Marketing Analytics, Performance and Decision-Making

Introduction

Healthcare marketing increasingly operates in an environment of measurable interactions. Patients search for symptoms and providers online, compare services, contact call centres, book appointments, receive reminders, attend or miss visits, evaluate experiences, and recommend—or criticize—organisations. Managers therefore need more than promotional creativity. They need a reliable system for converting data into decisions about market reach, service access, resource allocation, patient experience, growth and financial sustainability.

Marketing analytics is the systematic collection, integration, analysis and interpretation of marketing and service data for decision-making. In healthcare, its purpose is not simply to increase volume. Analytics must support appropriate access, clinically responsible demand, equity, privacy, patient autonomy and organisational value. A campaign that produces many inquiries but overwhelms a clinic, attracts inappropriate demand or excludes disadvantaged groups is not a successful campaign.

This chapter develops an integrated approach to dashboards, funnel analysis, cohort analysis, attribution, experimentation and forecasting. It also examines acquisition cost, retention, lifetime value, satisfaction, referral, reputation, market share and return on investment (ROI).

Learning objectives

By the end of this chapter, the learner should be able to:

1. Design a healthcare marketing measurement system and dashboard.
2. Calculate and interpret reach, conversion, acquisition cost, retention, lifetime value, satisfaction, referral, reputation, market share and ROI.
3. Apply funnel, cohort and attribution methods while recognising their assumptions.
4. Design ethically appropriate experiments and interpret their results.
5. Use forecasting to support capacity, budget and channel decisions.
6. Evaluate data quality, privacy, bias and governance risks in marketing analytics.

1. From activity measurement to decision intelligence

A useful measurement system connects four levels:

This distinction prevents the common error of treating activity as performance. Impressions and clicks are outputs of media delivery; they are not evidence that patients received appropriate care or that the organisation created value.

A measurement architecture should begin with a decision rather than with available data. For example, a manager may need to decide whether to increase investment in paid search, extend clinic hours, redesign a booking journey or target lapsed patients. Each decision requires a defined population, time period, intervention, outcome, comparator and action threshold.

A healthcare analytics system normally includes:

• a data dictionary defining events, fields, units and ownership;
• a patient or consumer identity-resolution method, subject to lawful governance;
• links among campaign, contact-centre, appointment, attendance, billing and experience data;
• documented inclusion and exclusion rules;
• a reporting layer for monitoring and a deeper analytical layer for explanation;
• controls for access, auditability, retention and correction of data.

The organisation should distinguish the person exposed to marketing, the person making an inquiry, the patient receiving care and the payer. They may be the same person, but they are not analytically interchangeable.

2. The healthcare marketing funnel

A funnel represents movement through stages such as awareness, consideration, inquiry, booking, attendance, treatment completion, repeat use and advocacy. It is a useful management model, although real patient journeys are not always linear. A patient may attend through a referral without seeing an advertisement, return after several years, or move between channels.

For stage i, the stage conversion rate is:

where N_i is the number entering stage i and N_i+1 is the number reaching the next stage. The formula assumes that both counts refer to the same population definition and period, and that individuals are not unintentionally counted multiple times.

Overall conversion from an initial audience to a later outcome is:

Consider 20,000 targeted impressions, 1,000 landing-page visits, 120 inquiries, 60 bookings and 45 attended appointments. The visit rate is 1,000 ÷ 20,000 = 5%; inquiry conversion is 120 ÷ 1,000 = 12%; booking conversion is 60 ÷ 120 = 50%; attendance conversion is 45 ÷ 60 = 75%; and exposure-to-attendance conversion is 45 ÷ 20,000 = 0.225%. The largest managerial opportunity is not automatically the stage with the lowest percentage. Managers should estimate the effect of improving each stage, its cost, operational feasibility and clinical appropriateness.

A funnel dashboard should show both counts and rates. Rates alone can conceal a small denominator; counts alone can conceal declining efficiency. It should also segment results by service line, geography, device, referral source, payer, age and other variables only where legally permitted and clinically meaningful.

3. Core performance measures

Reach and engagement

Reach is the number of distinct people or households exposed to a message during a specified period. Impressions are total exposures and may include repeated exposure. Frequency is:

A high frequency may support recall, but excessive repetition can waste budget or create irritation. Engagement measures—clicks, video completion, calls or content interaction—are useful diagnostic indicators, not final outcomes.

Conversion and acquisition cost

Customer or patient acquisition cost (CAC) is commonly expressed as:

The numerator must be defined. It may include media only, or media plus creative, agency, technology and relevant staff costs. The denominator must specify whether it means inquiries, booked appointments, attended first visits or clinically eligible new patients. A cost per inquiry is not equivalent to cost per attended patient.

If a campaign costs $12,000 and produces 80 new patients who attend an initial appointment, CAC is $12,000 ÷ 80 = $150 per attended new patient. If only 50 subsequently complete an appropriate treatment episode, the cost per completed episode is $240. These are different management measures.

Retention and churn

Retention should be defined around a meaningful event, such as a second visit within 12 months, completion of a care programme or renewal of a membership. For a cohort:

Churn is the proportion becoming inactive under a stated inactivity rule. Retention and churn are not necessarily exact complements when patients can be transferred, censored, deceased or lost to follow-up. Survival analysis may be preferable when follow-up periods differ.

Lifetime value

A simple contribution-based lifetime value model is:

where m is contribution margin per period, r is the probability of remaining active from one period to the next, d is the discount rate per period, and C_s is acquisition or onboarding cost. This is a perpetuity approximation. It assumes a constant margin, constant retention probability, a stable service model and an indefinite horizon. It is inappropriate when retention changes substantially with tenure or when capacity limits service delivery.

For example, if monthly contribution is $80, monthly retention is 0.90, the monthly discount rate is 0.01 and acquisition cost is $150:

LTV = (80 × 0.90) ÷ (1.01 − 0.90) − 150 = 654.55 − 150 = $504.55

A more realistic finite-horizon model is:

where m_t is contribution in period t, p_t is the probability of remaining active and T is the analysis horizon. LTV should be calculated by service, segment and payer where economics and ethical implications differ.

Satisfaction, loyalty and referral

Satisfaction is commonly measured using survey items, such as an overall rating or a question about whether expectations were met. Net Promoter Score (NPS) is calculated as:

Promoters typically select 9 or 10 on a 0–10 recommendation question, detractors select 0–6, and passives select 7 or 8. NPS ranges from −100 to +100. It is a descriptive indicator, not a universal measure of quality, and should not be used without examining response bias, sample size, language accessibility and comments.

A referral rate can be defined as:

Attribution requires a consistent question or coding process. A patient may have encountered several influences, so referral counts should not be interpreted as proof that one person caused the entire decision.

Reputation and market share

Reputation can be monitored through review volume, average rating, sentiment, complaint themes, response time and the proportion of reviews receiving a response. Sentiment classification should be validated because sarcasm, clinical terminology and cultural language can produce errors.

Market share depends on a defined market and denominator:

The relevant market may be geographic, service-specific, payer-specific or defined by clinical eligibility. If a hospital performs 2,000 relevant procedures in a market of 10,000, its share is 20%. Share growth may result from market contraction rather than organisational growth; therefore, share must be reported with absolute volume and market size.

Return on investment

A basic incremental ROI is:

“Incremental” means the change caused by the marketing intervention relative to a credible counterfactual. Revenue should not automatically be substituted for contribution, because revenue excludes variable delivery costs. ROI also differs from return on advertising spend (ROAS), which commonly uses attributable revenue divided by advertising spend and may ignore other costs.

4. Cohort analysis and attribution

A cohort is a group sharing a defining starting event, such as first appointment month, campaign exposure month or programme enrolment month. Cohort tables reveal whether performance changes for newer groups or whether an apparent overall improvement merely reflects a favourable mix of older patients.

For example, a retention table may list acquisition cohorts in rows and months since acquisition in columns. A January cohort might show 100%, 62%, 48% and 40% active at months zero through three, while a March cohort shows 100%, 70%, 55% and 46%. The March cohort appears healthier at equivalent tenure, suggesting a service or acquisition improvement. Analysts must distinguish retention from calendar-period effects and allow for incomplete follow-up, known as right censoring.

Attribution assigns credit for an outcome to marketing touchpoints. Common approaches include:

• First-touch attribution: all credit goes to the first recorded touchpoint; useful for demand creation but insensitive to later influence.
• Last-touch attribution: all credit goes to the final touchpoint; simple but often overvalues booking interfaces.
• Linear attribution: credit is divided equally among recorded touches.
• Time-decay attribution: later touches receive more credit according to a specified decay rule.
• Data-driven attribution: statistical models estimate contribution from observed paths.
• Incrementality analysis: compares outcomes with and without exposure and is generally stronger for causal decisions.

Attribution is not causality. People who search for a provider shortly before booking may already have strong intent. A high-performing channel may be harvesting demand created elsewhere. Privacy restrictions, missing identifiers, cross-device behaviour, offline referrals and consent choices further weaken deterministic attribution. Use attribution for description and experimentation for causal claims.

5. Experimentation and causal decision-making

An experiment compares an intervention group with a control group under a planned protocol. Random assignment is valuable because, in expectation, it balances observed and unobserved characteristics between groups. Appropriate healthcare marketing experiments include different appointment reminders, landing-page designs, outreach messages or call-centre scripts.

The absolute treatment effect is:

where p_T is the outcome proportion in the treatment group and p_C is the outcome proportion in the control group. The relative lift is:

If attendance is 78% in treatment and 72% in control, the absolute effect is 6 percentage points and relative lift is 6 ÷ 72 = 8.33%. Both should be reported because percentage points and percent change are not interchangeable.

A credible experiment specifies the primary outcome, unit of randomisation, eligibility, duration, sample-size rationale, stopping rule, exclusions and safety or equity checks. Avoid repeatedly checking results and stopping when a favourable difference appears. If randomisation is impossible, quasi-experimental designs such as interrupted time series, difference-in-differences or matched comparison groups may be used, but their assumptions must be examined.

In a difference-in-differences design, the estimated effect is:

The key assumption is parallel trends: without the intervention, treatment and comparison groups would have followed similar outcome trends. Seasonality, concurrent campaigns and service changes can violate this assumption.

Statistical significance does not establish managerial importance. A tiny effect may be statistically precise but operationally irrelevant; a useful effect may be uncertain because the sample is small. Report confidence intervals, effect size, costs, harms and subgroup results where justified.

6. Forecasting and resource decisions

Forecasting estimates future demand using historical patterns, explanatory variables or both. A simple moving average is:

where D_t is observed demand in period t, k is the number of periods averaged and F_t+1 is the next-period forecast. It is easy to explain but responds slowly to sudden changes and does not explicitly model seasonality.

Exponential smoothing is:

where 0 ≤ α ≤ 1. A larger α gives more weight to recent demand. Forecast accuracy can be assessed with mean absolute error:

MAE is expressed in the same units as demand. Forecasts should include uncertainty ranges where possible. A model trained on an unusual pandemic period, a changed referral system or a temporary advertising burst may perform poorly when conditions change. Forecasts are decision aids, not guarantees.

7. Dashboard design and managerial interpretation

A useful dashboard is a decision interface rather than a wall of charts. It should contain:

1. a stated purpose and audience;
2. a small set of outcome and diagnostic measures;
3. definitions, denominators, time windows and data refresh dates;
4. trend lines and comparisons with targets or prior periods;
5. segmentation that supports action without exposing unnecessary personal information;
6. alerts linked to named owners and response protocols;
7. uncertainty, data-quality flags and explanatory annotations.

A balanced healthcare marketing dashboard might include the following layers:

Dashboard governance should prevent metric drift. A “new patient” definition that changes from booked to attended can create a false performance trend. Every metric should have an owner, formula, source system, refresh frequency and escalation threshold.

Illustrative case: Riverside Integrated Care Network

Background

Riverside Integrated Care Network, a fictional multi-site provider, offers physiotherapy and chronic disease management. Its digital campaign generated many appointment requests, but attendance was low and clinics experienced uneven demand. Management wanted to determine whether to expand advertising.

Problem and analysis

Over three months, the campaign cost $30,000. It generated 15,000 landing-page visits, 900 inquiries, 450 bookings and 315 attended first appointments. The funnel rates were 6% from visit to inquiry, 50% from inquiry to booking and 70% from booking to attendance. Cost per attended first appointment was $30,000 ÷ 315 = $95.24.

A cohort analysis showed that patients booking through a mobile form had 62% attendance, compared with 78% for patients booking through the call centre. Interviews identified confusing eligibility language and limited reminder options. The apparent marketing problem was partly a service-design problem.

Approach

Riverside introduced a shorter mobile form, displayed eligibility information in plain language, offered text reminders and tested the changes using random assignment among eligible online bookers. The control group retained the original process. Attendance became 76% in treatment and 70% in control, an absolute improvement of 6 percentage points. The organisation also monitored cancellation, access by language group and appointment availability to ensure that improved conversion did not create inequitable access or unsafe congestion.

Outcome and decision

The intervention increased attendance without increasing inappropriate bookings. Riverside expanded the redesigned journey but did not immediately double media spending. It first forecast capacity, calculated incremental contribution, reviewed no-show effects and established a monthly cohort dashboard. The case demonstrates that analytics should identify the constraint in the whole system, not merely reward the channel producing the largest number of leads.

Learning points

1. A low outcome rate may originate after the marketing interaction.
2. Counts, rates, cost and capacity must be considered together.
3. Cohorts can reveal differences hidden by aggregate averages.
4. Controlled testing provides stronger evidence than last-touch attribution.
5. Patient access and equity are outcome constraints, not optional reporting fields.

8. Ethics, privacy and governance

Healthcare marketing data can reveal diagnoses, reproductive choices, mental-health conditions, insurance status and other sensitive information. Analytics therefore requires a lawful basis, purpose limitation, data minimisation, security controls, retention rules and transparent communication. Consent for care does not automatically authorise marketing use. Organisations must comply with applicable privacy, consumer-protection, health-record and professional advertising requirements.

Important safeguards include role-based access, encryption, audit logs, suppression of sensitive audiences, de-identification where feasible and procedures for correcting inaccurate records. Small-cell reporting can enable re-identification even when names are removed. Vendors and advertising platforms require contractual, technical and governance review.

Algorithmic targeting can reproduce historical inequity. A model trained on past utilisation may interpret underuse by disadvantaged groups as low demand rather than as an access barrier. Performance should therefore be examined across relevant groups, with attention to missing data, language, disability, geography, socioeconomic status and digital exclusion. Sensitive segmentation should be justified by patient benefit and legal authority, not merely by higher predicted response.

Marketing analytics should never manipulate fear, exploit vulnerability, conceal costs or imply guaranteed clinical outcomes. A high ROI campaign is unacceptable if it generates unsafe demand, misleading claims or discriminatory access.

9. Technology and contemporary developments

Customer relationship management systems, electronic health records, call-centre platforms, web analytics and survey tools increasingly exchange data through application programming interfaces. A unified data layer can reduce duplicate counting, but integration also increases the consequences of incorrect identity matching.

Artificial intelligence can support forecasting, text classification, next-best-action suggestions and anomaly detection. Its use requires validation, monitoring for drift, explainability appropriate to the decision, human oversight and controls against automation bias. Generative AI may assist with dashboard commentary or campaign drafts, but it should not receive identifiable health information without approved safeguards. Human review remains essential for clinical claims, accessibility, cultural appropriateness and regulatory compliance.

The decline of third-party tracking and the growth of privacy-preserving measurement make first-party consented data, aggregate reporting, clean experimentation and incrementality increasingly important. Organisations should treat privacy not only as a constraint but as a design principle that can improve trust and data quality.

Best-practice checklist

• Begin with a decision and define the outcome before selecting metrics.
• Use stable definitions, denominators and time windows.
• Separate descriptive attribution from causal evidence.
• Report absolute effects, relative effects, costs and uncertainty.
• Connect marketing results to operational capacity and patient experience.
• Use cohorts to distinguish lifecycle effects from calendar effects.
• Validate forecasts against holdout periods and changing conditions.
• Segment responsibly and inspect equity consequences.
• Govern access, consent, vendors, retention and auditability.
• Treat dashboards as managed products with owners and action protocols.

Chapter Summary

Marketing analytics in healthcare is a disciplined system for connecting data with responsible decisions. Funnel analysis locates movement and friction across the patient journey; acquisition cost, retention and lifetime value connect activity to economics; satisfaction, referral and reputation illuminate relationship quality; market share and ROI place performance in a competitive and financial context. Cohort analysis reveals lifecycle differences, while attribution describes touchpoints but does not by itself establish causality. Experiments and suitable quasi-experimental designs provide stronger evidence for intervention effects. Forecasting supports capacity and budget planning but must acknowledge uncertainty and structural change. Effective dashboards combine outcome, diagnostic, economic, experience and equity measures with clear definitions and accountable actions. Throughout, privacy, fairness, transparency, clinical appropriateness and patient welfare are central conditions of valid healthcare marketing performance.

Key Terms

Marketing analytics: Systematic use of marketing and service data for decision-making.

Reach: Number of distinct people exposed to a communication.

Impression: One recorded exposure to a communication.

Conversion rate: Proportion moving from one defined stage to another.

Funnel: Model of sequential stages in a customer or patient journey.

CAC: Cost attributable to acquiring a defined new patient or customer.

Retention: Continued activity of members of an original cohort.

Churn: Loss of active status under a stated inactivity rule.

Lifetime value: Expected discounted contribution from a relationship over its life.

Contribution margin: Revenue remaining after specified variable service costs.

NPS: Percentage of promoters minus percentage of detractors.

Referral rate: Share of new patients attributed to existing-patient referrals.

Reputation: Stakeholder perception reflected through ratings, reviews and related signals.

Market share: Organisation volume divided by relevant market volume.

ROI: Incremental contribution relative to incremental investment.

Cohort: Group sharing a defined starting event or characteristic.

Attribution: Assignment of outcome credit to marketing touchpoints.

Incrementality: Change caused by an intervention relative to a counterfactual.

Forecasting: Estimation of future demand or performance from data and assumptions.

Data governance: Policies and controls governing data quality, access, use and accountability.

Review Questions

Short Answer Questions

1. Distinguish reach from impressions.
2. Why is cost per inquiry different from cost per attended patient?
3. What is the purpose of cohort analysis?
4. State one limitation of last-touch attribution.
5. Why should a dashboard display metric definitions and denominators?

Long Answer Questions

1. Explain how a healthcare organisation can construct a balanced marketing performance dashboard.
2. Discuss the assumptions and limitations of a contribution-based lifetime value model.
3. Compare first-touch, last-touch, linear and incrementality-based approaches to attribution.
4. Explain how experimentation can improve healthcare marketing decisions while protecting patient welfare.
5. Evaluate the ethical and governance requirements for using sensitive healthcare data in marketing analytics.

Analytical / Application Questions

1. A campaign costs $18,000 and produces 120 attended first appointments, of which 75 complete an appropriate care episode. Calculate cost per attended appointment and cost per completed episode. Which measure should management use for a decision about clinical value, and why?
2. A reminder intervention produces attendance of 84% among 500 treatment patients, compared with 78% among 500 control patients. Calculate the absolute treatment effect and relative lift. Identify two additional checks needed before scaling the intervention.
3. A provider’s procedure volume rises from 1,800 to 2,000, while the relevant market rises from 8,000 to 12,000. Calculate market share in both periods and interpret whether the provider strengthened its competitive position.

Multiple Choice Questions with Answers

Question 1

A campaign generates 1,000 inquiries from 20,000 landing-page visits. What is the inquiry conversion rate?

A. 2%

B. 5%

C. 20%

D. 50%

Correct Answer: B. 5%

Question 2

A campaign costs $12,000 and produces 80 attended new patients. What is CAC per attended new patient?

A. $80

B. $120

C. $150

D. $960

Correct Answer: C. $150

Question 3

Which method provides the strongest basis for estimating whether a campaign caused an outcome?

A. Last-touch attribution

B. Impression counting

C. Randomised controlled experimentation

D. Average review rating

Correct Answer: C. Randomised controlled experimentation

Question 4

If treatment attendance is 78% and control attendance is 72%, what is the absolute treatment effect?

A. 6 percentage points

B. 8.33 percentage points

C. 6% relative lift only

D. 150 percentage points

Correct Answer: A. 6 percentage points

Question 5

Which statement best describes a cohort?

A. All impressions in a reporting period

B. A group sharing a defined starting event

C. All patients who gave a high rating

D. A forecast error measure

Correct Answer: B. A group sharing a defined starting event

Question 6

In the simple LTV model, what does r represent?

A. Revenue per advertisement

B. Referral volume

C. Probability of remaining active between periods

D. Reputation score

Correct Answer: C. Probability of remaining active between periods

Question 7

A provider performs 2,000 procedures in a relevant market of 10,000. What is its market share?

A. 2%

B. 5%

C. 20%

D. 50%

Correct Answer: C. 20%

Question 8

Which metric is expressed in the same units as demand?

A. Mean absolute error

B. Relative lift

C. Market share

D. NPS

Correct Answer: A. Mean absolute error

Question 9

Why can attribution overstate the value of a search channel used immediately before booking?

A. Search never affects behaviour

B. The channel may harvest demand created by earlier influences

C. Search data cannot contain dates

D. Booking is unrelated to marketing

Correct Answer: B. The channel may harvest demand created by earlier influences

Question 10

Which is an essential ethical safeguard for healthcare marketing analytics?

A. Using all available patient data indefinitely

B. Removing all human review from automated decisions

C. Purpose limitation and appropriate privacy controls

D. Targeting vulnerable groups without disclosure

Correct Answer: C. Purpose limitation and appropriate privacy controls