AI analytics explains where your patients actually dropped off

A huge share of prescriptions never turn into actual medication for the patient, costing the industry an estimated $637 billion a year. Drop-off happens at several points:

• About 30% of prescriptions never even reach the pharmacy
• Another 20% are abandoned when patients see a high copay
• More are lost while waiting for prior authorization approval.

‍

Analytics that track where patients drop off (doctor’s office, pharmacy counter, or PA queue) let you target fixes like copay support, faster authorization help, or clearer coverage communication at the exact failure point.

You can track patient persistence and identify early warning signs of discontinuation by monitoring refill timing, days on therapy, and adherence metrics such as percentage of days covered and amount of medication supplied over a time period (PDC and MPR). This helps you spot patterns of declining engagement.

‍

Common early indicators include:

• delayed or missed refills
• shorter days of supply
• switching pharmacies
• increased prior authorization activity
• rising out-of-pocket costs

‍

By comparing each patient’s behavior to expected refill schedules and typical adherence patterns, you can detect risk signals early. You can then intervene with support programs that help keep patients on therapy longer.

You can identify where patients drop off between prescription and first fill by tracking each step of the medication start process and measuring where progress stops. You do this by analyzing data from:

• prescribing
• benefits verification
• prior authorization
• pharmacy processing
• claims activity

‍

This shows which stage has the highest abandonment, such as:

• delays in approval
• high out-of-pocket costs
• lack of patient follow-up

‍

This step-by-step visibility makes it easier to pinpoint the exact barrier preventing patients from moving from prescription to first fill.

AI analytics identifies root causes by automatically scanning patient journey data to detect unusual patterns. It also pinpoints the factors most strongly associated with the drop-off. The system compares recent behavior to historical baselines. It analyzes variables such as:

• payer changes
• prior authorization delays
• pharmacy rejections
• rising copays
• inventory gaps
• patient disengagement

‍

AI highlights which factors changed at the same time as the abandonment spike. It also ranks their impact. This provides fast, data-driven explanations that help teams understand what happened and respond quickly.

AI agents can proactively alert you by continuously monitoring patient journey metrics. They automatically detect unusual changes within specific regions, payer groups, or physician segments.

‍

When the system identifies a deviation from normal patterns (such as a sudden rise in abandonment for a particular payer or geographic area), it triggers an alert that:

• highlights the affected segment
• explains the likely drivers behind the change

‍

This real-time, automated monitoring helps teams respond quickly to emerging access issues, payer behavior shifts, or provider-level barriers.

When abandonment jumps suddenly, teams usually lose days doing manual triage. A hub leader sees abandonment rise from 18% to 53% in the Southeast and asks what changed. The team then pulls specialty pharmacy rejection data, checks payer PA approvals, reviews benefits verification logs, looks at copay support usage, breaks results by prescriber, and searches for recent formulary updates. This work can take 2–3 days, and patients keep dropping off during the delay.

‍

Agentic patient journey analytics automates this investigation. When abandonment rises beyond expected levels, the system triggers an investigation automatically.

‍

• Automatic anomaly detection: The system monitors abandonment by region, payer, specialty pharmacy, prescriber group, and patient segment. It accounts for normal weekly variation and seasonality, and then flags true “breaks” in the pattern.
  
• Parallel root-cause checks: AI agents investigate common drivers at the same time:
  
  • Payer policy, access shifts and whether the effective dates match the spike.
  • Specialty pharmacy issues using rejection codes and dispense timelines.
  • Hub workflow bottlenecks
  • Copay assistance gaps
  • Prescriber concentration
    
• Quantified attribution with confidence: The output is not a list of guesses. It estimates how much each factor contributed and includes confidence based on timing and strength of evidence.
  
• Impact and action plan: It shows which payers, pharmacies, or practices drove the spike, estimates the NBRx risk, and recommends actions in priority order (for example, “PA documentation support for these practices” or “proactive BV for this payer segment”).
  
• Learning over time: The system remembers which patterns led to which causes and starts with the most likely paths first, so investigations get faster.

‍

Teams can move from “we noticed a spike” to “here is what changed, who is affected, and what to do next” in minutes instead of days, which reduces lost starts during access disruptions.

Yes. This is the practical way to monitor thousands of payer–region–pharmacy–therapy combinations without missing issues. Leaders cannot realistically scan hundreds of dashboards and catch early problems. Many issues are discovered only in weekly or monthly reviews, after patients have already been lost.

‍

Agentic monitoring continuously watches abandonment, time-to-therapy, adherence signals, PA delays, specialty pharmacy rejection patterns, and hub workflow bottlenecks across payers, regions, pharmacies, indications, and provider segments.

‍

• The system uses anomaly detection to separate real signals from normal noise by learning what “normal” looks like for each cohort, including seasonality and expected variation.
  
• When something meaningful changes, agents do more than alert. They automatically run a root-cause investigation, such as checking payer policy changes, PA approval shifts, pharmacy operational issues, copay friction, and provider-level clustering.
  
• Alerts are then prioritized, so leaders see only the few issues that matter most, ranked by business impact, confidence, urgency, and actionability.

‍

Each alert should include a plain-English summary of what changed, where it happened, what likely caused it, how big the risk is, and what actions are recommended. Over time, the system should learn which patterns usually lead to real losses and which interventions tend to work, so alerts and recommendations improve.

Early warning signs of drop-off include:

• Taking a long time to fill the first prescription
• Filling only part of what was prescribed
• Increasing gaps between refills

‍

Machine learning models can use patient characteristics, drug type, and these early fill patterns to flag patients who are likely to stop treatment. These models can often do this with around 75% accuracy.

This gives you a 90-day window to intervene before patients disappear. Typical interventions include: nurse outreach, payment help, or adherence programs.

Yes. Early signs in the first 30–60 days often indicate whether a patient will reach treatment goals. These early markers can include side effects, early lab results, and how regularly a patient takes the drug.

‍

Machine learning models can learn patterns from past patients and use them to flag new patients who look unlikely to do well if nothing changes. This allows care teams to step in early with dose changes, extra support, and therapy switches

‍

The key advantage is timing. You act early instead of waiting months to confirm that the regimen is not working.

Where a patient lives (their ZIP code) often predicts whether they will stay on medication better than many clinical factors. Social factors like income, education level, access to transportation can explain 30–40% of the differences in adherence. Yet many programs focus mainly on disease education or reminders.

‍

Analytics can help by combining:

• claims data
• outside data such as census information, maps of areas with few pharmacies, and local socioeconomic indicators.

‍

This can flag patients who are more likely to struggle because of their life situation. Then you can add the right kind of support, such as transportation help, financial support, or local resources, instead of relying only on clinical coaching.

Patient journey analytics integrates these sources by consolidating them into a unified data model that tracks each step of a patient’s access and adherence journey.

‍

Each source contributes different signals:

• Claims data shows fills and discontinuation.
• Specialty pharmacy data adds dispense details and rejection insights.
• Hub services data provides visibility into benefits verification and prior authorization steps.
• CRM systems contribute patient outreach and support activity.

‍

By linking these datasets at the patient or case level, platforms create a more complete view. This helps teams understand drivers of access, speed-to-therapy, and persistence, using a connected record instead of separate spreadsheets and dashboards.

This is hard because when patients switch specialty pharmacies due to insurance changes or service issues, their data becomes scattered across systems that often do not share a common patient ID. On top of that, each pharmacy may define and calculate adherence differently, which makes comparisons inconsistent.

‍

Tellius can address this using identity resolution, which matches patients across systems based on patterns in their data. This helps stitch together one continuous journey so you can see adherence and access even when patients move between pharmacies, without relying on manual stitching.

Most pharma companies do not have direct access to EMR (Electronic Medical Record - a full clinical picture) data, so they rely on claims data. Claims show billing and fills, but it does not capture clinical details like lab results or symptom changes.

‍

Some real-world evidence platforms solve this by aggregating de-identified (removing or masking information that can identify a person) EMR data and selling access, but those datasets can be expensive.

‍

Tellius can sit on top of the real-world data you already have such as the following, and analyze them together.

• claims,
• limited EMR feeds (if available),
• patient services data

‍

This gives you the richest view possible from your available data while staying within privacy and budget limits.

Working with patient-level data requires HIPAA compliance, which includes:

• De-identifying (removing or masking information that can identify a person like name, address etc.) data where possible,
• Using the minimum necessary information,
• Having proper Business Associate Agreements (BAAs) in place.

‍

Many generic analytics tools were not designed with this in mind.

‍

Life sciences-focused platforms are built to support HIPAA-compliant setups. For example, Tellius has SOC 2 Type II certification and can be deployed with:

• strong access controls and role-based permissions,
• encryption of data at rest and in transit,
• detailed audit logs showing who accessed what and when.

‍

The technology can support HIPAA, but you still need appropriate governance and operational controls on top of the platform.

Deployment timelines typically range from a few weeks to a few months, depending on the complexity of your data sources and integration needs.

‍

Most platforms can connect to claims, specialty pharmacy, hub, and CRM systems quickly. However, timelines vary based on:

• data availability,
• data quality,
• amount of required customization.
  
  

Many modern AI-driven solutions offer prebuilt connectors and templates that accelerate onboarding. This helps teams start exploring patient insights and abandonment patterns faster than traditional implementations.

Patient journey analytics should answer the questions reps care about in plain language, such as:

• “Where are patients in my territory dropping off?”
• “Which doctors’ patients wait the longest to start therapy?”

‍

The most effective approach is to push these insights into tools field teams already use, such as call prep context and alerts. By this way, reps are not running reports. They get clear signals on who to see, what to discuss, and which barriers to address in each account this week.

Building in-house usually means:

• buying large claims datasets,
• setting up systems to match patient records across sources,
• and putting strong privacy controls in place.

‍

This can cost millions up front, plus ongoing support.

‍

Partnering with an external vendor is faster, but it gives you less control and can expose some of your strategy and data to a third party.

‍

Tellius offers a middle option, where you run patient journey analytics on your own data, and plug in outside data only where needed, so you keep more control while still moving quickly.

Most patient journey programs still depend on claims that arrive 30–90 days late. If a patient stops filling in June, you might not see it clearly until August, after the patient has been off therapy for weeks. This delay hurts every workflow: abandonment reporting, adherence outreach, hub staffing, and program ROI measurement.

‍

Why the lag happens

• Claims adjudication takes time. Reversals, resubmissions, and coordination of benefits mean “final” claims often settle weeks later.
  
• Vendor and delivery cycles are slow. Aggregators standardize and project data on weekly or monthly schedules.
  
• Integration is batch-based. Files move through exports, FTP, ETL, and validation steps that add more days.
  
  

How modern platforms fix it without pretending claims will be instant

• A multi-speed data approach: Use faster operational signals for decisions, and keep adjudicated claims for final measurement.
  
• Proxy signals for early risk: The system watches leading indicators that appear weeks before claims confirm a stop, such as PA abandonment, repeated refill reminders with no response, copay card rejections, BV stalls, or repeated failed outreach.
  
• Streaming and API integrations: Instead of monthly file drops, the platform connects to hub and SP systems so “case moved to closed-no-response” is visible quickly.
  
• Data blending: The platform combines claims, SP, hub, and assistance data to create a more current journey view than claims alone.

‍

Traditional setups see discontinuation 60-90 days late and intervene when success rates are low. Modern setups see risk signals in 7-14 days, which gives teams a much better chance to re-engage patients while they still remember the therapy and the process.

Specialty pharmacies can look similar on paper but behave very differently in the real world. Some fill prescriptions quickly, support patients well, and drive strong adherence. Others are slower or offer weaker support.

‍

You can compare pharmacies using patient-journey and support signals such as:

• Time from prescription to first fill
• Refill patterns
• Use of support services, such as nursing or reminders

‍

These comparisons show:

• which pharmacy partners provide the best patient experience
• where you should steer more volume or push for performance-based contracts

Many patients drop out during prior authorization because the process takes too long or paperwork keeps bouncing back. Analytics helps by showing where the process breaks down most often, segmented by:

• payer,
• drug, or
• patient type.

‍

Tellius can map each step of the prior auth journey (from request to approval or denial). and spot where cases pile up or stall. This helps teams redesign workflows and hub support to move cases faster and cut down the 30% drop-off.

To measure real impact, you need to link program participation and longitudinal adherence behavior using patient-level data integration across program platforms and claims databases.

Analytics with proper attribution windows matters because education effects can fade over 6–12 months. With the right setup, you can quantify education ROI instead of guessing.

‍

Alongside prescription data, Tellius can analyze:

• the educational content delivered (PDFs, videos stored in Google Drive)
• nurse educator call recordings
• patient feedback forms

‍

This helps identify which specific messages and delivery methods are linked to sustained adherence (i.e., the patient keeps taking the medication consistently over time, not just for the first few weeks) improvement. It moves measurement beyond “who participated” to content effectiveness.

Patient support programs (nurse calls, injection training, reminder texts) often sit in a separate system from prescription data. As a result, you see activity (calls made, emails sent), but not whether patients actually stayed on therapy. This makes ROI hard to prove.

‍

The fix is to connect program enrollment and interaction data to long-term prescription data, so you can see which patients stayed on treatment and for how long.

T

ogether with claims data, Tellius can pull in:

• nurse call notes from Google Drive
• recorded support calls
• enrollment spreadsheets

‍

Then it can show which specific touchpoints and education content are linked to better adherence, not just which programs generated the most activity.

Different teams measure adherence differently using metrics like:

• PDC (Proportion of Days Covered) - percent of days a patient is covered by medication in a period
• MPR (Medication Possession Ratio) - total days’ supply dispensed divided by days in the period, based on fills
• persistence (how long a patient stays on therapy)
  
  

Often, each team uses its own formulas. When specialty pharmacies, payers, and internal teams all define adherence differently, it becomes almost impossible to compare results or have a clear discussion.

‍

What you need is:

• one standard way to calculate each metric
• one “source of truth” that everyone uses
• controlled flexibility for special cases (for example, oncology vs chronic primary care drugs).

‍

A single source of truth for adherence math brings clinical, access, and commercial teams onto the same page.

Patient services teams constantly face triage. You might identify 5,000 patients at risk, but only have capacity to proactively contact 800 per month. Simple rules like “call the highest risk first” waste effort because some patients are hard to reach or have barriers the hub cannot solve. AI optimization makes the tradeoff explicit and improves outcomes with the same capacity.

‍

• Impact scoring, not just risk scoring: The system estimates expected benefit for each patient by combining:
  
  • likelihood of discontinuation,
  • likelihood the intervention will work (based on similar past cases),
  • barrier type (PA vs copay vs operational delay),
  • effort required (time and complexity),
  • expected value of keeping the patient on therapy (duration and margin assumptions).
    
    
• Constraint-aware allocation: It optimizes within real limits like call capacity, nurse specialization, and the need to protect reactive support levels.
  
  
• Continuous re-prioritization: The list updates when new signals arrive. A patient can jump up the list when a PA stalls, a copay card gets rejected, or a refill date passes.
  
  
• Best intervention recommendation: It also suggests the best route, such as contacting the prescriber’s office for missing PA documentation instead of calling the patient when the patient cannot solve the issue.
  
  
• Scenario planning: Leaders can test “what if we add 2 FTE” or “what if we shift capacity from long calls to faster copay interventions”, and see the expected persistence lift.

‍

Instead of calling the “highest risk” 800 patients, you contact the highest expected-impact 800 patients. Organizations typically see meaningful persistence gains with the same staffing, because the work shifts toward barriers that can be solved and patients who are reachable.

Patient journey analytics in pharmaceutical commercial operations is the process of integrating data from:

• claims
• specialty pharmacies
• hub services
• patient-support programs

‍

to understand how patients move through the therapy journey, including:

• diagnosis → prescription
• first fill
• adherence
• long-term persistence

‍

It is important because it helps commercial teams identify barriers such as:

• access issues
• prior authorization delays
• high out-of-pocket costs
• pharmacy rejections
• where patients drop off

‍

By providing a complete, data-driven view of patient experiences, it enables better decision-making and targeted interventions.

Yes, but it requires linking datasets that are not naturally connected.

‍

The time from diagnosis to the first prescription fill can range from a few days to several months. Long delays often signal access problems and lost lifetime drug use.

‍

To measure time-to-therapy, you need to link the following for the same patient. This is technically hard to do.

• medical claims (where diagnosis is recorded)
• pharmacy claims (where the first fill appears)

‍

Platforms that can join these datasets using anonymous patient IDs (often called de-identified tokens) make it possible to track time-to-therapy by payer, region, and provider while still protecting patient privacy.

A full patient journey analytics setup (data licenses, technology, and people) often costs $2-5 million per year. The payback comes from three main areas:

• Lower abandonment (fewer patients dropping before first or second fill), often giving a 2-3% prescription lift
• Better adherence and persistence, where patients stay on therapy longer, giving 5-10% improvement in ongoing use
• More efficient patient services, where you spend support dollars on the patients who need them most, often improving efficiency by around 20%

‍

For a $1 billion brand, even modest gains in these areas can translate into $50-100 million in extra revenue, which is roughly a 10-20x return on the analytics investment.

Conversational analytics enables non-technical teams to explore patient journey data by letting them ask questions in natural language and get answers generated by the analytics engine.

‍

Instead of writing SQL or navigating complex dashboards, users can ask questions like:

• “Where do patients drop off?”
• “Which payers have the highest abandonment?”

‍

The system automatically retrieves the relevant data, analyzes it, and returns insights, visualizations, and explanations. This makes patient journey analysis accessible to brand teams access specialists and field teams who need fast, self-service insights without technical skills.

They should be able to get instant answers. Hub work happens in real time, so waiting days for analyst tickets breaks operations.

‍

Traditional setups force hub teams to submit requests and wait 2-5 days for custom reports. By the time the report arrives, patient statuses and priorities have already changed. This creates predictable friction. Hub teams triage with incomplete visibility, they repeat the same questions, and analytics teams get flooded with routine requests. The real constraint is data quality and refresh speed. If patient identity matching is weak or data updates are slow, “instant answers” will still be incomplete or stale.

‍

Conversational patient journey analytics fixes this by letting users ask questions in plain English and get answers from integrated data.

• It works when the platform has a patient services semantic layer that understands terms like “at-risk,” “PA barrier,” “Southeast,” and “intervention success rate,” and maps them to the right data fields and rules.
  
  
• It also supports follow-up questions with context, so users can drill down without restating everything (for example: at-risk → PA barriers → barrier types → highest success rates → assigned to my team → prioritize by expected impact).
  
  
• The platform must query multiple sources together, such as hub case status, specialty pharmacy signals, payer/formulary rules, copay assistance activity, and historical intervention outcomes.
  
  

The output should be action-oriented. It should return a prioritized list of patients, not just a chart, and it should support export or handoff to the team’s workflow.