Skip to content

TELUS Digital Agentic AI Delivery Playbook

Executive Summary

This playbook describes how I would lead agentic AI adoption without turning it into a tool-first experiment. The operating idea is simple: start with high-friction workflows, prove value with controlled pilots, measure business impact, and scale only when reliability, ownership, and governance are visible.

What This Artifact Is

This is not a fake transformation deck or a code demo wearing leadership clothes. It is a compact delivery packet meant to show how I would structure an Agentic AI and Automation practice: where to start, how to govern it, how to staff it, and how to decide whether it is working.

Delivery Principles

Principle Meaning
Workflow before model Define the business process, handoffs, risks, and success criteria before choosing tooling.
Evidence by default Every agent action should leave a trace: inputs, retrieved evidence, decision, action, and reviewer.
Human control at risk boundaries Autonomy is acceptable for low-risk routing and drafting; high-impact actions require approval.
Platform consistency Teams should share patterns for prompts, tools, evals, observability, and access control.
Measured scale Expand from pilot to portfolio only when KPIs show time saved, quality preserved, and risk reduced.

First Use Cases

These are selected for a telco/digital-services environment where customer trust, regulated data handling, and high-volume support operations are the operating reality.

  1. Customer support triage: route incoming service requests to the right queue with a draft response and severity flag, keeping a human reviewer in the loop for escalations.
  2. Internal knowledge routing: answer employee questions about policies, tools, and processes from a governed document corpus, with full retrieval evidence logged.
  3. Incident summarization: convert alert streams and on-call notes into structured incident summaries with suspected owner, impact scope, and recommended next check.
  4. Quality evidence generation: produce QA and data-quality reports for client-facing delivery milestones, with traceable inputs so audits are straightforward.

Governance Rhythm

Cadence Meeting Output
Weekly Delivery checkpoint Pilot status, blocker removal, metric review
Biweekly Risk and evaluation review Failed cases, unsafe action attempts, policy updates
Monthly Executive readout KPI trend, adoption progress, scale/no-scale decisions

One-Page Adoption Pattern

Select one workflow, baseline the current pain, build a narrow agentic pilot, instrument every step, compare against baseline, and graduate only when the workflow has a named owner, eval suite, runbook, and rollback path.