Last month at the European SharePoint Conference, I had the privilege of presenting alongside Jake Harvey on how we at Advania UK are delivering real value with Copilot Studio. Our session: “Real-World Success Stories with Copilot Studio”, was about practical wins, hard-earned lessons, and a repeatable approach to intelligent automation.
The Mental Model: Strategy → Capability → Solution
- Automation is the strategy.
- AI is a capability choice.
- Agents are a solution choice.
Keeping these lanes clear helps teams avoid “AI for AI’s sake” and select the right tool for the job. Our section on this (shown within the slide below), outlines some of the choices we have at hand. Of course, as this was a session on Copilot Studio, we’ll be covering the left-hand swim lane.
Deterministic + Generative: How to Choose the Right Tool
All AI is automation, but not all automation is AI. The sweet spot is combining rule based (deterministic) workflows with generative (non-deterministic) capabilities when the problem benefits from reasoning or language understanding. Use deterministic flows for consistency, auditability, and control; bring in generative steps when you need interpretation, summarisation, or bringing together non standardised data to make a decision.
We showed several implementations that follow this pattern. But a high-level overview of this can be seen in the slide below:
What This Looks Like in the Real World
We see automation projects falling into 3 main experiences:
1) Personal Automation – e.g. Project Tracking
A lightweight agent accelerates project tracking by pulling context and facilitating updates, keeping humans firmly “in full control” for judgment calls.
2) Defined Automation – e.g. 1st Line Support
We implemented a deterministic backbone (triage, routing, known error responses), augmented with agent skills for knowledge retrieval and follow-ups, reducing handoffs and time to resolution while maintaining reliability.
3) Intelligent Automation – e.g. Invoice Processing
A deterministic workflow orchestrates document parsing (using traditional OCR extraction), validation, and posting, with agents used for exception explanation and reconciliation notes, improving quality without compromising control. As this is autonomous, a “Human in the loop” connector is used to ensure that a human always has the final approval on all entries to the system of record.
Project Spotlights
Hiring Experience – Human Centred by Design
The agent approach supports candidates and managers end-to-end: Q&A, interview prep, insights, and follow-ups, keeping inclusivity and transparency front and centre while preserving human oversight.
Compliance in Financial Services – From Months to Hours
- Scale: 25,000+ users; 5,000+ Governance, Risk, & Compliance controls assessed.
- Before: A manual quarterly process taking ~3 months, often outsourced (£15k per quarter / £60k per year).
- After: Custom prompt framework + Copilot Studio agent + Power Automate ingestion reduced the cycle to ~2 hours (~£200 per run), with bulk/individual scoring and extensibility for new metrics.
Patterns We’ve Found Useful
1. Start deterministic, add generative where it moves the needle
Use rules for compliance, consistency, and SLA commitments; layer AI where judgment and language help.
2.Adopt an ecosystem mindset
Pair Copilot Studio with M365 Copilot, SharePoint Agents, and Power Platform to match the use case, from focused retrieval to end-to-end RFP support. The same use case can be built across technologies depending on your end users needs.
3. Keep humans in the loop
Review/approve steps at the edges (customer communications, compliance scoring) to maintain trust and accountability.
4. Data quality is a prerequisite for AI success
If your data isn’t ready, begin with traditional automation to create reliable pipelines and canonical sources, then switch on AI.
A huge thank you to everyone who joined, asked questions, and shared experiences. The most valuable part of this journey as we’re adopting these new technologies is the collective learning.