Movatterモバイル変換

[0]ホーム
URL:

Skip to content
The image shows a chat interaction between an Engine customer and Agentforce where the customer is looking to cancel a booking. On the right states, “30% autonomous case resolution.”
Engine logo

From takeoff to smooth landing: How Engine deploys Agentforce with confidence.

Learn how smart topic design, rigorous replica testing, and close monitoring help them build high-performance autonomous agents.

  • Share the story

Executive summary

Engine is growing fast. Their client services team handles more than half a million requests from travelers per year, often spending valuable time on routine reservation changes that leave less space for the complex, high-touch cases that really require their expertise. At the same time, the sales team grew fivefold in just one year – from 50 to over 250 sellers – adding new demands on HR, IT, operations, and finance as they worked to support both staff and customers. 

To continue offering exceptional service as they grew, Engine turned toAgentforce, the agentic layer of theSalesforce Platform. Their first AI agent, Eva, now manages over 30% of customer cases end-to-end — from rescheduling reservations to recommending accommodations based on preferences — cutting handle times and saving millions annually. 
For employees, Agentforce inSlack provides instant support across multiple specialized AI agents. Engine’s AI agent named Mae will act as a multipurpose admin expert, streamlining IT requests, HR support, and finance questions such as, “What’s my team’s budget this quarter?” Meanwhile, a second Slack AI agent called Cloe assists the client services team by delivering real-time case research and account summaries, helping reps respond quickly and accurately.

Tips for Success

See what Engine learned from deploying Agentforce.

Read the full story

Engine logo
Customer Story

Engine provides personalized service to 1 million travelers with Agentforce.

Autonomous agents handle 30% of customer support cases and help staff find critical info fast.

Consolidate related topics to prevent confusion.

Together with managed services partnerAstound Digital,Engine launched Eva in just 12 days. As they expanded the agent’s abilities, they noticed something surprising during testing. When too many similar topics existed – for instance, separate topics for “book a car,” “change a car,” and “update passengers" – Agentforce struggled to identify the right customer intent.

The fix was surprising yet effective. Instead of adding or changing Eva’s topics, Engine reduced them. They consolidated separate smaller topics into broader categories such as “car management,” and organized multiple related actions beneath each one. This approach gave Agentforce clearer choices, reduced response times, and made ongoing tuning easier for administrators.

Takeaway

More isn’t always better. Grouping related topics into broader categories can improve the speed and accuracy of agents, and make them easier to maintain long term.

Use replica testing with a wide variety of inputs.

Testing wasn’t an afterthought — it was central to Engine’s success. “We have an extremely high bar for Agentforce,” said Sarah Morton, Senior Salesforce Administrator. “Neither customers, nor employees, will interact with a new Agentforce topic before we’ve tested it about 100 times using different tones, with typos, without typos, and logged in versus logged out.”

Testing with a wide variety of inputs helps them see how Agentforce will respond in a variety of situations. The goal is to pressure-test edge cases and catch potential issues so customers don’t encounter them.

Engine also uses a replica testing approach to stay agile and manage risk while rolling out new capabilities. For each AI agent, they create and manage two versions – one live, and one experimental. New actions are built and tested in the replica first so updates don’t disrupt the customer experience. And if issues come up with one of their live agents, Engine always has a backup ready to deploy.

Takeaway

Test your agent with as many inputs as you can imagine to find and fix edge cases early. Consider keeping a backup replica agent to use for realistic testing, and ensure you have an agent ready to swap in if needed.

Treat performance monitoring as part of testing.

Engine tracks Eva’s customer satisfaction (CSAT) scores daily and uses out-of-the-boxData Cloud reports to spot patterns in conversation quality, latency, and request types. They also rely onTableau Next to monitor performance analytics — everything from time to successful resolution or escalation to abandon rates — giving both executives and frontline teams real-time visibility. Beyond the numbers, the team reviews transcripts to understand where customers are getting stuck and uses those real conversations to guide future optimizations.

Having real-time visibility into performance helps them move fast when issues surface. For example, they discovered that when customers entered numbers like confirmation codes without explanation, Eva sometimes mistook them for phone numbers or invoice numbers. To fix this, they configured Agentforce to pull context from earlier conversations — or ask clarifying questions when needed.

In another case, low CSAT scores pointed to reduced satisfaction when Eva couldn’t locate a previous booking contract, typically because it was cancelled or out of date. The fix was to build a flow to querySales Cloud for contract data and give clear answers like, “This booking was canceled.”

Takeaway

If your agents are designed to handle back-and-forth conversations, devote time to reading the transcripts. This will help you uncover hidden pain points or opportunities for improvement after launch.

[8]ページ先頭
©2009-2025 Movatter.jp