WISE – Mona AI Agent

Background

In this AWS agentic AI case study, WISE‘s operations team was running its day-to-day processes by hand. WISE is a telecommunications and managed IT service provider headquartered in Beirut, Lebanon, and a subsidiary of the Debbane Saikali Group, serving enterprise clients and residential communities across the country.

For every service order, the account manager composed and tracked several supplier emails, NOC messages and billing emails, then followed each one up personally. These steps were spread across Slack channels and email, guided by documented SOPs that people had to remember and apply correctly each time.

Moreover, many requests waited days on a supplier reply or an internal approval, so work had to be tracked manually and could stall without anyone noticing. Furthermore, some steps carry real cost when they go wrong, such as de-installing a client that still has an active WAN link, acting twice on a duplicate request, or sending an incorrect date to a supplier.

Finally, with so many manual touchpoints, there was constant room for missed steps, wrong information and forgotten follow-ups, which slowed service delivery and consumed the team’s time.

Proposed Solution

In this AWS agentic AI case study, Nebulane developed Mona, an AI agent that runs WISE’s operational workflows by following the company’s own SOPs. Each workflow was agreed with WISE as a process diagram before it was built, and Mona now covers more than twenty request types, from supplier communication to billing and service requests.

Mona works inside the tools the team already uses. Requests arrive through Slack and email, and Mona identifies the request type, gathers the details and drafts each outbound step, such as a supplier email, a NOC message or a billing email. The drafts are posted to a Slack review channel, where a team member approves them with a single emoji reaction, so nobody needs to learn a new system.

The agent runs on Amazon Bedrock AgentCore Runtime and uses Amazon Bedrock with two models split by task: a stronger model for understanding requests and a cheaper one for high-volume routine work. Rules that must always hold, such as duplicate detection, date validation and send-rate limits, are enforced in code rather than left to the model.

Because many requests wait days on a supplier or an approval, every workflow is checkpointed in Amazon DynamoDB, so it resumes at the exact step it stopped on. Mona chases follow-ups on a schedule, and events arriving through Amazon API Gateway are buffered in Amazon SQS so none are lost.

The level of autonomy is a setting, not code. WISE keeps human approval on supplier email, NOC messages, billing email, credit notes and equipment orders, and can loosen or tighten each gate without a new release.

Every message, state change and approval is recorded and visible on a live dashboard, including a Health view that flags stalled work. The whole solution runs on fully managed AWS services, with no servers to maintain.

Metrics for Success

The success of Mona was evident through several key indicators. Manual handling per service order dropped sharply: instead of composing and tracking every supplier email, NOC message and billing email, the account manager now only reviews and approves drafts with one emoji, while Mona chases follow-ups on a schedule.

Human errors were reduced as the agent runs each workflow step consistently according to the defined SOP, leaving less room for missed steps, wrong information or forgotten follow-ups. Because every message, state change and approval is recorded per workflow, the time and number of manual actions per request can be read directly from the audit trail rather than estimated.

The system also proved reliable in daily operation. No events are lost overnight, long-running requests resume where they stopped, and over 1,200 automated tests run before every change. Running costs were kept in check by scheduling the agent around working hours, after billing showed that most of the runtime cost came from idle time.

Conclusion

This AWS agentic AI case study demonstrates how Nebulane helped WISE turn manual, SOP-driven operations into a reliable automated workflow, while keeping people in control of every step that matters. Mona’s integration with existing tools, durable workflows and configurable approval gates reduced manual effort and errors without asking the team to change how they work. These results validate agentic AI on AWS as a practical, high-impact capability and a strong foundation for automating further operational processes.
Discover how your business can benefit with our free cloud and AI assessment.