Teams Insider

Copilot Practical Applications and Success Stories with Sarah Miller

Sarah Miller is Microsoft's UK product marketing manager for Microsoft 365 Copilot, working with organisations on adoption and value realisation. She breaks down what agents actually are, walks through the spectrum from declarative to autonomous agents, shares how Pets at Home is using them across r

Prefer audio?

Who should listen: Microsoft 365 service owners, IT leads and adoption/change managers planning or scaling a Copilot rollout, and anyone needing to explain agents and their business value to line-of-business stakeholders.

Guest: Sarah Miller — Senior Product Marketing Manager, Microsoft 365 Copilot, Microsoft

Many thanks to Crestron, sponsor of this episode.

Key insights

The framing Microsoft uses internally: "every person has a Copilot and every business process has an agent." An agent is a pre-prompted specialist assistant — e.g. a legal contracts agent that already knows the standard process and output format — built in natural language rather than code, so you stop re-prompting the same process every time. ▶ 5:38
Agents fall on a complexity spectrum: simple declarative agents pointed at a document set or SharePoint (an HR policy agent), task agents built in Copilot Studio where if/then logic is needed (a ServiceNow agent completing an action), and autonomous agents running in the background unprompted — such as lead-generation agents doing continuous market research and surfacing leads to sales. ▶ 9:17
Some of the most commonly built agents are now shipped in the box with a Microsoft 365 Copilot licence — including a prompt excellence agent and the facilitator — because Microsoft saw customers building the same ones repeatedly. ▶ 9:17
Pets at Home has deployed multiple agents: one unifying data from physical stores, the retail website and vet practices so a store leader on the shop floor can ask for stats rather than interpret a dashboard; another assisting retail fraud detection so the fraud team starts from pre-screened suspicious cases rather than zero; plus next-best-action recommendations for customers and patients. ▶ 12:23
Agents can reach into third-party databases, not just Microsoft sources — the value is replacing the "swivel chair" between multiple interfaces with one consistent Copilot experience and a single pane of glass for governance, instead of cobbling together point solutions. ▶ 14:26
The recommended data-readiness path: an IT audit of data security, sensitivity labelling and permissions, then a very small pilot purely to stress-test that permissions hold (users can only surface content they already have access to) — then go full estate at once, so the adoption and change-management spend happens one time and users learn from each other. ▶ 14:56
Adoption advice from promptathons: a prompt that saves one minute ten times a day is a far better starting point than one masterfully written prompt that solves an hour-long problem, because the latter takes much more trial and error. Target low-decision, repetitive tasks first. ▶ 18:03
First practical step for any organisation: switch on Copilot Chat — free, web-grounded, surfaced in Teams, available to anyone with a Microsoft login including frontline SKUs. It starts the adoption habit and keeps prompts inside enterprise-grade protection (nothing entered is used to train the model), which counters bring-your-own-AI shadow usage seen across every demographic, not just Gen Z. ▶ 20:05
For measurement and rollout, the Copilot Dashboard shows which apps Copilot is being used in and identifies hero users to recruit as champions, while the free Copilot Success Kit bundles use-case and persona mapping, launch-day posters and videos, and has been extended with an agents overview guide. ▶ 23:38