Get real work out of AI.

Madlytics is a data and AI consultancy. We help companies decide where AI belongs, build it properly and keep the costs in check.

Where the experience comes from

More than ten years of data science, analytics and AI, in very different kinds of organisation.

Madlytics is Alex Scriven's practice. Alex leads every engagement and draws on a small circle of senior specialists in AI strategy, data engineering and software, bringing them in when a project calls for it.

Large technology companies
Lead data scientist and tech lead on Atlassian's AI data science team, plus work with other large technology companies. Agent workflows, the information agents rely on, experimentation and how results are measured.
Government
Senior data scientist in a large state agency. Machine learning projects from scoping to delivery, analytics for senior decision makers and tools for non-technical teams.
Agencies
Built a software platform for marketing agencies, with a shared company brain and AI agents working from it. Also AI workflows for client delivery and AI cost optimisation.
Social enterprises
Practical AI on tight budgets, including a news hub with AI agents that support advocacy.
Universities
Twelve years of teaching and research. Built and delivered Masters-level courses in machine learning and AI at UTS, and held a data science research fellowship at UNSW.
Startups
Founder, Startmate First Believers investor and mentor, and a moderator of Sydney Startups, a community of 13,000.

About Madlytics

Where are you with AI?

Starting out

A sensible first step

You know AI matters and want to start without wasting a year. We find the work where it will pay off first and get your team using it.

  • Choose where to start, by value
  • Set up the first agents
  • Train the people who will use them
  • Agree how you will judge success
Already running

More value, lower bill

AI is in daily use, the costs are climbing and results vary by team. We work out what is earning its keep and tune the rest.

  • See what you spend and what it returns
  • Cut token costs without losing quality
  • Rework the workflows that are not paying off
  • Give leaders numbers they can trust

A skill runs. A loop learns.

Much of what gets called an agent is a skill: instructions written once that run the same way every time. A skill is one and done. It cannot get better, and as the business around it changes it gets worse. A loop adds the parts that make it learn.

A skill: a request goes in, the skill runs, an output comes out, and nothing comes back to improve it. Request Skill Output

A skill

Written once, then left. Nothing checks it and nothing updates it, so it falls behind as your tools, data and business change.

A loop: do the work, check the result, record the lesson, update the agent, then do the work again. Do the work Check the result Record the lesson Update the agent

A loop

Each run is checked, the lesson is written down, and the next run starts from it. It takes more effort to set up. In return it keeps up with the business and gets better with use.

Where the AI bill goes

Companies that have run agents for a while tend to overspend in the same few places. A review checks each one and tells you what it is costing.

Model choice
Routine steps running on the most expensive model.
Context
Agents re-reading material they do not need.
Caching
The same long instructions paid for in full every time.
Retries
Agents that loop with no rule for when to stop.
Workflow
Many small calls where one planned pass would do.

We teach this, and wrote a book on it

The practical work rests on years of teaching, research and writing about data and AI.

The book

Data Science and Analytics Strategy: An Emergent Design Approach

Kailash Awati and Alex Scriven. Routledge, 2023. A guide to building data capability without betting everything up front, drawing on interviews with data leaders across Asia-Pacific.

See the book at Routledge

12 years
teaching at universities, including Masters-level machine learning and AI
100,000+
enrolments across four DataCamp courses
Hundreds
of colleagues trained in data and AI inside a large technology company
Published
research on AutoML in industry, big data in education and group reasoning

How an engagement runs

Start small, prove it works, then grow what works.

  1. Audit

    We look at how your team works now, what you spend and where AI would help.

  2. Plan

    A short roadmap: priorities, owners, success measures and what it will cost.

  3. Build

    We set up and build alongside your team, one useful piece at a time.

  4. Hand over

    Your people run it. We train them and stay on call while it beds in.

Tell us where you are with AI.

One conversation is usually enough to know whether we can help, and what a sensible first step looks like.

Talk to us