From AI Spend to AI Strategy: A GenAI Roadmap for Enterprise Scale
Protagona partnered with a fast-scaling SaaS employee experience platform to assess GenAI and agentic AI readiness, validate AWS architecture patterns, and deliver a fundable business case — all within a two-week advisory sprint.
Industry
Startups & Software
Teams & Services
AI/ML Strategy, Data Architecture, Cloud Architecture, Technical Advisory
Tech & Tools
Amazon Bedrock, AWS Inferentia, Amazon SageMaker, Amazon EC2 GPU Instances, Amazon ECR, Amazon ECS, Amazon EKS, OpenAI, Anthropic, DynamoDB, S3 Vector, OpenSearch
Key Data Points
The Vision
A fast-growing SaaS platform powering intelligent intranet and communication tools for enterprise workforces had reached an inflection point. The agentic AI platform was live across hundreds of customers, enterprise search was processing massive datasets, and AI spend was accelerating. Leadership recognized that decisions around model selection, infrastructure, and cost structure would only get harder to unwind as the platform scaled toward 1 million employees served, and chose to get ahead of them.
The Goal
The engagement aimed to identify and prioritize the highest-impact GenAI and agentic AI opportunities, assess readiness across technical, data, security, and organizational dimensions, validate AWS service fit and architecture patterns, and produce an ROI-backed business case to support internal funding decisions and guide the platform's next phase of AI investment.
The Challenge
Several critical architecture decisions were deeply interconnected and could not be addressed in isolation. Enterprise search required locking in an embedding model before data was ingested at scale — a choice that becomes increasingly costly to reverse as more customers onboard. With one customer already at 900 GB and the platform growing quickly, the stakes were immediate, not hypothetical.
At the same time, the agentic platform was processing 400 to 600 requests per second across 800 customers, meaning infrastructure decisions carried real financial consequences today. A single third-party provider represented 95% of total AI spend, and token-based pricing was becoming a structural cost risk as the platform approached 1 million employees served. Evaluating self-hosted small language model deployment added further complexity. Each decision influenced the others, making a coordinated strategy essential.
The Solution
Protagona structured the engagement as a four-phase advisory sprint designed to move from discovery to a fundable recommendation within two weeks. Enterprise search and the agentic platform were identified as the highest priorities, as those architecture decisions are the hardest to reverse and carry the largest cost implications at scale. LLM cost optimization through dynamic model switching, self-hosted small language model infrastructure, and a bring-your-own-model control layer were sequenced as near-term follow-on work with clear business value.
The AWS architecture evaluation assessed model serving options across multiple approaches and validated the embedding and retrieval patterns best suited to the platform's enterprise search requirements. A dynamic model switching architecture, routing requests based on task complexity, was validated as a concrete mechanism for reducing token costs as volume grows. The engagement delivered a scored use case backlog, a sequenced one-year roadmap, an AI readiness report, and a full ROI model with optimistic, expected, and conservative projections including payback period and return on investment analysis.
.png)