Stephen Bartlett — AI Systems Consultant

AI Systems Consultant

Specialist across research, system design, and implementation, focusing on integrating AI into business workflows to deliver measurable, safe, secure, and long-lasting productivity gains.

30 minutes. No pitch.

Diagram: several data sources feed a central model, which produces a single measured output.

The questions

Capability alone doesn’t establish value.

AI is capable. Whether it creates value depends on the conditions around it: the information it has, who holds authority, what it costs to keep it true, and whose definition of improvement is being used. These are the questions I start with.

95% of organisations reported no measurable financial return. MIT Project NANDA, The GenAI Divide: State of AI in Business 2025. Preliminary findings from 300+ public initiatives, 52 interviews, and 153 leaders; ROI measured six months after pilot.

Resolution
Does the information preserve what matters for this decision, and can the system gather what’s missing?
Authority
Who shapes the choices, who decides, and can that influence be traced?
Learning
Will the system change because of what it encounters, or only accumulate more data?
Cost
Do the gains hold once operation, data gathering, and consequences over time are counted?
Value
Whose priorities define improvement, and how are competing interests weighted?

How I work

Research, design, implementation

I can take an engagement end to end, or work on the part you need help with.

Research

  • Map workflows, employee knowledge, and friction.
  • Assess data resolution, gaps, and collection costs.
  • Identify authority drift and competing priorities.
  • Define outcomes and measures of success.

Design

  • Determine where AI adds value.
  • Define human, AI, and conventional technology boundaries.
  • Build in traceability, accountability, and compliance.
  • Make decisions and system behaviour understandable.

Implementation

  • Lead teams from development through production.
  • Map delivery directly to agreed outcomes.
  • Validate against real workflows and consequential exceptions.
  • Measure gains, costs, and employee friction.

Track record

Evidence

Reported outcomes, and the delivery behind them.

$1M+

First-year benefit reported by Warner from its multi-location platform.

30%

First-year increase in room usage reported by Belmont University.

End-to-end delivery. Founder and CTO of Studio Assistant. Led an eight-person team across client discovery, system architecture, backend and frontend development, and Azure and Kubernetes production operations, delivering enterprise systems for Warner, Belmont University, and the world’s largest streaming platform.

Selected work

Selected systems

Human-in-the-loop AI copilot
Turns real-world session activity into structured data proposals, with supporting evidence, human approval, and a record of every change.
Research and evaluation platform
Explicit criteria, weights, and scoring rules, with the evidence behind every result. It has been used for acquisition screening and enterprise research.
Real-time clinical music system
Combines live gait data and therapist input to generate patient-specific audio in real time, with therapist override. Winner of the 2017 Intelligent Health Award.
Stephen Bartlett speaking at an event.

Background

About

I’ve worked across music production, clinical systems, and enterprise software. Across those settings, my work has involved understanding how the parts interact, working with people who bring different priorities, and taking responsibility for the result.

I spent more than fifteen years as a producer, engineer, and composer, with multiple top-ten records across eleven countries. Much of that work happened at a large analogue console, managing dozens of inputs, hardware processors, gain stages, people, egos, and automation while the session continued around me. You need an accurate mental model of the whole signal path, because a change in one place affects what happens elsewhere. You also need to understand what the musicians are trying to achieve. Everyone may want their part louder, and you can guess how that sounds.

In large organisations, too, everyone is trying to hear their part. An outside perspective can help identify what needs more attention, what needs turning down, and when the arrangement, or the system itself, needs to change. Maintaining that perspective means understanding the different priorities and making deliberate decisions about the trade-offs.

In 2016, I began building a clinical music system with therapists and doctors. It combined live gait data with therapist input to generate patient-specific audio in real time. The design preserved control over patient data and gave therapists the ability to override the system, requiring a thorough understanding of authority bounds. This brought together technical design and clinical judgment, so the person treating the patient could respond to circumstances the system could not fully capture. The project won the 2017 Intelligent Health Award and was a finalist in the 2018 SXSW Interactive Innovation Awards.

Working with clinicians also introduced me to the neuroscience of learning, memory, and adaptation. It gave me questions that continue to shape my interest in AI: how a system changes through experience, what information it retains, how consequences affect what it pays attention to next, and many more.

As founder and CTO of Studio Assistant, I subsequently led an eight-person team through client discovery, architecture, delivery, and production operations for Warner, Belmont University, and the world’s largest streaming platform. That work required understanding each organisation’s workflows and turning the needs of different stakeholders into systems they could use.

Those experiences inform my AI research and consulting today. I look at how information, technology, people, and authority interact, and how a change intended to improve one part affects the whole. I work with a small number of clients across research, system design, and implementation, carrying those decisions through to working software.

Stephen Bartlett

Next step

Start with the problem.

Intro calls are 30 minutes. Tell me what you’re trying to do and I’ll tell you whether I can help.