Two engagements running. Five behind them.
Situation, what I did, and what it produced. Clients are described rather than named because this sits under commercial confidentiality, and live engagements are described in less detail than closed ones for the same reason. Named references come once we are talking properly.
01 / Running now
Current client work
What The Modern Analytics is doing this quarter. Both engagements are live, which is why the detail stops short of where a closed case study would go.
Data, analytics and agentic transformation
The situation
An agency whose reporting describes what already happened, delivered after the moment to act on it has passed. Routine operational work absorbing senior time that should be going to client thinking. And a genuine appetite for AI arriving before the foundation underneath it was ready to carry any.
What I am doing
- Leading the whole chain rather than a layer of it, because the value that reaches the end is the product of every link and not the best one.
- The data foundation first. Ownership, lineage and a governed layer the analytics can actually stand on. This is the unglamorous part and skipping it is why the interesting part fails later.
- Then the analytics layer. Governed metrics and self service, so answering a client question is a query rather than a rebuild.
- Then agentic capability, sequenced onto something solid. Routine operational work moved to agents with authority boundaries and rollback defined before anything runs unsupervised.
- Capturing the baseline before each phase goes live, which is the only point at which it can be captured at all.
Why healthcare makes it harder
Regulated content raises the bar on every automated step. It is not enough for an agent to be right. The reasoning has to be explainable to a client, to a brand team, and eventually to a regulator, and it has to be reconstructable months later. That constraint shapes the architecture from the first decision rather than being bolted on at review.
It is also why the governance work runs alongside the build rather than after it. Nothing here goes live and gets governed afterwards.
An agency does not have a data problem or an AI problem. It has a sequencing problem, and the cost of getting the sequence wrong is a year.
Three AI platforms, definition through to release
The situation
A company building in the digital trust and AI space needed capability that did not exist to buy, where the thinking and the building had to sit close enough together that a specification handed to strangers would have lost most of it.
What I did
- Took a digital and AI trust platform from product definition through to released software. Live.
- Took an AI oversight platform the same route: what is running, who owns it, how it is classified and the evidence behind it. Live.
- Currently building an AI adoption index scoring access, usage, workflow integration, trust and business value, comparable across business units and repeatable year on year.
- Owned architecture, data model, measurement design, workstreams and release cadence, with each release instrumented from the first day rather than the first review.
What it produced
This is also the reason building is not on my services list. Having done it is useful to you. Selling it to you would not be. The argument is here.
02 / Before this
Leading it in role
Five enterprise transformations, closed and measured. The numbers below survived a finance function whose job was to disbelieve them.
Commercial and operations transformation
The situation
A commercial and operations function with real data volume, real margin pressure and no single owner for data, analytics and AI. Investment was happening in pockets. Nothing connected the spend to an outcome anybody in finance recognised.
What I did
- Set a three year data and AI roadmap with the C suite, tied to growth and efficiency targets the business had already committed to.
- Built and led a 50+ person cross functional team across India, the US and Europe. Data engineering, analytics engineering, data science, product management and BI, on an agile pod model with three spoke leads.
- Established a target operating model and demand governance, which meant saying no in public. It removed 30%+ of low value initiatives from a $17M portfolio.
- Shipped 20+ AI and analytics products: churn prediction, dynamic pricing, customer segmentation, demand forecasting, opportunity scoring and predictive inventory.
What it produced
The operating model work was worth more than any single product. Deciding what not to build is where most of the money was.
Conversational AI, agentic workflows and model governance
The situation
Insight turnaround was measured in days. Analysts fielded the same questions repeatedly, and the people who needed answers were not the people who could write a query.
What I did
- Architected a RAG based conversational platform on AWS and Databricks, giving governed natural language analytics to the business rather than a new tool to learn.
- Established a model governance framework on Databricks Model Registry and MLflow: versioning, lineage, performance monitoring and drift detection.
- Then, and only then, deployed agentic workflows for customer service and sales operations, with authority boundaries and rollback defined before anything ran unsupervised.
What it produced
Governance first was not caution. It was the only way the agentic work got approved to run at all.
Rationalising a legacy platform estate
The situation
An estate that had accumulated rather than been designed. Overlapping tools, business critical logic living in Access and Alteryx, and a cost line nobody could attribute to an owner.
What I did
- Attributed cost to owners first, which is almost always the highest return move, because consumption stops being abstract.
- Retired what nothing depended on, after checking what nothing depended on rather than assuming.
- Modernised the legacy layer that quietly ran the business, and automated the manual workflows around it.
What it produced
Enterprise BI platform and analytics portfolio
The situation
A global BI platform serving a research organisation, with a large analytics portfolio and adoption that did not match the investment.
What I did
- Owned the enterprise BI and visualisation platform and a $12M+ analytics portfolio across drug development globally.
- Built the drug development data, analytics and AI function inside the corporate centre of excellence, including talent development and succession.
- Delivered 18+ data and analytics products and standardised governed self service across all regions.
- Drove automation led delivery improvement rather than adding headcount.
What it produced
Building an analytics function from zero
The situation
No analytics function. A global services organisation making decisions on reporting assembled by hand, with no product model and no roadmap.
What I did
- Built a globally distributed cross functional organisation from zero to one: BI, data engineering, data science, QA and UX.
- Set a three year data, analytics and AI product roadmap aligned to the business unit strategy, with value realisation tracked rather than assumed.
- Architected the platform, semantic models and integration pipelines underneath it.
- Institutionalised standards, governance and KPI models so the function outlasted the people who started it.
What it produced
Starting from zero is easier than fixing an operating model that already has defenders. Different problem, different difficulty.
Also / Built rather than advised
See the platformsRecognise any of this?
Bring one engagement that has not moved the way you expected, and we will find the gate it is stuck at.