Salesforce runs on the same CRM we sell to customers. But even our own sellers had to jump between tools to find the right customer signal, act on it, and update the record. We fixed that in weeks, not quarters. And the people who helped build the fix weren’t the engineers you’d expect. The story of how it came together at agentic speed, using a small squad, matters as much as the tool itself.
The problem: too many tabs, no single place to act
SMB sellers juggled more than 18 tools every day and lost roughly 2.3 hours to context switching. Website visits, product usage spikes, renewal windows, and other buying signals lived across disconnected systems. Sellers had no single place to see what needed attention, act on it, or track what happened next. They spent too much time chasing updates instead of moving deals forward.
The solution: one persistent surface, in the flow of work
The fix is a persistent surface inside Slack that consolidates detection from dozens of sources into one clear, trackable daily work list. Sellers see the right signal at the right time, take the next best action, whether that’s logging a call, sending an email, or reaching out, without ever leaving Slack, and the outcome logs straight back to our CRM. Alerts auto close on the conversation outcome, so there are no stale notifications and zero manual housekeeping. Managers get accountability too: a simple question like “how many of my reps acted on signals today?” returns a dashboard view instantly.
The core principle is what makes it work: users never open Salesforce. They consume, act, update, and close entirely in Slack. Salesforce comes to where people already work.
How we built it: a small squad, not a roadmap
The most repeatable part of this story is the how. This wasn’t a 12-month project plan handed to engineering. It was built by a small, cross functional squad that formed around the problem, shipped in days, and disbanded. The business didn’t file a ticket and wait; they rolled up their sleeves alongside engineering, with no layer in between.
The squad ran on a 24-hour loop: plan, build, swarm, review, and ship. That compressed what used to happen over two-week sprints into daily microcycles, leading to roughly 10x faster decision making. AI coding agents parallelized the grunt work so the humans stayed on the hard calls. Claude Code turned intent into working software in hours, while a three-model check caught errors fast in place of slow manual review gates. Governance and security weren’t a trade off for speed; because access runs through MCP, the data rules travel with it automatically. Claude Code even played QA analyst, determining the right sample size, calculating the confidence interval, and comparing CRM data against signal status, turning what used to be hours of manual CRM inspection into about 20 minutes of intelligent review.
The results, and what’s next
The foundation shipped in weeks. Today 47 signals live across sales, product, and marketing sources, with 2 to 3 new ones landing every week, roughly 400,000 records surfaced, and more than 2,500 accounts launched. The roadmap scales on the same governed foundation: richer context per signal through the Next Best Engagement Engine, dynamic multi step plays instead of one off alerts, expansion to new segments and personas, and a horizon of nearly 400 signals, close to 9x today’s coverage, with every seller, every surface, closed loop by default.
The problem isn’t new. Who gets to fix it is.
That’s the real lesson, and it travels to any company with sellers. Start where people already are, let small squads out ship big roadmaps, treat speed and safety as traveling together rather than as a choice, and just get started, because one success story is the best recruiting pitch for the next squad.
Want to build at agentic speed?
See how Salesforce uses Rapid Innovation Squads (RIS) to redesign processes, move faster, and scale AI with the right governance in place.
