Clearwater Analytics: 20% More Assets, Flat Headcount — the Internal-First AI Playbook

Clearwater Analytics (NYSE: CWAN), the investment accounting platform that reports on more than $7.3 trillion in assets every day, grew its assets under management by over 20 percent while operational headcount stayed flat. The engine behind that equation is a generative AI program launched in May 2023, built on Amazon SageMaker and Amazon Bedrock, and documented in a technical case study on the AWS Machine Learning Blog in December 2024.

A business where growth once meant hiring

Investment accounting is a volume business. Every institutional client arrives with thousands of positions to reconcile, daily data feeds from custodians and asset managers, and reporting deadlines that repeat monthly, quarterly, and annually. Clearwater Analytics, founded in Boise, Idaho in 2004 and listed on the NYSE since 2021, built its business automating that work for insurers, corporations, and asset owners — a platform that today reports on trillions in assets daily with a workforce of about 1,600 professionals.

The structural challenge is familiar to every services-heavy SaaS company: growth in assets under management historically pulled operational staffing up with it. More clients meant more reconciliation exceptions, more service tickets, more analysts. That linear coupling between revenue and operating cost defines the margin ceiling of the category. In early 2023, as generative AI matured, Clearwater's engineering leadership framed a precise ambition: absorb double-digit asset growth with the operations team already in place, and turn the platform's own documentation — thousands of pages of manuals, help content, and domain knowledge — into working software.

The decision that made it possible: employees first, platform second

The sequencing is the most instructive part of this story. In May 2023, Clearwater launched a private, secure generative AI chat assistant for its internal workforce, built on retrieval-augmented generation over the company's own knowledge base. The internal-first move answered the hardest questions — accuracy, data governance, retrieval quality — on an audience of employees, months before any client touched the system. By September 2023, at the Clearwater Connect user conference, the company announced its customer-facing generative AI offerings.

The second decision was architectural. Rather than bolting a chatbot onto the product, Clearwater transformed its existing console into the Clearwater Intelligent Console (CWIC), an AI platform organized around three layers: knowledge awareness, which delivers domain answers — book value calculations, reconciliation processes — through RAG; application awareness, which guides users toward the right reports and features; and data awareness, which runs complex portfolio queries with real-time calculations. Alongside CWIC sit Crystal, an internal operations assistant with access to broader data sources and APIs, and CWIC Specialists, domain-specific agents for accounting and regulatory workflows trained on thousands of pages of documentation.

The model strategy stayed pragmatic. Clearwater pairs advanced foundation models — including Anthropic's Claude 3.5 Sonnet — for complex reasoning with smaller fine-tuned models, deployed through Amazon SageMaker JumpStart, for rapid domain-specific response. An 11-step evaluation framework compares fine-tuning against RAG for every use case, measuring accuracy against operating cost before anything ships. Data governance drew a hard line from day one: public help content trains shared models, while client data lives exclusively inside dedicated, client-approved models.

The result, with full context

By September 2024, roughly sixteen months after the first internal deployment, the numbers were on the table. Assets under management grew by over 20 percent while operational headcount stayed flat. Crystal, the internal assistant, delivered efficiency gains between 25 and 43 percent in the operational workflows it touches. Client-facing response times improved and service ticket volumes fell.

SAGA reads vendor-published numbers with discipline, and this case deserves both credit and context. The figures appear on the AWS Machine Learning Blog in a piece co-authored by Clearwater's own engineers and AWS architects — an official channel with commercial interest in the outcome. Two anchors strengthen the claim. First, Clearwater is a public company: its headcount and asset trajectories are auditable through quarterly filings, which makes an inflated efficiency claim expensive. Second, the claim itself is precise: commercial growth had multiple drivers, and the verified operational fact is that the existing team absorbed that growth — the AI layer expanded service capacity at stable staffing. That is exactly the before/after equation CFOs ask vendors to prove.

What other organizations can learn

Three conditions made this result possible, and all three travel well beyond fintech. First, internal-first sequencing: Clearwater answered its hardest governance and accuracy questions on an internal audience, produced measurable evidence, and walked into client conversations with sixteen months of operational proof. Organizations that invert the order face those questions in public. Second, documentation as raw material: thousands of pages of manuals and help content powered both the RAG systems and the fine-tuned models, so companies with mature, well-maintained knowledge bases start this journey with their most valuable asset already built. Third, measurement over faith: the fine-tune-versus-RAG evaluation framework turned model selection into an empirical, per-use-case decision grounded in cost-benefit analysis — a discipline that keeps spending aligned with outcomes as model prices move.

The scale point matters most. Clearwater employs about 1,600 people — a mid-size enterprise, far from hyperscaler budgets. The playbook — internal pilot, platform transformation, empirical model selection, hard data boundaries — is within reach of any organization with proprietary knowledge and a measurable operations baseline. The destination is the equation on the first line of this story: service capacity up more than 20 percent, operating cost flat.