|E-commerce / MarketplaceZero seasonal hiring for the first time ever. Black Friday included.
How Groupon, handling 300K+ conversations a month across 13 countries, went from zero automation to eliminating seasonal hiring, replacing four platforms with one, and transforming agent productivity. In six weeks. Before Q4 hit.
The Situation
Many tools. Many contracts. Minimal automation.
By mid-2025, Groupon's customer service leadership was managing 273 agents across two locations, serving 16 million active customers in 13 countries. The stack had grown organically over the years: Standalone app for customer-facing FAQ and self-service, Salesforce Service Cloud for agent work, a legacy translation service, and Salted AI QA for quality assurance. Many tools. Many contracts. Almost no integration between them.
Automation was nearly nonexistent. The only "deflection" was a FAQ tree with limited self-service options. Every customer who got past those basic branches was transferred to a human agent.
After every conversation, human agents spent up to a minute manually tagging outcomes, selecting reasons, and writing wrap-up notes. A minute doesn't sound like much until you multiply it by 300,000+ conversations a month. That's thousands of agent-hours each month spent on data entry that added no value to the customer.
And because Groupon operates across 13 countries, agents and customers often didn't share a language. A legacy translation service added up to 10 seconds of latency to every single chat message, creating yet another tool in the chain, another contract, another integration to maintain.
Leadership saw the opportunity to fundamentally transform how they delivered support. The current stack couldn't get them there.
The Stakes
Q4 was coming. There was no time for a 12-month implementation.
The weeks around Black Friday represent Groupon's highest-volume period. Every year, the playbook was the same: hire 50 to 100 seasonal agents weeks in advance, train them on the existing tools, absorb the productivity dip, then let them go in January.
Leadership had seen transformation projects stall before. The question wasn't whether to modernize. It was whether anyone could actually deliver before Black Friday.
"What impressed me most about Salted wasn't just the technology, it was the partnership. They didn't show up as another vendor with a product to install. They embedded with our team, understood our operation inside out, and helped us rethink how we deliver customer support at scale. And the timing was critical. They completed the deployment right before our Q4 high season hit and conversation volumes surged dramatically. We went into our busiest period of the year with a fundamentally different operation, and the numbers speak for themselves."
Dušan Šenkypl
CEO, Groupon
The Approach
Not advising. Operating.
Groupon's CEO tasked Salted with a CX transformation pilot: take automation from zero to global coverage, and put agents on a new platform. Not as a consultant. As an operator, accountable for the numbers.
This wasn't a software deployment. It was a complete customer experience transformation: new platform, new agent workflows, new automation logic, new analytics, and built-in translation, all delivered end-to-end. Groupon had been running Salted AI QA for over a year at that point, giving Salted unprecedented visibility into every conversation, every failure mode, every friction point. That foundation made Salted uniquely positioned to go beyond quality assurance and take on the full transformation, including automation logic, agent platform, and operations.
A year of data before day one
Groupon had been running Salted AI QA for over a year before the transformation began. Every conversation, every failure mode, every friction point had already been mapped. That foundation meant Salted didn't start with assumptions. They started with evidence. It's why six weeks was possible.
Engineering-native automation
Groupon's engineers build automation the same way they build everything else: same programming languages, same CI/CD pipeline, same testing frameworks, same deployment process, same AI stack. No proprietary workflow builder. No black box. No new stack to learn.
Built-in translation, 13 countries
Automatic, seamless translation from any language to English and back. Agents only communicate in English. A customer in Germany writes in German, the agent sees English, responds in English, and the customer receives German instantly. This replaced Groupon's legacy translation provider entirely.
The Rollout
Three agents to thirteen countries in six weeks
Pilot launch
Launched with 3 agents to verify the ergonomics of the agent interface, the integration with Groupon's automation logic, and overall feedback from agents.
Edge cases surface
Agents hit friction in workflows that looked fine on the whiteboard. Years of muscle memory in old tools, including Salesforce shortcuts, manual tree navigation, and wrap-up routines they could do in their sleep. Every habit had to be unlearned.
US, then English, then international
United States first to 100%, then the rest of the English-speaking countries, then international markets, all with real-time AI QA visibility throughout the migration.
Full coverage
100% of chat traffic on Salted across all 13 countries. Black Friday handled without seasonal hiring.
"Our engineers build customer service automation and AI agents the same way they build the rest of Groupon: same languages, same CI/CD, same testing, same deploy process. No new stack. Most platforms don't let you work this way. They force you into black boxes, proprietary instruction formats, or low-code tools that collapse under complexity. Salted gave us full control and ownership of the automation logic, and the speed to adopt AI advancements that are frankly happening on a weekly basis."
Aleš Drábek
CTO, Groupon
The Change
The transformation in operational terms
After the initial transition, Groupon incrementally adopted Salted's productivity features: the bot finishing conversations after the agent leaves, seamless multi-modal chats combining text and interactive menus, and custom buttons that let agents trigger complex business processes with a single click. These compounding improvements meant agents on Salted were handling materially more conversations per shift than they ever did on Salesforce Service Cloud, without working harder.
AI-native from the ground up
Bot and agent work on the same conversation, on the same platform. AI stays connected during human takeover, helping formulate responses and execute actions. No one-way handoffs. The AI and the agent collaborate on the same thread, with the same context. Once the bot knows how to handle the wrap-up, the agent can leave and start serving a new customer.
100% QA coverage on one platform
Every single conversation is now scored automatically on one platform that understands the full customer journey, including which part was handled by the bot, which part involved an agent, and how they worked together. Same rules. Same lens. No silos. No blind spots. The highest level of operational transparency Groupon's service leadership had ever had.
Shipping every week
Salted ships new platform capabilities every week, and because Groupon owns the automation logic, their engineering team can iterate just as fast. New productivity features, refined bot behaviors, updated business rules — both sides push improvements on a weekly cadence. This isn't a platform you configure once and hope for the best. It's a system that gets smarter and faster with every deployment, from both directions.
A new way of working
Agents love working in the Salted platform because it gives them what actually matters: a tool that is intuitive, fast, and automates the work they shouldn't be doing in the first place. Onboarding takes less than an hour. Agents started picking up extra chats on their own, not because they were told to, but because the platform made it effortless. No more selecting reasons, outcomes, or describing what the conversation is about. AI handles all of that automatically.
"With Salted, trainers are not required. A 30 minute training session followed by 30 minutes of shadowing and new agents can start handling customer conversations. One hour from onboarding to live, that's it."
Nithya Subramani
Area Manager, Customer Service, Groupon
The Impact
Operational impact that maps directly to the P&L
Transition
Scale
Savings
Systems Replaced
Where the transformation comes from
The operational transformation is driven by two compounding effects. First, most conversations now resolve end-to-end without agent involvement, handled entirely by AI from first message to close. That directly reduces how many agents the operation needs. Second, when agents do step in, they move through conversations faster than they ever did on Salesforce because the platform removes the manual work that used to slow every interaction. Wrap-up alone used to cost up to a minute per conversation. Across 300,000+ monthly conversations, that's thousands of hours returned to productive work. Add the consolidation of four vendor contracts into one platform, and the compounding effect is substantial.
The Bigger Picture
See yourself in this story?
As the CTO
Your engineering team can write automation logic using the programming languages, frameworks, and agentic AI tools they already use. Deploy to staging and production through your existing CI/CD pipeline. Write tests the way you work today. Do pushes, rollbacks, and A/B testing the way you already do. No proprietary runtime. No vendor-controlled deployment. Full ownership.
As the Head of CX Ops
Six weeks from pilot to full coverage. No 12-month implementation. No seasonal hiring panic. Every conversation across 13 countries, in any language, is automatically scored, categorized, and analyzed on one platform. Whether handled by AI or by a human. 100% QA coverage from day one.
As the CFO
Eliminate seasonal hiring cycles of 50 to 100 agents. When most conversations resolve without an agent, you need fewer agents. It's that direct. When the agents you keep move through conversations faster on a better platform, you need even fewer. Replace four vendor contracts with one. These aren't separate savings. They compound.
