AI activation
2025
A PLG system from scratch: Sage Copilot activation in Sage 50 Accounts
Sage Copilot was launched in Sage 50 Accounts with three new feature emails. Before 75k users could see any AI value, they had to complete five-step setup, including moving their whole team's login to the cloud after 10+ years on the desktop software. It became clear the problem wasn't awareness; users needed proof that AI was worth the change before committing to it. That insight led me to build a PLG system from scratch: in-product guidance, email flows, value content, and the measurement to learn what worked.
Problem area
The hardest onboarding in the product
Before seeing Sage Copilot work, users had to complete five steps, including migrating their whole team's login to the cloud. They'd chosen desktop software deliberately, so it felt like risk with no visible reward. The launch had awareness but a baseline activation of around 3%. No support team could walk 75k users through it, so the product had to. But there were no PLG channels beyond three new feature emails, and no unified product analytics to show where users dropped.
Onboarding flow

Where design could move it
Map the funnel before trying to move it
Data sat across disconnected systems, so no one could see where users dropped or tell if a change worked.
Design decision 01
Building a funnel from disconnected data
Product analytics data sat across Pendo, internal engineering dashboards, Google Analytics and Power BI, with no unified view. By collaborating with Product Manager and Data Analyst, I designed a manual and AI-assisted ways to structure them into one funnel, which was showing activation rate of around 3% at the start. It took 2 weeks for data analysis before any visible design, but without it every experiment would have been guesswork.
Design decision 02
Messaging created for setup state
I was handed three new feature emails sent to the whole user base. But before anyone could set up Copilot, they had to be cloud connected and on the Copilot-ready desktop version. I segmented the 75k users by which setup requirement they hadn't met, and reached out each cohort through email and in-app messages with the step they needed next. It meant more flows to build and track than three emails, but every user got a relevant message, and every step became measurable.

Email content used in the flow
Design decision 03
Intent over traffic
My first hypothesis was that guides on the highest-traffic screen would reach out more users. They did, but engagement barely moved: people on a general screen weren't in the mindset for a five-step setup.
Round 1 assumed a linear journey. Users would update to the new version, land on the Customers screen, and be ready to finish setup. So the guide picked up where the update left off, telling them how to activate Copilot next. It reached plenty of users, but few acted on it. They hadn't come to that screen to set anything up, and the message asked for effort before showing any reason to give it.
In Round 2, I started from what users were already trying to do. The guide moved to the Transactions screen, where small business owners go when they're checking their numbers. The content changed with it as well: instead of setup steps, it showed Copilot catching errors in transactions they were reviewing. Setup came last, as a single next step. Reaching fewer users at the right moment outperformed reaching many at the wrong one.
1
Placed by intent
Round 1 assumed users had just updated and would land on Customers ready to finish setup. Round 2 met them on Transactions, where they came with a question about their numbers.
2
Framed by its benefit
Instead of setup steps, the guide led with an outcome, catching errors before they cost time or money, and showed Copilot doing it on transactions like their own.
3
Setup showed up last
One clear next step after the value, so the ask for effort follows the reason to give it.
Retrospective
Onboarding flow itself as the constraint
Impact
Guide to completion
57% of users who interacted with the guides were more likely to complete 5-step onboarding





