Cost Optimization: Feature-Management Service Rolling Fingerprint Refactor

The client is the family-fintech category leader: debit card, investing, savings, safety, and senior protection. The platform serves 6.5M parents and kids, with $100M+ ARR by late 2021 and $2B+ in family-managed assets by 2025. The product is direct-to-consumer across four subscription tiers (Core, Max, Infinity, Family Shield) and B2B2C through the B2B2C banking platform. Taller's engagement at this client is the largest sustained mobile-plus-backend engineering footprint in the portfolio: three master-deck case studies, more than sixty filled roles across 2022 to 2026, and pods covering Backend Node.js, Android, iOS, Fullstack, QA, SDET, SRE, and Security. The two cases below describe what Taller has delivered against the family financial platform's most consequential strategic bets.

A core function in the feature-management service maintains a "rolling fingerprint" at the user-by-feature level, writing a record to a DynamoDB table for every user::feature_key combination. At peak, this generated up to 1,500 writes per second, driving significant and largely unnoticed DynamoDB costs amounting to over $200K per year. The expense had gone unflagged until it surfaced internally, leaving the team without a clear picture of how much the table was costing or whether the underlying granularity was even required by downstream teams.

The engagement began with a deep investigation into the original implementation, tracing how the rolling fingerprint logic worked, instrumenting and tracking usage metrics to quantify actual write volume, and identifying where writes could be reduced without affecting functionality. The team then executed a refactor to lower the number of DynamoDB writes required to maintain the fingerprint, followed by thorough QA to validate that behavior remained correct. The work also scoped a potential next-phase optimization: consolidating to a single set of attributes per user rather than one per user::feature_key, contingent on confirming whether the per-combination granularity is actually consumed by teams across the organization.

Direct cost reduction. The refactor cut DynamoDB write volume against the fingerprinting table, delivering savings of over $200K per year on a previously unmonitored expense.

Cost visibility. Usage metrics were instrumented and tracked, giving the team a quantified, evidence-based view of write patterns where none existed before.

Validated quality. The refactor was QA'd thoroughly, ensuring savings were achieved with no regression in functionality.

Identified further upside. A clear, scoped path to additional savings was defined, pending a functional decision on fingerprint granularity.

Beyond the immediate savings, the work demonstrated a repeatable approach to cost governance: surfacing hidden infrastructure spend, grounding decisions in measured usage rather than assumptions, and de-risking changes through disciplined QA. A single overlooked table was costing more than $200K annually. The same investigative method can be applied across other services to find comparable savings. It positioned the team to make an informed build-or-defer decision on the next optimization phase, weighing further cost reduction against a change in functionality, rather than guessing. This reflects the contractor's value not just as an implementer, but as someone who investigates root causes, quantifies impact, and frames the trade-offs leadership needs to decide where to invest next.