Confidential
Wilmington, DE
Mental Health / Digital Wellness / SaaS
Amazon QuickSight Amazon Q Amazon Athena Amazon S3 AWS Glue Data Catalog AWS IAM QuickSight SPICE
A leading digital mental health platform serves millions of users with free and premium wellness support services, generating rich behavioral and clinical data across every interaction. Despite sitting on a valuable data lake in Amazon S3, the organization’s analytics capabilities were locked behind technical SQL expertise, leaving product, operations, and clinical teams unable to independently answer critical questions about user behavior, wellness outcomes, and subscription health. Avahi designed and deployed a GenAI-powered analytics platform using Amazon QuickSight and Amazon Q, enabling non-technical stakeholders to query their own data in plain English and access interactive dashboards consolidating key business KPIs, all within a three-week engagement. The result is a democratized, accurate, and scalable analytics foundation ready to support a full production rollout.
The client is a Wilmington, Delaware-based digital health and wellness company offering one of the world’s largest mental health support platforms, providing free peer-to-peer emotional support alongside premium therapy and counseling services. Operating at the intersection of mental health, behavioral data, and SaaS subscription models, the platform serves a broad and diverse user base and generates high volumes of interaction, wellness assessment, and subscription event data, making timely, accessible analytics critical to understanding and improving user outcomes at scale.
Despite having years of valuable user interaction data stored in Amazon S3 as parquet files, the client’s analytics workflow was entirely dependent on technical staff who could write and execute SQL queries. Product managers, operations leads, and clinical staff had no way to independently explore data or answer time-sensitive business questions, creating a persistent bottleneck that slowed decision-making across the organization.
The absence of interactive dashboards meant there was no consolidated view of the platform’s most important KPIs: member growth, listener conversions, wellness outcomes, subscription trends, or feature engagement. Insights that should have been available in seconds required engineering time to produce, and even then, inconsistent metric definitions across teams created further risk of misaligned decisions.
Most critically, the platform had no natural language interface, no way for a non-technical team member to ask “What percent of users improved their wellness score in the last 90 days?” and receive an instant, accurate answer from their own data. In a domain where understanding clinical outcomes and user behavior directly informs product and operational decisions, this gap had real consequences. Without a solution, analytics would remain gated behind technical resources indefinitely, slowing growth decisions and limiting the platform’s ability to measure and act on the user wellness trends at the heart of its mission.
The client’s data was already resident in Amazon S3, making AWS the natural home for any analytics solution. Rather than introducing a third-party BI or NL query tool that would require data movement, additional security review, and new cost structures, AWS offered a native, integrated path — Amazon Athena for serverless SQL over S3, Amazon QuickSight for dashboards and visualization, and Amazon Q for natural language query — all operating within the client’s existing AWS account boundary. This meant sensitive mental health and behavioral data never left the client’s own infrastructure, a non-negotiable requirement for a platform operating in the digital health space.
AWS’s SPICE in-memory engine within QuickSight also provided the performance characteristics needed to make dashboards fast and responsive over high-volume event data — without generating runaway Athena query costs. The combination of security posture, native integration, and cost efficiency made AWS the clear platform of choice.
The client engaged Avahi for their demonstrated ability to architect production-grade analytics solutions on AWS, specifically the combination of rigorous data engineering and GenAI layer tuning required to make Amazon Q perform accurately on real-world, high-volume, domain-specific data. Building a natural language analytics interface that non-technical users can actually trust requires more than enabling a feature. It requires careful data architecture, metadata enrichment, and honest documentation of what the system can and cannot reliably answer. Avahi brought all three.
The three-week time-boxed engagement structure also gave the client a clearly scoped, low-risk path to validating the technology before committing to a full production rollout. Avahi’s ability to take an organization from raw S3 parquet files to a working GenAI analytics platform within that window — on a sensitive and analytically complex data domain — made them the right partner for the engagement.
Avahi began with a structured discovery phase, reviewing the client’s Amazon S3 data layout across their Research and Sales Analytics buckets, analyzing parquet schemas, identifying the highest-priority analytics use cases, and configuring least-privilege AWS IAM roles governing access across all services. AWS Glue Data Catalog was used for schema discovery and cataloguing, enabling Amazon Athena and Amazon QuickSight to connect to S3 data with consistent, governed metadata.
The central architectural challenge was SPICE capacity. The client’s raw event tables were too large and high-cardinality for direct QuickSight ingestion. Avahi’s solution was to design five purpose-built Amazon Athena views, each scoped to a specific analytical domain with targeted time windows, pre-aggregations, and field pruning, reducing data volume to analytically relevant subsets while preserving all required business logic. The five views were: base event slice, 30 days; subscription health, 90 days; first vs. second wellness assessment outcomes; 365 days, feature DAU and engagement intensity; 30 days, and members, listeners, and conversions, 1 year.
A particularly complex design challenge was computing accurate cumulative metrics, total members, listeners, and conversions, within windowed datasets. Avahi solved this by building carry-in CTEs within each Athena view, calculating opening balances so that lifetime totals start from the correct historical baseline regardless of the dashboard’s display window. The wellness delta view introduced a clinically meaningful capability: surfacing a before-and-after comparison of user wellness scores between first and second assessment, giving the platform a direct, data-driven lens into whether users are improving under care.
Nine interactive Amazon QuickSight dashboard visuals were built across these five datasets, consolidating the platform’s most important KPIs into a single, accessible interface for non-technical stakeholders. Amazon Q was then enabled within QuickSight and systematically tuned, dataset topics were enriched with field aliases, synonyms, and descriptions, and high-cardinality or ambiguous columns were hidden from natural language query scope, to ensure that plain-English questions returned accurate, trustworthy answers. This metadata enrichment is what distinguishes a GenAI analytics layer that stakeholders trust from one they quickly abandon.
Critically, the hard business logic — carry-in calculations, delta scoring, device classification, service type normalization — was computed in SQL within the Athena views, not left to Amazon Q to infer. This architectural decision means the GenAI layer only needs to filter and aggregate pre-validated fields, dramatically reducing the risk of plausible-but-incorrect answers on complex analytical questions. The engagement concluded with end-to-end validation of all data flows, dashboards, and natural language queries, a curated performance evaluation documenting supported and unsupported Amazon Q query types, a knowledge transfer session, and full technical documentation and runbooks.
Avahi delivered a complete GenAI analytics platform, from raw S3 parquet data to natural language queries, in three weeks, on a sensitive and analytically complex data domain. Non-technical stakeholders across product, operations, and clinical functions can now ask plain-English questions directly of their own data and receive instant, accurate answers, without writing a single line of SQL. The platform’s most critical KPIs (member growth, listener conversions, wellness outcomes, subscription trends, and feature engagement) are now consolidated in interactive dashboards accessible to the entire team.
The Athena view architecture and SPICE optimization pattern established in this engagement serve as a reusable foundation for the client’s production analytics rollout. The performance evaluation and recommendation report delivered in Phase 3 provides the roadmap for scaling the platform to the full dataset, additional dashboard domains, and broader team access, with a clear understanding of Amazon Q’s capabilities and boundaries already documented and tested.
Engagement Scope and Success Criteria Delivered:
Let’s explore your high-impact AI opportunities together in a complimentary session
