₹25 crore in savings over five years for a 300+ person sales team

Scaler is an edtech company (online & offline courses) It trains working professionals and 12th grads for software and business careers. The sales team works inside the CRM all day. If the CRM is slow, sales slow down.

The goal was to replace a third-party CRM. The old tool cost about ₹5 crore a year. Common actions took 7-10 seconds. Sales managers had to do manual work in Sheets to get a basic lead set. We rebuilt the full system from the ground up. After release, we added powerful new features that saved time, effort, and increase sales productivity.

Pagination took 8-15 seconds and sometimes failed, so BDAs waited on loading screens instead of calling. The filters could not select the leads managers needed, so every new lead set meant a 30-35 minute detour through Sheets, about two days of delay before calls started, and private lead data leaving the system.

A BDA's day revolved around making sales calls. The next lead must be on screen when they sit down. If the tool is slow, they make fewer calls, this leads to lower productivity. The old CRM cost about ₹5 crore plus ₹40-60 lakh in extra charges for additional features a year. Even at that price, it could not filter leads the way Scaler needed.

Pagination took 8-15 seconds, and sometimes failed. People waited on loading screens instead of making calls. Thier worklfow and focus broke everytime they move between leads and other pages.

The filters could not select the leads that managers needed. So managers exported leads to Sheets, applied their own filters and formulas. Then they imported the data back. A manager spent 30-35 minutes every time they needed a new lead set. Sharing leads or filters was hard. Managers had to verbally walk teams through filter steps on screen-share calls to apply a filter.

They taught the same steps again each time the plan changed. Lead allocation could delay calls by about two days. Every export also moved private lead data out of the system, with no way to track it.

Familiar screens, a new system underneath. 300 people moved across in a single day and sales did not drop.

We changed the product, not people's habits. Users saw familiar screens. Under the screens, everything was new. We stopped the old vendor at the end of Q4 2024. On the first day of Q1 2025, 300 people logged into the new tool. Sales did not fall.

Filter buckets combine inside the CRM, so managers get the exact lead set without an export.

Managers could now combine many filter groups inside the CRM. They got the exact lead set without export or import. No more manual tags. No more copies of lead data outside the system.

One shared link applies a manager’s filter to each person’s own leads. The 40-minute walkthrough became a link.

Before, a manager guided 20-25 people through many filter steps. This took 40 minutes at a minimum. Now the manager shares one link. The link applies the same filter to each person's own leads. The walkthrough became a link.

Automations sit on top of filters: message a lead after a call, or assign leads by condition.

Admins and managers could build automations on top of filters. Examples: send email and WhatsApp messages after a call, or assign leads based on set conditions.

With filters doing the real work, unnecessary exports were blocked and lead data stayed in the system.

The new filters did the real work. So we blocked exports that were not needed. This saved time. It also kept lead data inside the system.

Dark mode was a small change at the token level. It became one of the most loved changes at launch.

The safety metric was 4 sales per BDA per month. That number did not fall after the switch. We did not break the sales team.

Five AI agents run on every lead today. Every call was already recorded, but no one used the recordings. The agents turn those calls into information the system can act on.

The migration ran team by team. This project taught me how queues, cache, and background jobs keep a large system fast. Working with Redis, Kafka, and SQS showed me how large-scale systems stay responsive. That knowledge now feeds my own building work.

In v1 we did not rebuild the lead details page. BDAs used it only as a contact card. They did not use it to understand the lead. In v2 we rebuilt that page. Now a BDA sees the lead stage and the key facts before the call starts. Part of this is design. Part of this is the AI features above.

• Stakeholders wanted core flow changes in the first release. I pushed back. I kept the existing BDA workflows. Familiar screens were my adoption strategy.

One problem could break the whole AI engine: call recordings never reached the server. BDAs killed the upload with various tricks. No recordings meant no AI.

I built an end-to-end system that mimicked a large-scale calling setup for our use case. I rooted an Android phone. When a BDA clicks call, an automated call goes out from that phone. The numbers stay masked, so BDAs do not use their personal phones to contact leads.

More on this in the full story: The Calling POC case study.