that has stopped fitting the world is noticed.

  5. 05

    Explainable where it matters

    Where a result affects a person or a payment, the reasoning is available, not only the score.

Data work we have delivered

Each sits inside a product people use daily.

GeoTech / data intelligence platformGeoVerdantSix vegetation indices per pixel, Mann-Kendall trend tests and z-score anomalies, served as map tiles. Weeks of analyst work down to minutes.Read case studyEnterprise data infrastructureBusiness HubA PowerBI setup replaced by a web dashboard fed by our own ETL, with a daily AWS Glue dump from Snowflake.Read case studySupport operations consoleMeridian ConsoleDashboard and reports reading one calculation, computed as a Postgres stored procedure, so they stop disagreeing.Read case study

How a data project runs

  1. 01

    Name the decision

    About a week. We agree the decision the analysis must improve and the number that will show it did.

  2. 02

    Fix the data path

    One pipeline and one calculation, so everything downstream is built on numbers that agree.

  3. 03

    Model against a baseline

    Statistical or machine learning methods, tested against a simple rule, with errors analysed.

  4. 04

    Deliver it where it is used

    A first result is typically live in three to four weeks, inside the dashboard or product where the decision happens.

Machine learning and data science, answered

It is working out where machine learning would improve a decision in your business, whether you have the data for it, and what it would take to run in production. It should end with a recommendation, including when the answer is a simpler method.

Often better reporting comes first. If reports disagree or rely on manual exports, fixing the pipeline and the calculation delivers more than a model would. We tell you which you need before building anything.

It depends on the task. Simple statistical methods work on modest data, while some models need far more. We check feasibility against your actual data during the audit rather than guessing.

Data engineering gets reliable data to the right place on a schedule. Data science analyses it and builds models. Good models depend on good pipelines, so we treat them as one job.

Yes. Models can be served behind an API or run on a schedule, with results written into your dashboard or application, so people use them without leaving the tools they already use.

The audit takes about a week and a first result is typically live in three to four weeks. Cost depends on the data, the pipelines and the models. We scope it after the audit, so you know what you are buying before you commit.
Related services
AI Development Company for Vision and NLPAI development for computer vision, language and documents: models measured on your data and built into the software your team uses.Learn moreAI Integration & Automation ServicesConnect AI to your CRM, ERP, billing and messaging so records move without a person in the middle, with monitoring on every automation.Learn moreAI Consulting Services for BusinessAI consulting that ends in a plan: a free systems audit run by an engineer, showing what to automate, what it returns and what to leave alone.Learn more
Read more
How We EngageName the number before you write the codeIf a build has no agreed metric, its success becomes a matter of opinion. Opinion always favours whoever is presenting.Document IntelligenceYour documents are not a knowledge base yetA folder of contracts is not something you can ask questions of. Getting from one to the other is data engineering, not a model choice.Systems IntegrationYour integration layer is a person with two browser tabs openMost businesses already have an integration layer. It is a human being, and they are the least reliable and most expensive part of the stack.

Tell us the decision you wish you had data for.

Thirty minutes with an engineer. You leave with a view on whether it needs a model or just a better pipeline, whether you hire us or not.

Book a data scoping call