Illustrative example
E-commerce · Level 1
Automatic ticket categorization in customer support
The support team manually sorts incoming emails into queues (complaint, product question, order status). AI classifies each ticket, routes it to the right queue, and drafts a reply.
Implementation: a model classifies the ticket’s content and metadata
(including customer history) into one of several categories, sets a priority,
and generates a draft reply for the agent to approve. Typical build time: 2-3
weeks.
Measurable effect: shorter handling time per ticket and fewer misrouting
mistakes during peak hours.
This illustrates a typical engagement at this level, not a description of a completed project for a specific client.
Illustrative example
Logistics / freight forwarding · Level 2
Detecting discrepancies between a customer order and TMS status
A shipping order arrives by email, but the actual status in the TMS sometimes differs from what was promised to the customer. AI compares the order's content against the system state and flags discrepancies before the customer notices.
Implementation requires integrating with the TMS API and the company’s mail
inbox. The model extracts key order parameters from the email (route,
deadline, terms) and compares them against the order’s current state in the
system.
Measurable effect: fewer escalations and complaints caused by discrepancies
noticed too late.
This illustrates a typical engagement at this level, not a description of a completed project for a specific client.
Illustrative example
SaaS / IT · Level 3
A knowledge base that answers in Slack
Internal documentation and the knowledge base are scattered across several tools, and the team wastes time searching for answers. A RAG pipeline connected to Slack answers questions with a link to the source.
Implementation: indexing internal documentation (wiki, documents, historical
tickets), a RAG pipeline with a clearly defined source of truth, and a Slack
bot that replies in-thread and links its source. Critical pieces: monitoring
answer accuracy and an escalation path to a human when the model isn’t
confident.
Measurable effect: less time spent searching for answers, fewer repeated
questions directed at the same people.
This illustrates a typical engagement at this level, not a description of a completed project for a specific client.