Predictive analytics and decision science

Demand planning, customer acquisition, retention and pricing all depend on estimates of what may happen next. The useful model is the one that gives people the right information in time to make a better decision.

I start by establishing that decision, the information available when it must be made and how the result will be used. My work spans lead scoring, customer retention and retention automation, price automation, and demand forecasting across thousands of series, from data preparation and model development through operating integration.

A forecast for staffing, purchasing or financial planning needs to arrive at the right level of detail and early enough to affect the decision. Large collections of related series also need a repeatable process for updating forecasts, handling missing data and identifying changes that require attention.

I develop forecasting pipelines and compare their predictions with the approach already in use. Validation follows the information and timing available in operation, while error analysis examines the consequences of over- and underestimating demand. Prediction intervals and scenario analysis help the people planning capacity assess uncertainty.

Customer analytics connects acquisition, retention and growth. Lead scoring helps prioritize attention; retention models identify customers at risk of leaving; customer-value estimates relate expected behavior to future contribution. Lifecycle and campaign analysis show how those patterns change over time.

I connect these estimates to the available business response: who can receive attention, what action is possible, what it costs and how it reaches the customer. A high risk of leaving does not establish that an offer will help. That distinction leads from prediction to the intervention decisions below.

When the question is what a changed price, offer or service action would achieve, experiments and causal analysis provide the relevant comparison. I help define the eligible population, alternatives and outcome, then turn the estimate into a policy that reflects costs, capacity and commercial constraints.

Pricing makes the connection concrete. Expected demand and margin need to become a price that the operating system can apply, with clear rules for update frequency, exceptions and feedback. My work has included price automation alongside forecasting and customer analytics.

The customer-intervention resource develops the same reasoning for choosing whom to contact and which action to take, including constrained allocation and an operating worksheet.

A model needs to work under the conditions in which people will use it. I validate against the information available at the time, investigate errors that change the business response and connect the output to the relevant workflow. Explanations help people understand a result, challenge an assumption or identify a defect.

For an existing system, improvement may mean a revised target, a repaired data pipeline, a better decision threshold or a different model. Monitoring and feedback from the people using the results show where further work is needed.

The practice draws on multi-year client engagements in healthcare, insurance and financial services, as well as advice to portfolio-company teams building their data-science capabilities. Organizations I have worked with include RxSense / SingleCare, iptiQ (Swiss Re's digital insurance business), and Loeb Enterprises and its portfolio companies.

Across that work, the recurring challenge is to connect analysis with an operating decision. The experience spans customer acquisition and retention, campaign decisions, pricing and forecasting, with responsibility for the models, data pipelines and integration that make the analysis usable.

My financial-markets work includes research infrastructure, predictive models and live trading systems. In Machine Learning for Trading, I develop the path from data and research design through model validation, portfolio decisions and implementation.

The book's worked applications and software let readers examine these methods in detail. They complement the commercial forecasting and customer-analytics work described above.

You can engage me for a specific analysis, an independent review or continuing development alongside your analysts and engineers. The starting information is the decision, the available data, the current approach and what would count as an improvement.