Giving sales reps something to say, mid-call
Salesken.ai·5 MONTHS
Co-Browser, a real-time AI assistant that surfaces the right prompt, product context, and next step while a rep is still on the phone, designed end-to-end from research through shipped UX.
- Avni Garg (Me)Product Designer
- Raj SinghSenior Backend Developer
- YogeshSenior Frontend Developer
1. Who, Why & What
Salesken’s own app needed the same fix it was meant to give customers. Scheduling calls, checking deal insights, and getting real-time tips meant hopping between separate tools mid-workflow, a messy setup that cost reps time they should’ve spent talking to the customer, not searching for what to say to them.
“There are moments mid-call where I just don’t know how to convince the customer, and that’s exactly when I lose them.” That was the pattern behind lost conversions, not a knowledge gap, but a freeze, at the one moment a rep couldn’t afford it.
Co-Browser is a two-sided product, an agent on a client company’s sales floor, and a customer who registered for a demo on that company’s site.
Customer (Demo Registrant)
Registers for a product demo on a client’s site, e.g. a policy demo on Policy Bazaar, then joins a live session and needs to follow the walkthrough clearly, with zero technical setup on their end.
Sales Agent
Picks up the session, shares their screen, and needs to see exactly where the customer is looking or pointing so they can respond to what the customer’s actually asking about, in real time.
I didn’t have access to Salesken’s own reps as interview subjects, so research ran on two tracks in parallel: a competitive teardown of how the market already solved this, and primary interviews with sales reps at other companies to get an unbiased read on the job itself.
Competitive teardown
What I did2 weeks. Watched 10 product-walkthrough videos from competing platforms, used AI transcription to pull structured feature notes from each, and built a start-to-finish comparison in Notion.
Primary interviews
What I didInterviewed sales reps at other companies directly about call volume, daily workflow, and where they got stuck; since they weren’t Salesken customers, this stayed independent of any one product’s framing.
Three connected surfaces, not one screen; one for the customer, two supporting the agent through the session.
Co-browsing session
The customer registers for a demo, the agent picks it up, and the two share one screen: full view for the customer, cursor-only visibility for the agent.
In-session assistance
A live panel for the agent with product and plan details plus suggested talking points, turning into a summary once the session ends.
Live agent monitoring
An ops-facing view of every agent's status, lead, and session duration, so incoming demo requests always land on someone available.
A co-browsing session, not a phone call; the customer and agent share one screen, but see different things on it.
Agent guides the customer live through the demo, responding to exactly where they’re pointing.
2. Product Scoping
User story
3. Design Evolution
Several rounds, directly in Figma.
I designed across three tracks, the customer view, the agent view, and how they connect. Both the layout and the information on screen changed a lot as I iterated.
Early versions packed too much into one panel, so I broke it into clearer, focused sections. It took several rounds before the structure felt right. The final design kept only what the agent needed in that moment, and moved the rest into separate tabs.
Stakeholder Review
Two asks came out of the same review. First, agents were juggling a full day of Google Calendar meetings with no way to tell customer sessions apart from internal ones at a glance, so a “Smart To-Do” surface was added that pulls the agent’s calendar and separates upcoming customer meetings from internal meetings automatically.
Second, stakeholders wanted agents to walk into a call already knowing who they’re talking to, so a lookup tab was added where the agent can check a customer’s email, LinkedIn, and deal history before the call, entirely at their own discretion.
Making the AI write, not just surface.
The second round of feedback was more specific: follow-up emails needed an AI assist built in, the same way ChatGPT can take a rough draft and turn it into something professional on request.
So the panel’s email composer got that same rewrite step, the agent writes loosely, the AI tightens tone and structure before it goes out, rather than leaving follow-up quality to whatever the agent had time to polish.
4. Outcome 🏆
Where it is now: Co-Browser went into development after I left, and it’s live today as part of Salesken’s AI Sales Assistant. The product still ships the pieces I designed: real-time in-call prompts, AI-drafted follow-up emails pulled from the conversation and deal history, and Smart To-Dos. Salesken sells it per seat to sales teams, and the company reports it’s used by 200+ businesses.
This internship was the most challenging, most rewarding stretch of work I’d done, mostly because of how ambiguous and fast-moving it was.
I didn’t stay on to see it through to launch, so I can’t claim specific usage numbers from my time there. But knowing the direction shipped, and is still what agents use today, is the outcome I can stand behind.