What It Takes to Make AI Customer Service Actually Work

Agents can do a great many things. They can inspect a photo of a moldy shower or help dispatch technicians to the owner of said shower to get that grout looking good again. They can help football fans — millions of them — figure out ticketing and memberships and stadium events. They can help people navigate tricky financial realities.

For companies willing to rethink the role of AI agents in their technology stack and the operations underneath it, they can be a huge customer service advantage. But how to get started? 

When the team at Salesforce was evaluating submissions for this year’s Customer Success Awards — a program that honors customers who’ve used Agentforce to drive real business results — they saw a consistent pattern. Different organizations had different challenges, but the most successful transformations consistently did three things: they started with the right problem, built the right foundation, and brought the right people together. Here’s what we learned from their success.

Start with the right problem

True transformation starts when teams identify a specific operational friction point.

Take Tottenham Hotspur Football Club, based in London. As one of the most storied institutions in all of sports, Spurs have fans all over the world. Supporting this gigantic fanbase means fielding an immense volume of inquiries, particularly around match-day ticketing, stadium access, and account management. Historically, resolving fan questions about ticketing or membership required human service agents to log into three separate systems, consuming four to five minutes per interaction.

Rather than chasing generalized AI capabilities, Spurs leadership focused a recent initiative squarely on alleviating this friction point. First, they leveraged Salesforce Data 360 to consolidate disparate data sources into a single, real-time fan context layer, unifying more than 4.6 million fans and 30 disparate systems into a single “golden record.” Next, they built the AI-powered Ask Spurs service agent, equipped it to handle queries in eight different languages, and grounded it directly in that unified profile data. Finally, they applied the unified data context to both self-service (unassisted AI chat) and agent-assisted channels.

Today, according to team CTO Rob Pickering, when fans reach out with inquiries, the Ask Spurs agent handles the majority of them instantaneously. What’s more, if calls must be handed off to human agents, the unified profile empowers those humans to resolve issues in an average of 10 seconds or less.

These improvements might not spark as much pride as winning on the pitch, but they’ve certainly made a difference to the bottom line, saving 80,000 minutes in the first month and on track to handle over 200,000 calls a year. By starting with a defined business challenge rather than an abstract goal, Tottenham Hotspur created measurable value for all its customers.

In other words: Focus on real-world impact. The most effective solutions prioritize clear ROI and time savings. For Tottenham Hotspur, that meant that targeting a single, repetitive interaction can save thousands of hours of human labor.

Build the right foundation

Another way to create lasting value is anchoring an AI project on a solid operational foundation. Even the most sophisticated agents will stumble if they’re built on fragmented data or outdated practices. For most businesses, success necessitates clean data, scalable infrastructure, and adaptable processes.

A family-owned business, The Grout Guy is Australia’s largest tile and grout restoration network, and in the early days of the company, management tracked scheduling and client books using physical paper planners and basic spreadsheets. As business demand scaled rapidly across the country, managing dispatch, lead capture, and job quotes through manual methods created inevitable bottlenecks.

Recognizing that rapid growth required an equally scalable backbone, The Grout Guy leadership undertook a total digital overhaul. They centralized workflows and replaced legacy logs with an integrated Salesforce suite including Data 360, Agentforce Marketing, Agentforce Field Service, Agentforce Service, Agentforce Builder, and Slack. They also automated lead capture and speed-to-quote workflows, allowing requests to transition seamlessly into quotes and field dispatch. As part of the Agentforce implementation, The Grout Guy built a multiagent system, including one that automatically processes inbound work orders from emails and PDFs. 

Separately, its customer-facing agent, Groutie, asks website visitors to upload photos of their bathrooms via chat, then uses optical image recognition to count tiles, identify mold or discoloration, and help generate a quote. 

These investments eliminated human blockers and evening voicemail backlogs, drastically shortening the time from customer inquiry to quote delivery. According to CTO Anthony Messina, with the help of Groutie, quotes are now generated within 20 minutes, a vast improvement over the three to five days it took when the system was manual. The Grout Guy has increased its field capacity from four technicians per administrator to more than 25, without significant growth in its dispatch team. Perhaps most impressively, it now takes less than two minutes for the company to process work orders, 24/7.

In other words: Scale your infrastructure before your ambitions. For The Grout Guy, transitioning away from legacy tools onto a unified platform was a prerequisite for seamless automation.

Bring the right people together

Technology alone doesn’t drive change; people do. Achieving sustainable AI adoption requires uniting cross-functional teams, maintaining strong feedback loops, and ensuring that employees view automated tools as helpful collaborators rather than job replacements.

Sammons Financial Group illustrated this principle when it explored how agentic AI could create new ways to support customers and distribution partners while expanding the capacity of its service organization, headquartered in West Des Moines, Iowa. To establish rapport with customers, technologists worked with Agentforce to develop the agent with a friendly, empathetic voice and intuitive conversational guardrails. As they extended agent coverage to nights, weekends, and Friday afternoons, they noticed that customers were responding positively.

“Our customers were having meaningful, safe conversations with the agent,” said Andrew Walling, the company’s AVP, Capability Planning & Delivery. “To me it conveyed a real trust.”

This was only part of the puzzle. From the outset, Sammons leaders recognized that internal buy-in was just as critical as technical performance. To ease employee anxiety, managers created an internal team SharePoint site that framed the AI agent’s development as an evolving persona, visually depicting the agent’s journey from a learning “apprentice” to a full-fledged service colleague.

By combining empathetic design with strong human oversight, Sammons’ new AI-based system has handled more than 16,000 policy calls in the first six months of deployment, automatically providing nonjudgmental customer service for most inquiries while freeing up human advisors to tackle the more complex and emotionally sensitive requests. This has enabled the company to absorb seasonal peak call spikes without hiring temporary staffing or increasing hold times, which translates into another win for customers.

In other words: Pair human empathy with machine efficiency. AI can unlock and remove constraints on a company’s capacity to improve self-service and free the human workforce to deliver better experiences. 

Ultimately, by focusing on the right problem, laying a solid foundation, and bringing people along for the journey, modern customer service organizations can turn ambitious visions (and smaller ones too) into daily operational truths. 

Learn more:

• How Salesforce pilots its own software
• New study of 2,025 agentic AI leaders: First to launch isn’t fastest to ROI