Agents that deliver real results are built with Agentforce.

Latin America's leading retailer scaled customer service and sales support with Agentforce. Now they deliver faster responses and higher satisfaction all powered by agents that work around the clock.

Retailer Grupo Falabella needed to meet a growing number of complex support needs. Agentforce provides fast, 24/7 support via WhatsApp while Salesforce apps help personalise service and marketing at scale.

Before Agentforce, it was very common to see customers start out on WhatsApp and then call our contact centre because they couldn't find the right answer. Now that's changing. People are staying on WhatsApp because they see that Agentforce interactions actually work."
Mariana Albarracn, Customer Service and Quality Manager, Grupo Falabella

Read Grupo Falabellas story

Autonomous agents handle 50% of customer support cases and help staff find critical info fast. Plus, Agentforce handles customer cancellations in seconds on their website and provides always-on HR, IT, operations, and finance expertise to employees right in Slack.

Salesforce is helping enable scalability

Read Engine's story and Implementation Tips

Finnair wanted to better support travellers at every stage from planning to arrival. With Agentforce, they'll provide instant answers for customers at scale. Their digital workforce will resolve 80% of customer service questions.

"Agentforce is going to be the heart of customer service; I see it as revolutionary."
Taina Kunelius, Finnair, Head of Before and After Journey CX

Read Finnair's story

AI agents help SMB advertisers navigate the platform, launch campaigns fast, and stay engaged.

With the Einstein chatbot, we had to tell it what to do and how to assist. Agentforce understands what an advertiser needs no matter how they say it."
Zee Hence, Salesforce Senior Product Manager, Reddit

Read Reddit's story

Together with managed services partner Astound Digital, Engine launched Eva in just 12 days. As they expanded the agents abilities, they noticed something surprising during testing. When too many similar subagents existed for instance, separate subagents 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 Evas subagents, Engine reduced them. They consolidated separate smaller topics into broader categories such as car management, and organised multiple related actions beneath each one. This approach gave Agentforce clearer choices, reduced response times, and made ongoing tuning easier for administrators.

Takeaway
More isnt always better. Grouping related subagents into broader categories can improve the speed and accuracy of agents, and make them easier to maintain long term.

See more tips from Engine:
Engine implementation story

Before the release of Agentforce, Nexo recognized where Salesforce (and technology more broadly) was headed and took early steps to clean up their data. As Nexo Salesforce Architect Kostadin Stoev said, Its really important for everyone to know that you wont get good results unless you have clean data.

But clean data was only part of the equation. Once Agentforce went live, Nexo discovered that the agent also needed clearer context to understand and respond accurately. It struggled to interpret help articles written for crypto-savvy users. This content made sense to customers and support agents, but not to a language model starting from scratch. To solve this, Nexo worked with their client care team to rewrite the articles with both human readers and the agent in mind, adding foundational crypto context to guide Agentforces behavior and improve its accuracy.

Takeaway
Clean data is essential, but not enough. Your content also needs to be written with humans and AI in mind so Agentforce has the context to understand and respond accurately.

See more tips from Nexo:
Nexo implementation story

Safari365 started with internal agents so they could monitor everything closely. It was an intentional, iterative process: They would test a conversation with Agentforce Testing Centre, spot an issue, then tweak the instructions or adjust the subagents. Early on, for example, the agent began pulling internal notes like Customer isnt ready to book into messages meant for travellers. Because Safari365 built their internal agent with a human-in-the-loop, nothing went out, but it served as a strong reminder to pay close attention and ensure proper guardrails were in place.

Because Agentforce is configured in (and understands) natural language, Safari365 CEO Marcus Brain could make changes himself without waiting on developer time. In fact, Brain built two of their agents personally. Being able to build agents and refine them independently made Agentforce the fastest time-to-value of any Salesforce product they have deployed: just six weeks from signing to live.

To this day, Safari365 keeps a tight feedback loop. If someone catches a hallucination or issue, the team can jump in and fix it right away.

Takeaway
Safari365 attributes its success to the simplicity of the tools and a mindset of radical ownership. Brain can sit down and create 10 iterations of a conversation in the time it would've taken to schedule one dev meeting. That level of control keeps them close to both the customer experience and the risk.

See more tips from Safari365:
Safari365 interview

While building OpenTable's AI agents with Agentforce, RoseTree and OpenTable tackled one of the trickiest challenges in AI service design: knowing when to escalate a chat to a human. They created the deflection score, a real-time metric that tracks sentiment and intent throughout the conversation.

Every chat session starts with a baseline score. From there, the Atlas Reasoning Engine adjusts the score dynamically based on the diner's inputs. For example, asking for help carries a score of 5, and requesting a rep is a score of 10. Typing in all caps and extreme frustration take the score all the way up to 20. Agentforce uses this live score to determine whether to continue helping, create a case, or escalate to a human.

The best part is that the deflection score is a living metric. OpenTable's team can adjust escalation thresholds at any time. If staffing is tight, they can increase the threshold to reduce escalations, and during high-stress times like holidays, they can ease the threshold to offer more human support. A companion tenor score helps track how sentiment changes throughout the conversation, offering a full picture of what's working and where to improve.

Takeaway
Build flexibility into your escalation logic to match your staffing levels, customer sentiments, and seasonality. The bonus is that this will also help your AI agent improve with every chat.

See more tips from OpenTable:
OpenTable implementation story

Endress+Hauser treated Agentforce like a new teammate, not just a tool you could turn on and expect to work perfectly. From the beginning, they focussed on teaching the agent how to answer questions in the right way, what tone to use, and which parts of long help articles were most important. They gave real examples of customer questions so the agent could learn how to respond like a helpful team member. The team carefully checked the answers it gave, looking for anything that was confusing, too general, or didn't match what customers needed. If the agent gave a weak or wrong answer, the team fixed the instructions and tested it again.

They kept track of every change using a shared checklist, and nothing went live until it was fully reviewed. But during this training, the team also discovered something else: Many of the knowledge articles Agentforce used were outdated or too hard to understand. That made it harder for both people and the agent to find the right answers. So the team took time to clean up those articles, rewrite parts, and organise the information better. This restructuring of their existing content ensured Agentforce had clean, accurate information to draw from.

Takeaway
Just like any new hire, Agentforce improved with hands-on training, regular feedback, and better resources. The more it learnt, the more consistent it became at delivering clearer, more accurate answers with every iteration.

See more tips from Endress+Hauser:
Endress+Hauser customer story

reMarkable's deep partnership with Salesforce made it possible for them to launch Mark on their Experience Cloud site in just three weeks. Collaboration on everything from vision-setting to implementation and enthusiastic executive sponsorship helped the team meet their ambitious launch goal despite an intense workload and tight deadlines.

That early alignment paid off when the work got hard. It helped them overcome initial growing pains such as long hours, unexpected agent behaviour, and manual workarounds, and provided the necessary support to build a strong foundation for future growth.

Having Salesforce Professional Services, who knew more than us, was great because then we could move faster," said reMarkable's System Manager Pernille Bucher-Johannessen. Working with the people who are closest to the product was super valuable.

Takeaway
Close collaboration with your implementation partner and strong executive support will speed up your launch and help you take your agents from good to great.

See more tips from reMarkable:
reMarkable implementation story