For global pharma companies, LoE is one of the most financially significant events in a brand’s lifecycle. When generic or biosimilar competitors enter the market at significantly lower price points, the originator brand can experience rapid and deep sales erosion. These dynamics directly inform high-stakes decisions such as pricing strategy, market prioritization, and whether to defend or de-invest in late-lifecycle brands.

Yet, forecasting LoE impact remains difficult. Because LoE happens only once per product, traditional forecasting methods - largely based on a brand’s own historical sales - struggle to predict post-LoE behavior accurately. As a result, many organizations still rely on manual, market-by-market erosions estimates and bespoke external analyses. These approaches are time-consuming, expensive, hard to scale, and often inaccurate, leading to both premature de-investment and over-investment in brands post-LoE.

The client needed an internal, scalable machine-learning capability to foreacst LoE erosion faster, cheaper, and with higher accuracy than human planners, while reducing reliance on subjective judgement.

‍

For global pharma companies, LoE is one of the most financially significant events in a brand’s lifecycle. When generic or biosimilar competitors enter the market at significantly lower price points, the originator brand can experience rapid and deep sales erosion. These dynamics directly inform high-stakes decisions such as pricing strategy, market prioritization, and whether to defend or de-invest in late-lifecycle brands.

Yet, forecasting LoE impact remains difficult. Because LoE happens only once per product, traditional forecasting methods - largely based on a brand’s own historical sales - struggle to predict post-LoE behavior accurately. As a result, many organizations still rely on manual, market-by-market erosions estimates and bespoke external analyses. These approaches are time-consuming, expensive, hard to scale, and often inaccurate, leading to both premature de-investment and over-investment in brands post-LoE.

The client needed an internal, scalable machine-learning capability to foreacst LoE erosion faster, cheaper, and with higher accuracy than human planners, while reducing reliance on subjective judgement.

‍

Historically, our client estimated LoE impact using a small set of erosion observations from manually selected local analogues from its own portfolio. This exercise had to be done market by market and was frequently supplemented by costly external consultant analyses. The limitation of this approach is structural: LoE behavior is highly contextual and driven by a complex interaction between product characteristics and local market dynamics. A handful of hand-picked analogues cannot reliably capture this variability.

The strategic shift was to move from one-off analyses to a reusable, data-driven forecasting engine - leveraging a large-scale external data foundation to learn from hundreds of historical LoE events across markets, rather than a small number of internal examples. Together with finance and forecasting stakeholders, we set out to:

1. Establish a unified data model across markets, manufacturers, and brands with standardized inputs (e.g. sales history, time on market, LoE and generic entry dates)
2. Build a large-scale analogue library and derive comparable historical erosion curves across products and countries
3. Generate forward-looking erosion predictions that integrate seamlessly into existing forecasting and planning processes

Historically, our client estimated LoE impact using a small set of erosion observations from manually selected local analogues from its own portfolio. This exercise had to be done market by market and was frequently supplemented by costly external consultant analyses. The limitation of this approach is structural: LoE behavior is highly contextual and driven by a complex interaction between product characteristics and local market dynamics. A handful of hand-picked analogues cannot reliably capture this variability.

The strategic shift was to move from one-off analyses to a reusable, data-driven forecasting engine - leveraging a large-scale external data foundation to learn from hundreds of historical LoE events across markets, rather than a small number of internal examples. Together with finance and forecasting stakeholders, we set out to:

1. Establish a unified data model across markets, manufacturers, and brands with standardized inputs (e.g. sales history, time on market, LoE and generic entry dates)
2. Build a large-scale analogue library and derive comparable historical erosion curves across products and countries
3. Generate forward-looking erosion predictions that integrate seamlessly into existing forecasting and planning processes

Using IQVIA MIDAS data, we constructed an analogue library covering thousands of historical brand-country LoE events. For each analogue, we modelled 50+ product and market characteristics, grouped into three main categories:

1. ‍Clinical & product features (e.g. route of administration, molecule type, formulation complexity)‍
2. Competitive & lifecycle features (e.g. time on market, pre-LoE market share, pioneer status)‍
3. Country & market features (e.g. procurement centralization, substitution enforcement, pricing mechanisms)

These features were combined with observed post-LoE sales erosion patterns. The model runs thousands of simulations to learn which characteristics most strongly influence erosion behavior and how different feature combinations translate into distinct erosion trajectories. Based on this learning, the model generates a unique brand- and market-specific erosion curve for each target product-country combination. The approach captures non-linear effects, reduces subjective bias from manual analogue selection, and produces more accurate and explainable forecasts to drive better decision-making.Importantly, the model also provides feature-importance insights, allowing users to understand why a given erosion profile is predicted - not just what the prediction is. This enables teams to explore questions such as:

1. Does longer exclusive time on market result in patient loyalty and translate into higher post-LoE retention?
2. Are high-revenue brands more attractive targets for generic or biosimilar manufacturers and thus expected to erode faster?
3. Do brands sold primarily through hospital channels face faster substitution and steeper erosion?

Using IQVIA MIDAS data, we constructed an analogue library covering thousands of historical brand-country LoE events. For each analogue, we modelled 50+ product and market characteristics, grouped into three main categories:

1. ‍Clinical & product features (e.g. route of administration, molecule type, formulation complexity)‍
2. Competitive & lifecycle features (e.g. time on market, pre-LoE market share, pioneer status)‍
3. Country & market features (e.g. procurement centralization, substitution enforcement, pricing mechanisms)

These features were combined with observed post-LoE sales erosion patterns. The model runs thousands of simulations to learn which characteristics most strongly influence erosion behavior and how different feature combinations translate into distinct erosion trajectories. Based on this learning, the model generates a unique brand- and market-specific erosion curve for each target product-country combination. The approach captures non-linear effects, reduces subjective bias from manual analogue selection, and produces more accurate and explainable forecasts to drive better decision-making.Importantly, the model also provides feature-importance insights, allowing users to understand why a given erosion profile is predicted - not just what the prediction is. This enables teams to explore questions such as:

1. Does longer exclusive time on market result in patient loyalty and translate into higher post-LoE retention?
2. Are high-revenue brands more attractive targets for generic or biosimilar manufacturers and thus expected to erode faster?
3. Do brands sold primarily through hospital channels face faster substitution and steeper erosion?

The solution enables pharma organizations to anticipate LoE erosion faster, at lower cost, and with greater accuracy, allowing teams to focus on the strategic decisions that matter most around LoE. Phase 1 delivered a scalable forecasting capability and operating model that significantly outperformed traditional approaches. Compared to manual analogue-based forecasting, the model demonstrated:

1. Materially improved predictive accuracy across initial pilotmarkets (90+% avg. accuracy)
2. Learning from 1,000+ historical analogues, providing farbroader empirical grounding than traditional methods
3. Systematic incorporation of product and market drivers,reducing subjectivity and forecast variability

Beyond accuracy, the approach delivers several strategic and operational advantages:

1. Reduced reliance on costly, one-off external analyses
2. Shorter lead times and lower manual effort per forecast
3. Full internal ownership with the ability to continuously improve the model
4. Transparency into key erosion drivers, supporting stronger cross-functional alignment

The solution enables pharma organizations to anticipate LoE erosion faster, at lower cost, and with greater accuracy, allowing teams to focus on the strategic decisions that matter most around LoE. Phase 1 delivered a scalable forecasting capability and operating model that significantly outperformed traditional approaches. Compared to manual analogue-based forecasting, the model demonstrated:

1. Materially improved predictive accuracy across initial pilotmarkets (90+% avg. accuracy)
2. Learning from 1,000+ historical analogues, providing farbroader empirical grounding than traditional methods
3. Systematic incorporation of product and market drivers,reducing subjectivity and forecast variability

Beyond accuracy, the approach delivers several strategic and operational advantages:

1. Reduced reliance on costly, one-off external analyses
2. Shorter lead times and lower manual effort per forecast
3. Full internal ownership with the ability to continuously improve the model
4. Transparency into key erosion drivers, supporting stronger cross-functional alignment

In practice, the solution supports multiple stakeholders looking to answer different key business questions:
‍

Finance Leaders: How much LoE downside risk are we carrying, and when does it materialize?
‍

Portfolio & Lifecycle Strategy Teams: Which assets are worth defending post-LoE, and which should be deprioritized?
‍

Commercial Leaders managing late-lifecycle assets: Where and how much should we continue to invest in sales and marketing?
‍

Data & Analytics Leaders: How can we move beyond traditional forecasting approaches towards scalable and explainable ML-driven models?

In practice, the solution supports multiple stakeholders looking to answer different key business questions:
‍

Finance Leaders: How much LoE downside risk are we carrying, and when does it materialize?
‍

Portfolio & Lifecycle Strategy Teams: Which assets are worth defending post-LoE, and which should be deprioritized?
‍

Commercial Leaders managing late-lifecycle assets: Where and how much should we continue to invest in sales and marketing?
‍

Data & Analytics Leaders: How can we move beyond traditional forecasting approaches towards scalable and explainable ML-driven models?