Copilot Growth: Low Friction Trials

The Challenge

"Can I just try it? That would be so much better than telling me what I can do."

— User research participant

For commercial Microsoft 365 users, paid Copilot is hard to grasp. Generic value props don't tell anyone what it'll actually do for them. And the standard path is to ask IT or submit a license request, so the people who'd benefit most couldn't decide for themselves.

We tried a lot of ways to explain or show it generally. Feature previews. Painted-door flows where clicking "Upgrade" showed you what would happen if you upgraded. But people wanted to understand how their own work and day could actually change.

Paid Copilot's whole point is access to your Microsoft Graph: your emails, Teams conversations, documents, spreadsheets, and PowerPoints, with a conversation across all of it. Turning that on for one user isn't a switch. Connectors have to populate, indexes have to warm, and every paid query runs expensive inference on the back end.

What we shipped

Low Friction Trials. 30 days of full paid Copilot. No credit card. No admin assignment required. The implicit user contract: experience the value, then make the case to IT to assign you a license.

Research surfaced five trial phases, each with its own user goal. I led design for the on-product moments; the lifecycle email and admin license-request teams partnered in parallel.

Strategic Constraints

Three constraints shaped every decision.

Designing for hesitant users

Research surfaced two primary anxieties.

"Am I allowed to do this? Should I tell my manager?"

— User research participant

The first was permission. A "free trial" sounds like a future bill, even when it isn't, and people weren't sure whether starting one was something they should be running by their manager.

The second was privacy. Paid Copilot reads your private files and conversations. Even when your admin already has access, "Copilot is now reading my email" lands differently as a feature than as a fact.

We answered both directly in the in-product copy.

Permission. The offer-screen subline carries it in one breath: "…approved by your business, no payment needed." Both reassurances land before the user has reached the buttons.

Privacy. The welcome state spells out what Copilot can and can't see, so the user doesn't have to ask.

But I don't think any in-product copy can fully convince someone who's hesitant, no matter how well it's written. That belief pushed us off the screen. A parallel workstream let IT admins decide whether to surface the trial to their users at all. When the trial shows up because your own company turned it on, half the permission question is already answered. And at the end of the trial, instead of leaving the user to handle conversion alone, the experience hands the conversation back to IT.

A signal from inside the org beats anything Microsoft can say in-product.

What do I do with this?

The same people who told research "I just want to try it" often had no idea what to do once they had it.

Copilot's best moments are personal: combine these two files into a presentation; summarize what my manager said this week. That makes a suggested prompt risky. Putting "Try Researcher!" on the homepage and getting back twenty minutes of irrelevant output would have hurt the value we were trying to prove. A bad nudge is worse than no nudge at all.

Time-to-value first

One task in the checklist is something Copilot almost always nails: "Show me five cool things you can do for me." Copilot already has your role, your manager, your company. The answer comes back specific to who you are. That's the trial's first impression, and it's a safe one.

Educational scaffolding

Four prompts isn't a lot of room. The split was my hypothesis going in: two prompts that actually get something done for you, and two that quietly teach you how Copilot works. Research and content signed off, the team agreed, and that became the brief.

The doing prompts came from where the data already pointed. Recap /meeting showed up over and over in our DS and UXR readouts — one of the most-used features in the product, and one of the easiest to get a useful result on the first try. Prepare a report from /file is the same idea with more ambition. Harder to nail. Bigger payoff when it does.

The teaching prompts do two things at once. They give people the personalization users keep asking for — every answer is grounded in who you are and what you do. And they teach the move underneath: Copilot itself is the place to ask "how do I use this thing?" That's a skill that pays out long after the trial is over.

Of the two, five cool things has been the breakout. It works across surfaces — users get a personalized list back and it's their first real "oh, this is for me" moment with Copilot. Agents are the harder one. A meaningful chunk of what users are paying for, and most of them have no idea agents exist. Pointing at a specific agent would have repeated the Researcher problem, so we ask Copilot itself: "How can I use agents in my role?" The answer isn't always perfect. For someone whose first reaction was "what is this," it's still a useful place to land.

We could have personalized the prompts themselves. The machinery existed. We didn't need it. The four prompts stay the same for everyone, but Copilot already knows your role, your team, your company, so the responses come back custom anyway.

The Word version uses the same shape. Two doing, two teaching, Copilot doing the personalization underneath.

The cohesion job

Some of the most important work on this project was the least visible.

The trial touches the Copilot app, Word, Excel, PowerPoint, mobile, the transaction dialogs that handle license assignment, onboarding side panes, and the IT admin's license-request flow. Each surface has its own design partner, PM, engineering team, and release cadence. A parallel team was independently building their own side-pane for a different onboarding effort that ended up overlapping with ours; we're in the process of merging them.

A lot of the design work was being the place where coherence lived. That meant pulling stakeholders into the same room, watching for decisions on one surface that would break another, and pushing back when "we can't" was really "we won't." When the mobile design team picked up their version of the trial, they used my framework as their starting point. That's when I knew the work had landed.