A GrowthPad session recap with Vikas Kansal Product Lead for Gemini AI Subscriptions at Google
Guest: Vikas Kansal, Product Lead, Gemini AI Subscriptions · Google
The Day-Zero Monetization Problem
In most consumer technology businesses, the playbook has always been the same: grow users first, monetize later. AI broke that assumption on arrival.
Vikas joined Google a decade ago as a software engineer, moved into product management through Google’s storage subscriptions business, and found himself at the centre of AI monetization two and a half years ago — when the cost reality of large language models made “wait and monetize later” impossible.
“Monetization was required across the industry — ChatGPT, Anthropic, Google — anybody using AI needed to start monetizing at the get-go. One, the supply constraints. Two, it was an expensive technology.”
That framing — scarcity of compute, high cost to serve — shapes almost every decision Vikas and his team make, from acquisition campaigns to pricing tier design to churn management.
Acquisition: Trials, Partnerships, and Skin in the Game
Traditional subscription acquisition relies on a clear value proposition locked behind a paywall. With AI, the free product itself is often so capable that the conversion funnel becomes genuinely difficult. Vikas described a two-layer approach: growing the top of funnel by working closely with product teams on consumer-facing apps, and then running targeted trial campaigns to drive what he called “instant magic” — moments where users feel the product’s power directly.
Google ran a 12-month free trial for students in India, partnering with Jio Telecom — a move almost unheard of in traditional SaaS. Vikas explained the logic: long-term ecosystem retention, not immediate revenue. But perhaps the more counterintuitive finding was this: free trials often underperform versus discounted paid entry points.
“Users who put some skin in the game — even at 50% or 75% off for a month — retain better than users on a free trial. There’s not enough urgency to actually use the product when you’re not paying anything.”
On paid channels, Google runs performance campaigns across YouTube, Instagram, and Google Discover, particularly tied to moment-based offers like New Year promotions or student campaigns. Partnerships with telcos and OEMs add local-market reach that direct channels can’t replicate, along with benefits like existing billing relationships with end users.
Why the Traditional Freemium Model Collapses
In conventional SaaS, the cost to serve additional users is nearly flat. Adding another Zoom seat, another Slack user, the incremental infrastructure cost is minor. AI inverts this. Every query triggers a compute event. Every agent step burns tokens and generates a cloud bill. As multi-agent workloads grow, costs scale directly with usage rather than flattening.
This makes the classic freemium model hide premium features behind a paywall, offer a generous free tier to grow users, structurally unsound. The free tier costs money at scale. The premium tier has to recoup it. And because the free tier is itself so powerful in AI, users have less incentive to upgrade.
Subscription vs. Usage-Based Pricing — It’s Not Either/Or
One of the more nuanced points Vikas made is that the “subscription vs. usage-based” debate is largely a false choice. Subscriptions are usage-based pricing just bucketed. The real question is which customer segments you’re serving and how they think about cost predictability.
Consumer users, solo builders, and hobbyists typically want fixed, predictable costs. Google’s Plus, Pro, and Ultra tiers serve this segment. Power users and professionals are more comfortable with variable costs paying for usage above a baseline when they need it.
Vikas sees seat-based pricing evolving rapidly, particularly in the agentic era: charging a fixed seat price when a user can spin up 100 simultaneous agents creates obvious fairness problems. Usage pooling across agents, with a consumption layer on top, is the likely direction.
How AI Pricing Gets Decided (It’s Messier Than You Think)
Vikas shared a well-known story in AI circles: ChatGPT’s original subscription price was reportedly chosen through a Reddit poll with around 100 responses. The options were $10, $20, or $50 per month. The industry converged on $20, and has mostly stayed there for flagship tiers, not because of rigorous conjoint analysis or willingness-to-pay studies, but because the economics are still being figured out in real time.
Traditional pricing research tools haven’t disappeared, but their usefulness is constrained. By the time a study is designed, deployed, and analyzed, the product and competitive landscape have already moved. Competitive benchmarking and elasticity modeling are doing more of the work.
The cleaner finding from experimentation: higher discounts bring lower-intent users. Retention is consistently better among users who signed up at or near full price.
Retention: The Playbook Is Mostly the Same, the Stakes Are Higher
On churn management, Vikas drew a sharp line between voluntary and involuntary churn — and had strong language about the latter.
“Involuntary churn is a sin. A user didn’t want to churn, their payment failed or their card expired. That’s completely unacceptable. These are predictable events. You should be handling them proactively.”
On voluntary churn, the AI context makes things harder. The free tier is good enough for most use cases, switching costs between AI platforms are genuinely low (especially for developers whose code lives in GitHub), and AI is not yet a utility the way cloud storage or music streaming is.
Google’s response: lean into bundling. Including cloud storage inside the subscription raises the retention floor significantly, storage is a necessity in a way that paid AI still isn’t for most users. Personalization and in-product memory are the longer-term bet for raising switching costs.
The Future: From Scarcity to Abundance
Asked where AI pricing goes in five years, Vikas was candid that nobody has a reliable forecast but offered a directional view. Right now, demand for tokens far outpaces supply. Data center and chip constraints keep costs high and create the token anxiety many users feel usage limits, paused tasks, “come back in five hours” messages.
As the infrastructure buildout accelerates, Vikas expects that anxiety to give way to abundance. Pricing will still exist, but the spirit behind it should shift from managing scarcity to enabling access.
He also raised the longer-term question of what software even is in an AI-native world pointing to concepts like appless phones where an AI operating system handles tasks that currently require separate applications. For founders and builders, his advice is unchanged: lower build costs don’t solve the problem of knowing what to build. User empathy and problem clarity remain the irreplaceable input.
Key Takeaways
- Monetize at day zero. AI compute costs are too high to run a “grow now, monetize later” strategy. Paywall design starts at launch.
- Discounted paid beats free trials. Users with even a small financial stake engage more and retain better than those on pure free trials.
- Subscriptions and usage-based pricing aren’t opposites. Design buckets for predictability-seekers; add consumption layers for power users.
- Seat-based pricing will evolve. In an agentic world, usage pooling replaces per-seat models — you can’t charge a seat fee for an ephemeral agent.
- Involuntary churn is preventable. Card expirations and failed payments are foreseeable. Handle them proactively, every time.
- Bundling raises the retention floor. Pairing AI with utility products (storage, health) gives users a reason to stay beyond AI value alone.
- The build cost has fallen; the idea cost hasn’t. Cheaper and faster building doesn’t tell you what to build. Problem clarity is still the bottleneck.
Audience Q&A
Q: You’ve talked about using discounted trial pricing to bring users in. How do you handle the expectation that the lower price will continue? When users see the full price coming, what do you do?
Transparency is the foundation. When users sign up on a trial, we’re upfront: “You have 50% off for three months. After that, the price is X.” Some users do set a deferred cancellation the moment they sign up — they know they won’t pay full price.
Before the transition, we send a clear nudge to all users: your price is changing next week to $X, take action or the new price takes effect. For users who are pending cancellation, we also share a summary of their usage — what they’ve done with the product, how much value they’ve generated — alongside an encouragement to use it more before the price changes.
The framing matters too: this isn’t a price increase, it’s an introductory period ending. Users knew from day one what the real price was going to be.
Q: What’s the most interesting way you use Gemini in your own work as a PM — beyond PRDs and meeting transcripts?
Meeting prep. We have an internal tool where Gemini is connected to my calendar, my past interactions with a group, the agenda, and my private talking points. Two minutes before a meeting, I get a full brief: prior conversations, context, what I want to say. It’s like having a business partner without actually having an assistant. Meeting prep has essentially stopped being a manual task for me.
Rapid Fire
| Question | Answer |
|---|---|
| Most underrated AI company right now? | Lovable |
| One skill every PM should learn? | User empathy |
| Favourite Google product? | Photos |
| Biggest pricing lesson? | Free is good enough |
| What will surprise people most about AI in the next decade? | Abundance |
| Where would you build tomorrow? | Gemini on AI Studio |
Vikas is active on Twitter / X and building a community around AI monetization — reach out via DM. More GrowthPad sessions are coming. Stay tuned on growthpad.blog.
Leave a Reply