Growth marketing in the agentic era: what happens to CAC, ARPU and retention when an AI agent does the browsing
My view is agents wont kill upselling. They will kill lazy upselling and whoever learns to sell to a machine first will win the growth game.
Take a simple example. A user tells her AI assistant "order dinner from the usual Thai place under ₹600 and top up groceries for the week". Two orders get placed. She doesnt open Zomato or Zepto at all. So she never sees the banner, the "you might also like" rail, the ₹49 chocolate at checkout or the Gold membership nudge.
Every growth team in India has built everything on one assumption that customer will come and see our screens. This is going to break for large share of transactions and what you decide in next 12 to 18 months will decide if its a threat or opportunity for you.
What changes with agents
Today every screen in the funnel is a monetisation slot. Home feed, search, product page, cart, checkout. Agent compresses all of it into intent, shortlist and transaction. The screens in between are never even rendered.
1. There are now two customers. The human who sets the rules responds to brand and habit. The agent who executes responds to price, reliability and policy.
2. Attention doesnt matter much. Retail media, restaurant ads, financial cross-sell all depend on someone scrolling. Agent doesnt scroll past a sponsored listing, it just filters it out.
3. Comparing is free now. Human compares two apps on a good day. Agent compares every provider on every order. Whatever you win today because of friction, you lose.
Will agents kill browsing and upselling
No. What agents will kill is chore browsing and placement based upselling. Chores like groceries, recharges, bills and daily cab go to agents first. Considered chores like insurance or a fridge get shortlisted by the agent and approved by human. Pleasure stuff like a new restaurant, fashion scroll or a holiday stays human because browsing itself is the fun.
The ₹49 chocolate at checkout is in trouble for sure. But a membership that provably saves the user ₹300 a month will get recommended by the agent faster than any human would have noticed it.
What it means for the big consumer apps
- Amazon. Most exposed because ads assume a human on results page, thats why it sued Perplexity over its Comet agent. Its playbook is walled garden and push its own agents.
- Flipkart. Agent compares prices perfectly so commodity margin goes. Defence is exclusives, private labels and Flipkart-only finance.
- Uber. Rides become an auction across Uber, Ola, Rapido. But agents read calendars and book rides the night before, which is great for planning supply.
- Zomato and Swiggy. "The usual" is perfect agent prompt. Let agents own reorders, keep discovery human and make Gold or One the agent's default.
- Zepto, Blinkit, Instamart. Weekly pantry goes on autopilot. Impulse add-ons die but whoever agent trusts on fill rate gets the whole basket.
- Paytm. If agents pay via UPI directly the payment screen goes unseen. Opportunity is becoming the place where people set rules and limits for their agents.
- CRED. Agent pays the bill on due date and never visits. But CRED members are exactly who will delegate money first, so CRED can become their financial operator.
Impact on CAC
Acquisition now means getting on the agent's shortlist and getting the user to link you once. Agents reward what they can verify.
- Be callable. Ship an MCP server or agent API for search, order, track and cancel.
- Be legible. Real stock, exact delivery windows, total price with fees. Hidden fees get you marked unreliable.
- Be predictable. Return and refund policy as machine readable rules.
- Be fast and correct. Agent keeps score and a wrong item costs more than a slow page ever did.
The new install is the account link. Optimise it like you once optimised onboarding. And paid acquisition moves from cost per click to fee per completed order, which is better economics if you watch the take rate.
Impact on ARPU
Agent is the most rational buyer you will ever deal with. It wont buy on impulse but will buy anything provably good for the user. Ask for upgrades at the moment user sets standing rules, not at checkout. Pitch memberships as savings ("Gold would have saved you ₹412 last month"). Move ad money into agent-visible offers and sell bundles instead of add-ons.
Charge premium, agent users can afford it. Think about who is actually using agents today. Its mostly well paid folks who value their time way more than ₹50 here and there. They are not using an agent to find the cheapest option, they are using it so they dont have to think. Credit to my friend Kuldeep for this thought, he was the one who pointed out that brands can actually charge premium from agents.
- Make agent access a pro feature. Want your agent to order from us? Thats in Pro plan only.
- Nudge the agent towards the expensive option. Make your ₹780 option look best for a "best under ₹800" rule. Agents can be gamed same way SEO got gamed.
- Let your agent talk to their agent. Your brand agent negotiates, offers bundles and says "upgrade to annual and save 20%". Two bots haggling and yours is trained to upsell.
- Sell priority. Faster slot, first pick in a sale, no substitution. Well paid users will set rules that allow it.
Early on this audience is the best margin you will ever get from agents. Dont waste it by discounting.
Impact on retention
Retention used to mean habit. Now it means being the default in user's standing instructions, and nobody changes a default till something breaks. You cant push notify an agent back, so reliability matters most. Every missed slot or wrong substitution gets logged and the next order quietly goes elsewhere. Still keep a direct line to the human through brand, because brand is what makes user tell the agent "use Zomato" instead of "find cheapest".
What I would do now
- Measure what share of orders already come from agents.
- Ship an agent API or MCP server and give it a PM.
- Audit pricing for every hidden fee a machine would flag.
- Rebuild membership pitch as maths for every user.
- Pilot agent-visible offers with top advertisers.
- Decide if you are open, metered or walled for third party agents.
And track these metrics:
| Metric | What it tells you |
|---|---|
| Agent share of orders | How much of your business already runs without your screens |
| Account link rate | Users who connected you to their agent |
| Shortlist inclusion rate | How often agents consider you |
| Agent win rate | How often you win once shortlisted |
| Default rate | Share of users with you in a standing instruction |
| Failures per 1,000 agent orders | Predicts churn before it shows up |
| Agent take rate | What agent platforms charge per order, your new CAC |
Final thoughts
Growth teams who keep optimising banner CTRs while agents go around them will lose. Agents remove friction and what remains is the oldest growth strategy. Be the better deal, be reliable and make people like you enough to write your name into their rules.
