Tristan Gayford, Microsoft Teams Product Manager at EY on Teams, UC and AI Journey

Prefer audio?

Who should listen: Teams and M365 service owners, product managers and IT finance leads in large, regulated organisations working out how to govern Copilot in meetings and fund consumption-based AI.

Guest: Tristan Gayford — Microsoft Teams Product Manager, EY

EY runs Microsoft Teams for more than 400,000 people in well over 100 countries, each with its own regulations, budgets and risk appetite. Tristan Gayford has worked in the space since the OCS days and now manages Teams there. He walks through how EY handles transcription and Copilot in meetings across different regulatory regimes, and why the move from predictable per-user licensing to consumption-based AI billing means the business, not IT, has to own the cost.

Many thanks to Landis, sponsor of this episode.

Key insights

• Gayford has been in the space since 2008, when he joined Deloitte and was asked to run a training course on OCS. He said yes, then looked up what OCS stood for (Office Communications Server). ▶ 1:08
• EY operates in well over 100 countries, so the Teams service has to cope with differences in local technology, regulation and the business itself. Consultants in technology want to be on the bleeding edge because that is what their clients are asking to see. ▶ 2:14
• Professional services firms look like one organisation but run as partnerships, each with its own cost centres, budgets and readings of regulation. In the early days of cloud that meant data sitting in the US or perhaps Germany, which made firms hesitant, and Gayford admits EY was not on the bleeding edge of the move to cloud. ▶ 6:18
• EY has rolled out Copilot since pretty much day one, expanding location by location as each gets comfortable, with Teams meeting recaps and notes as a key driver. Tom notes the firm has also made a public bet on E7 across the board. ▶ 11:51
• Recording, transcription and AI decisions pull together regulatory requirements in some countries, differing risk appetites by location and service line, and clients who either want a documentary record or explicitly do not. Gayford says the answer is guidance and change management, otherwise controls such as muted microphones when Copilot starts just look like IT throwing spanners in the works. ▶ 13:57
• The during-the-meeting-only option for Copilot has been 'an absolute godsend', because retained transcription still carries risk exposure many people do not want. Users capture the notes at the end of the meeting and copy them out, and can switch Copilot off entirely for HR conversations or one-to-ones. ▶ 17:27
• IT funding has been per-user and predictable for years, so a consumption cost on top is 'jarring'. With Copilot Cowork and other consumption-billed AI, the technology group signs the cheque but is not the one driving the usage: Copilot's flat rate balances out heavy and light users across 100,000 people, while consumption billing leaves IT with costs that are potentially astronomical, unbudgeted and unpredictable. ▶ 21:03
• Tom pushed Copilot Cowork as hard as he could while it was in the Frontier Program and ran a single query that cost $270. The output was good enough that he could justify it to a CFO, but the question is how anyone makes that call at scale when the spend was never budgeted. ▶ 22:33
• Gayford agrees with Tom that Cowork is effectively an employee: 'my Cowork is my employee', and he has to justify its cost and output to his manager, who justifies it further up the chain. IT does not carry the business's people costs, so the business should own AI credits too. ▶ 24:32
• Users cannot see what a prompt will cost before they run it, or control which model it uses, and only find out afterwards. A misplaced word can send a prompt off on a tangent costing hundreds of dollars, so Gayford wants more upfront transparency on how prompts will be charged. ▶ 26:32
• Asking business users to pick between models does not work: Tom points out that a user who does not own the budget will always pick the best one, and Gayford admits to sliding it up to 'ultra high' himself. On being told to just choose auto, his view is that if auto can work out the right model, it should be able to estimate the cost as well. ▶ 28:26
• Copilot at $30 a user once felt expensive and now looks like a bargain, because users can be told to use it as much as they like with no consumption cost. Gayford finds himself pushing Copilot as far as he can to keep spending on the fixed-licence side of the equation. ▶ 31:09

Insights summarised by AI from the episode transcript, reviewed by the Empowering.Cloud team.

Listen: Apple Podcasts · Spotify · Other platforms