What this is
Aderant builds practice management software for law firms: 268 customer environments globally and a 38-person SierraOps support team. Previously, each of the 34-40 weekly support tickets required 15-25 minutes of manual context gathering. Now an AWS Lambda function runs hourly, calls the lightweight model Nova Lite on Amazon Bedrock to read the ticket, queries related information across Jira, Confluence, and SharePoint, and produces routing recommendations. Humans step in only on low-confidence cases.
The numbers are concrete: 2.5 weeks in production, 109 tickets processed, 96% accuracy, 8-14 engineer hours reclaimed weekly, and under $30 per month in cloud spend. AWS case studies are routinely cherry-picked, but this batch is unusually complete—what we find worth noting is that it represents a standard pattern for how mid-size B2B companies put AI into production.
Industry view
Supporters see this as the textbook answer to "quiet AI deployment": 96% accuracy, under $30 a month, near-zero ROI drag. For mid-size SaaS firms, no AI team is needed—Bedrock's off-the-shelf models plus Lambda orchestration get a production-grade scenario live in weeks.
Critics push back. The first line is "narrow win, not replicable": ticket routing is a well-bounded classification problem; shift to fuzzier support conversations or bug investigation and accuracy falls off a cliff. The second critique is sharper: the $30 figure in the case study covers inference cost only—not integration, data cleaning, or ongoing ops iteration. On a total cost of ownership (TCO) basis, mid-size companies' real spend is often 10x that number. We lean toward believing AWS's writing logic naturally surfaces the most flattering data points.
Impact on regular people
For enterprise IT: the "auto-triage + human review" architecture is becoming SaaS support standard—a ticket system plus a model plus a scheduled trigger, no in-house team required. Worth evaluating internal scenarios now.
For careers: support, technical support, and operations—"white-box process" roles—are the first to be absorbed by AI. But "reviewer" roles—people who read model output and make high-risk decisions—will emerge and demand new skill combinations.
For consumers: almost no short-term impact. Triage happens in the backend; what improves is service response stability, not the direct user experience.