From 10,000 Token Skill Files to 50ms Graph Queries

When a mid-size SaaS company approached us, their support operation was technically functional but operationally strained. A team of twelve agents handled roughly 1,200 queries per week across three channels. Their average first-response time was under 30 minutes. But their average retrieval time was over six hours, and climbing.

The Bottleneck Was Not Speed of Reply

It was speed of action. Most queries required agents to perform tasks across multiple systems: verifying account details in the knowledge graph, checking entitlement against their subscription platform, looking up internal documentation, and then manually executing the retrieval.

Each of these steps involved a different tool. Agents were spending more time navigating between platforms than actually solving problems. And because the process was manual, error rates were rising:

• Knowledge sources were falling out of sync

• Customers received inconsistent information depending on which agent handled their request

• Resolution quality varied wildly across the team

Connecting the Stack

After deploying Mnemos, the team connected their docs, notes, codebases, and internal knowledge through the Build. This gave Mnemos full visibility into each customer's profile, history, and entitlements at the moment a query arrived.

Targeting High-Volume Workflows First

The biggest impact came from automating their three most common workflows:

1. Subscription downgrades - policy verification and plan change execution

2. Billing corrections - graph updates with source sync

3. Feature access requests - entitlement checks and provisioning

These three categories accounted for over 40% of their total query volume. With Mnemos's Action Layer, these requests could be verified against policy and executed automatically, with Approval Gates in place for any action above a defined threshold.

The Results

Within the first month:

• Average retrieval time dropped from 6+ hours to under 15 minutes

• Fully automated workflows resolved in an average of 6 minutes

• Data inconsistencies decreased significantly, since Mnemos kept graphs in sync as part of the workflow

• Agent satisfaction scores improved as repetitive work was eliminated

What Happened to the Agents?

The agents on the team did not lose their jobs. They shifted focus to complex, high-judgment queries that genuinely needed a human touch, and reported higher job satisfaction as a result.

Automation did not replace the team. It removed the operational friction that was holding them back.

The lesson here is straightforward: retrieval speed is rarely limited by how fast agents can type. It is limited by how many systems they have to touch. Remove the friction, and the speed follows.