Reframing Customer Journey Analytics

The digital analytics industry has expanded rapidly, yet it continues to face a persistent challenge: explaining why customers behave the way they do. Despite widespread adoption of sophisticated tracking platforms, most analytics systems remain better at describing outcomes than identifying the underlying causes behind customer decisions. Industry forecasts estimate that the global digital analytics market will reach approximately $50 billion by the early 2030s, underscoring how central behavioral data has become to enterprise decision-making.

Against this backdrop, a sequence-based analytical framework developed within Fractal Analytics proposes a structural shift in how customer journeys are interpreted. The framework was architected by Vinothkumar Kolluru, who defined the core representation approach and technical direction while collaborating with a broader analytics team at Fractal Analytics. Rather than relying on linear funnels and aggregate metrics, the approach treats customer journeys as structured behavioral sequences—preserving context so intent can be inferred rather than assumed.

Why traditional funnels fall short

The conversion funnel assumes customers progress through predictable stages—awareness, consideration, and purchase. In practice, modern journeys rarely follow such a path. Customers routinely switch devices, revisit decisions over extended periods, and interact across platforms that cannot always be fully observed. Management consulting research has shown that customer journeys now involve dozens to hundreds of touchpoints across channels, making linear funnel models increasingly inadequate. Funnels summarize outcomes efficiently, but they struggle to represent intent. Two users can generate nearly identical metrics while being in fundamentally different decision states—an ambiguity traditional analytics tools are not designed to resolve.

Limitations of prior analytical approaches

Before sequence-based modeling, most customer-journey analysis relied on approaches that simplify behavior in ways that limit explanatory power. Rule-based funnel analytics, still embedded in many commercial platforms, reduce journeys to predefined stages and aggregate metrics. While useful for reporting, these methods assume linear progression and fail to capture looping, cross-device, or delayed decision-making—limitations increasingly noted as journeys fragment across channels.

A second class of approaches applies “bag-of-events” or summary-feature models, where journeys are represented as unordered counts of page views, clicks, or time spent. This mirrors early bag-of-words techniques in language processing, which ignore word order and context. Foundational work in language modeling has shown that removing sequence information leads to substantial loss of meaning. Some analytics teams have adopted Markov-chain attribution and transition models to introduce order.

While these approaches model transitions, they typically assume short memory and homogeneous behavior, limiting their ability to represent long, irregular journeys and making them difficult to interpret at scale. More recently, deep learning sequence models have been explored in academic settings, but many remain impractical for enterprise use due to limited interpretability, sensitivity to missing data, and challenges translating outputs into actionable business decisions. These limitations left a gap between descriptive analytics and decision-making: systems could model behavior statistically, but not in ways that preserved intent, context, and actionability simultaneously.

A linguistics-inspired reframing of behavior

The conceptual shift behind the framework draws inspiration from natural-language processing. In language models, the meaning of a word depends on its context—its position relative to other words. Customer behavior follows a similar principle. A product-page visit after a comparison search often signals evaluation, while the same visit after a cart abandonment may indicate hesitation. Conventional analytics treats both as identical events, discarding the surrounding context that gives them meaning. By preserving sequence and transition information, the framework interprets journeys as ordered behavioral structures rather than isolated actions.

Three original technical contributions

The methodology introduced by the Fractal Analytics team rests on three core contributions, documented in technical publications released in 2025.

1. Sequence-preserving behavioral representation

Instead of collapsing journeys into averages and totals, the framework encodes event order directly. This allows journeys with similar intent to cluster together even when surface-level actions differ.

2. Data-driven discovery of intent patterns

Rather than relying on manually defined segmentation rules, the system identifies recurring behavioral patterns directly from observed data. This reduces analyst bias and allows intent categories to emerge organically from user behavior.

3. Validation focused on actionability

A central design principle is distinguishing between correlations that merely describe past behavior and patterns that remain stable and actionable when tested—addressing a widely recognized limitation in applied analytics research.

These principles are detailed in Fractal Analytics’ published research:

Taming the Clickstream: Transforming Raw Clicks into Actionable Intelligence; and Machine Learning in Digital Analytics: Unlocking Insights with the Semantic Layer.

From research to enterprise deployment

Unlike many academic approaches, the framework was designed for operational environments, with attention to interpretability and real-world constraints such as incomplete data and cross-platform behavior. Fractal Analytics has stated publicly that this work informs decision-making across large enterprise clients, particularly in personalization, journey analysis, and product prioritization.

Independent recognition and company context

The work has gained visibility during a period of increased scrutiny of Fractal Analytics’ research capabilities. In February 2026, Reuters reported that Fractal Analytics filed for an initial public offering, targeting a valuation of up to approximately $1.6 billion, positioning it among the first AI-focused firms in India to pursue a public listing.

Separately, Fractal’s AI research division achieved a leading position on MLE-Bench, a public benchmark developed by OpenAI to evaluate end-to-end machine-learning engineering systems. Within the organization, Kolluru received formal recognition for research contributions and advanced within Fractal’s technical career track in 2025, reflecting peer acknowledgment of his influence on applied research direction. The significance of this work extends beyond digital marketing. Any domain involving sequential decision-making—financial transaction monitoring, healthcare patient pathways, enterprise sales engagement, or media consumption—faces similar challenges in interpreting intent over time. Progress in these areas, the research suggests, may depend less on collecting more data and more on representing behavior in ways that preserve meaning.