Market Analytics

Bringing new products to market, sizing an opportunity, understanding your customer base, and protecting revenue from churn, all come down to the same underlying discipline: rigorous, evidence-based, decision-grade analytics. Whether you're validating a new concept before launch, quantifying the size of an addressable market, segmenting a customer base for sharper targeting, or building the models that explain why customers arrive, spend, and leave — this is where research design meets applied statistics to give you answers you can act on.

I have delivered advanced analytics across a wide range of business sectors, with particular depth in healthcare — spanning pharmaceuticals, medical devices, pharmacy robotics and automation, health insurance, and pharmacy benefit management. Additional industry experience includes telecommunications, financial services, home improvement, information technology, automotive financing, publishing, restaurants, office and print services, transportation, and media polling. Beyond primary research, my work has included exploratory analysis of secondary data sources — identifying predictors of medication adherence, benchmarking digital marketing spend across large and small businesses, and analyzing national retail chain performance against revenue growth, among other projects.

Below is an overview of my core service areas. Every engagement is scoped to the specific decisions you're trying to make — get in touch to talk through your project.

Concept & Product Testing

Before you invest in building or launching something new, you want defensible evidence that it will resonate — not just a gut check. Well-designed experiments are the most reliable way to establish that a message, feature, or price point actually causes a change in customer response, rather than merely correlating with it.

Output will inform which of several competing messages or value propositions has the strongest pull? How would customers respond to different combinations of product features? Does a new concept land differently across customer segments? What's the minimum customer base size needed to detect a meaningful effect?

Methods used in my past studies include:

A/B tests and sequential monadic designs: randomly assigning respondents to two or more alternative concepts to isolate causal effects. Getting this right means answering some foundational design questions up front: Are you comparing means or proportions? Is the outcome measure continuous or a hard threshold? What's the smallest effect size worth detecting? Is the design within-subject or between-subject? These choices determine both the statistical test used downstream and the power analysis needed to size the sample correctly — undersized studies risk missing real effects, oversized ones waste budget.
Factorial designs and discrete choice modeling (conjoint/CBC): testing combinations of features or claims rather than one variable at a time. Built on Random Utility Theory, discrete choice models estimate the latent "utility" customers place on each attribute, decomposing it into a systematic component (driven by actual feature and respondent differences) and a random error component that supports the external validity of the estimates. In practice, this means defining attributes and levels, generating a full factorial design, deriving an efficient fractional/blocked design, fielding balanced choice sets, and modeling results with conditional logit or ANOVA to quantify relative feature importance, main effects, and interactions — often delivered as an interactive simulator so your team can test "what-if" feature bundles on demand. MaxDiff approach can be layered in for prioritizing long feature or claims lists, and menu-based conjoint works well for configurable products and pricing tiers.
Price sensitivity and willingness-to-pay projections: pricing is, structurally, a special case of feature testing, but it deserves its own toolkit and techniques including Van Westendorp and Gabor-Granger, monadic price-point testing, price laddering, and discrete choice models that isolate price as one attribute among many. My work has spanned multi-stakeholder pricing simulations (e.g., patients, prescribers, and payers evaluating list price versus net price, access restrictions, and co-pay assistance for new pharmaceuticals), employer price sensitivity to bundled versus standalone wellness programs, bank customers' fee tolerance, and e-commerce pricing structure evaluation. Where applicable, these estimates feed directly into econometric revenue forecasts.

Market Segmentation

Segmentation partitions a target population into distinct groups that share underlying needs, attitudes, values, and behaviors — and, critically, are likely to respond similarly to product, pricing, and messaging decisions. Done well, it's one of the highest-leverage analytics investments a business can make, because a single segmentation solution can inform product development, competitive positioning, promotional targeting, and channel strategy all at once.

A well-executed segmentation does more than sort customers into labeled clusters — it becomes a working framework that shapes decisions across time and business verticals. Once you understand the distinct needs, attitudes, and behaviors that separate one group of customers from another, messaging and targeting stop being guesswork: campaigns can speak directly to what each segment actually cares about, rather than relying on a single generic pitch that resonates weakly with everyone. The same clarity extends to pricing and packaging — segments often differ in what they're willing to pay for and how they want to buy, which opens the door to differentiated pricing tiers and bundles that capture more value than a one-size-fits-all structure ever could. Segmentation also reveals which channels actually matter to which customers, so marketing and distribution spend can be directed toward the touchpoints each group genuinely relies on instead of being spread evenly across every possible channel. Just as valuable is what segmentation surfaces that wasn't visible before: pockets of under-served or niche customers whose needs aren't being met by current offerings, representing real opportunities for new products, features, or positioning. And when an organization is pulled in different directions by customers who seem to want contradictory things, a good segmentation resolves that tension — it shows that those aren't actually competing demands from the same market, but distinct demands from distinct groups, giving leadership a clear, evidence-based way to prioritize investment

My approach to segmentation generally follows below process:

1. Clearly define and operationalize the underlying constructs (needs, attitudes, behaviors) in valid, reliable measures

2. Examine the resulting variables with attention to their distribution and dispersion

3. Select measurement scales appropriate to each construct, grounded in psychometric best practice

4. Run clustering algorithms - e.g. k-means, latent class models, gaussian mixture models (GMM), partitioning around medoids (PAM), etc. chosen to fit the data structure

5. Test the internal coherence and stability of candidate cluster solutions to land on the most meaningful segmentation

6. Build out full segment profiles highlighting the dimensions that differentiate each group

7. Deliver a segment-typing tool so field and marketing teams can classify new customers quickly

I have produced segmentations across highly varied audiences — clinicians, pharmacists, and patients across multiple therapeutic areas; health plan beneficiaries; bank customers; users of curated e-commerce and marketplace platforms; small business owners; school teachers; park visitors; and non-profit program participants — which is precisely the point: the framework generalizes, while the specific constructs and clusters are always custom-built to your market.

Market Sizing

A credible market size estimate is one of the most scrutinized numbers in any business plan or investment case — and one of the easiest to get wrong. Two problems undermine most market sizing efforts: samples that don't represent the true target population, and respondents' well-documented tendency to overstate their likelihood of purchase.

Getting the sample right is arguably the most important step in market sizing. Representative projections depend on careful sample design and recruitment, sufficient sample size, and attention to coverage and nonresponse bias — especially with opt-in web panels, where self-selection bias is a real risk. Sample matching, propensity score adjustment, and post-stratification weighting are standard tools for correcting these issues before any projection is made.

Another essential step lies in correcting for stated-preference inflation. A substantial body of research shows that stated purchase intent systematically overstates actual market share, largely because real-world purchase decisions are shaped by frictions and constraints that never show up in a survey. Robust market sizing applies a calibrated adjustment factor to stated-preference estimates, typically informed by econometric data on comparable and competitive products already in market.

Market sizing requires consideration of the full cascade from the Total Addressable Market, the Serviceable Addressable Market your business model can realistically reach, and the Serviceable Obtainable Market you can capture given current constraints. Where segments are created, top-down sizing will refine share estimates starting from industry wide macro data and narrowing to your target segment's share. Annual market revenue projections can be built bottom up via unit economics (addressable customer count × penetration rate × average revenue per customer), generally the more defensible approach for investor and operating plans.

Revenue Projection & Growth Forecasting

Turning market and customer data into a credible revenue forecast means combining rigorous statistical modeling with a realistic view of how customers actually convert and behave over time.

Bottom-up funnel forecasting: building revenue from traffic, conversion rates, and average revenue per customer, broken out by channel, segment, and cohort rather than blended into a single top-line growth rate.
Cohort-based forecasting: tracking how each customer cohort's retention and expansion behaves over its lifecycle, now the standard approach for subscription and recurring-revenue businesses.
Time series and machine learning forecasting: classical ARIMA/SARIMA models, increasingly supplemented with tools like Prophet or gradient-boosted trees (XGBoost, LightGBM) that incorporate external drivers, plus hierarchical forecasting methods for businesses with many products or regions to project simultaneously.
Bayesian forecasting: structural time series methods that produce a full uncertainty distribution around a forecast, not just a single point estimate, most critical when a plan or investment case needs to show a credible range of outcomes.

Driver Analysis for Customer Acquisition

Understanding why customers arrive, and which levers actually move that number, requires careful separation of correlation from causation across your marketing and growth channels.

Marketing Mix Modeling (MMM): regression-based, often Bayesian, estimation of each channel's incremental contribution to acquisition, using tools like Meridian or Robyn. MMM has seen renewed adoption as privacy changes (iOS tracking restrictions, cookie deprecation) have degraded the accuracy of digital attribution.
Multi-touch attribution (MTA): still relevant for digital-heavy funnels, though best triangulated against MMM given growing signal loss on major platforms.
Incrementality and geo-lift testing: holdout-region or holdout-audience experiments that isolate the true causal lift of a channel or campaign, generally considered the gold standard wherever a clean test is feasible.
Uplift modeling / causal machine learning: causal forests and double machine learning approaches that estimate which customers respond to which interventions, enabling acquisition spend to be targeted by actual responsiveness rather than broad propensity scores.

Customer Churn & Attrition Modeling

Retaining an existing customer is almost always cheaper than acquiring a new one, which makes churn modeling one of the highest-ROI analytics investments available — provided the model tells you not just who is at risk, but why, and with enough lead time to act.

Predictive classification: gradient-boosted tree models (XGBoost, LightGBM, CatBoost) are now the standard for churn scoring, generally outperforming simpler logistic regression baselines on the nonlinear interactions typical of customer behavior data.
Survival analysis: Cox proportional hazards and discrete-time hazard models, which estimate *when* a customer is likely to churn rather than a simple yes/no, and correctly handle customers who haven't churned yet (censored data) — a meaningful advantage over standard binary classifiers.
Explainable driver analysis: SHAP-value analysis layered on top of ML churn models to translate "black box" predictions into clear, actionable drivers of churn risk, both in aggregate and at the individual customer level.
Customer health scoring: composite indices combining usage frequency, support interactions, satisfaction scores, and engagement trends, used operationally (especially in B2B and SaaS) to trigger retention outreach before a formal churn model flags the account.