From Pilot to Playbook: What We Learned from Our First Year Using Agentforce

It started out simply enough. A customer who'd signed up for a Slack webinar was contacted by an AI agent for a sales follow-up. The customer expressed dissatisfaction with the webinar experience in response to the opening line "hope you enjoyed the webinar you recently attended," explaining that their expectations weren't met, topics weren't covered, and more.

The agent addressed the complaint, apologized, and then offered relevant information to address the prospect's concerns - not a generic response. Then it was able to seamlessly pivot the interaction back to Sales and have a productive conversation that closed a deal. This flip from a complaint to a genuine opportunity would have been impossible just one year earlier.

This is just one successful example of an interaction handled by Agentforce, our platform for building and deploying AI agents. As "Customer Zero" for Agentforce, we scaled the technology across the company and have seen the impact across every part of the organization.

After one year, our service agent has handled 1.5+ million support requests on our help site, the majority of cases without humans. Our sales development rep (SDR) agent has worked on over 43,000 leads and generated $1.7 million in new pipeline from dormant ones. Agentforce in Slack gave our teams 500,000 hours back this year by handling routine tasks.

Behind these metrics lies a story of challenges, experiments, missteps, and continuous refinement. Early iterations were humbling: When we first launched Agentforce for Sales, ‌for example, one of the agents we deployed was the SDR, which can prospect, conduct outreach, and qualify leads autonomously. We envisioned an army of these agents acting as entry-level sales coordinators, working around the clock at a scale no human could ever approach. But in those early months, the agent responded "I don't know" 30% of the times we asked it for details on a lead. Through painstaking data cleanup and iterative training, we've reduced that number to under 10%.

By deploying Agentforce, we learned critical lessons in areas like agent testing, data, and finding the right human-AI balance, making a million mistakes so our customers don't have to. These lessons are now directly informing the Agentforce product roadmap and creating a blueprint for customers embarking on their own transformation to an agentic enterprise.

The counterintuitive testing and tuning agents need

Customer adoption hinges on efficacy - if agents don't work well, people won't use them. This reality drove our most important lesson: Agent success hinges less on comprehensive training and more on avoiding the nuanced mistakes that only emerge at scale.

One unexpected challenge came when we first launched our customer support agent. While the agent's answers were factually accurate, the experience felt too transactional to customers. When human agents handle incidents, they express concern, saying things like "I'm really sorry to see that. That must be disappointing." Our initial AI agent simply opened tickets.

We knew we had to do better. So we reviewed every customer conversation and consulted with colleagues to identify where there were gaps. Then we built tools to help scale these more-human insights.

We knew we had to do better. So we reviewed every customer conversation and consulted with colleagues to identify where the human touch was lacking.

Our most enlightening mistake involved putting in overly restrictive guardrails. Initially, we instructed Agentforce not to discuss competitors, creating an extensive block list of rival companies. This backfired when customers asked legitimate questions about integrating Microsoft Teams with Salesforce. The agent refused to help because Microsoft appeared on our competitor list.

The solution was elegantly simple. We replaced rigid rules with an instruction to act in Salesforce's - and by extension, customers' - best interest in everything the agent did. This shift revealed a crucial insight: We had been treating our AI agent like an old-school chatbot with overly prescriptive directions, when what we really needed to do was give the agent a goal and let it determine how to deliver on it. Agents perform best when you tell them what to achieve, not how to achieve it. We call this "letting the LLM be an LLM."

Now our agents improve through continuous, relentless iteration based on real user feedback, measuring business value, performance, satisfaction, and engagement. This refined approach enables Agentforce to let customers customize their agent's brand tenor, ‌acknowledging that while some brands may prefer a bit of snark, others require a supportive, reassuring tone.

Data fidelity is tantamount

In the past, machine learning demanded exceptionally precise and manicured data. With agentic AI, the critical factor is having a single set of consistent facts: AI agents are only as good as the data they use.

We discovered that if an agent encounters two conflicting "right" answers within its vast dataset, it will attempt to reconcile them, potentially making up an answer. For instance, in our rollout of Agentforce on Salesforce's Customer Support site, an agent once pulled an outdated set of facts from an old page that wasn't linked to and rarely updated, which contradicted regularly maintained help articles.

This realization forced us to take a multilayered approach.

This realization forced us to take a multilayered approach. We had to focus heavily on data governance (the rules and processes for managing data), source cleanup (getting rid of duplicate or outdated information), and consolidation (bringing all the scattered data from different systems into one central location) while investing in our knowledge article curation and validation processes to ensure agents are using a consistent library of resources.

We also applied Salesforce Data Cloud as our data activation layer, connecting, harmonizing, and unifying data from more than 650 of our own data streams while resolving fragmented profiles to create a single source of truth. This gives agents a complete history of a customer's interactions with the company, helping our agents offer more personalized support.

The lesson: Agents are probabilistic, not deterministic. They don't always produce the same result for the same input, and their outcomes may even drift over time with new information. Continuous iteration and tooling to ensure data consistency isn't optional‌ - it's the foundation that makes agents more powerful than chatbots.

Redefining roles: The human-AI partnership

Agents have become the perfect complement to human strengths, as they handle the grunt work humans find tedious or demoralizing, like answering repetitive requests, nurturing dormant leads with one-in-a-thousand conversion rates, and persisting through rejection without fatigue or frustration.

This creates a powerful dynamic and a strategic recalibration of jobs. SDR agents are revolutionizing sales by handling the volume-intensive prospecting work to generate millions in pipeline from what was previously "sawdust on the floor." Meanwhile, human salespeople focus on what they do best: building relationships, understanding complex customer needs, and closing significant deals.

In customer service, agents are responding quickly to after-hours requests and answering the repetitive questions that used to dominate support teams' time. Now customer service teams can focus on the stickiest, most complex cases and nurture relationships with their customers. As a result, we're able to redeploy our people into areas of the business that are growing, like professional services, forward-deployed engineering, and other roles within customer success and sales.

This partnership model works because it leverages natural advantages: Agents don't mind doing the same thing 7,000 times, while humans excel at the nuanced conversations that turn prospects into customers. Together, everyone can do more.

Video: Agentforce boosts sales productivity by 49% for Salesforce ANZ teams

The journey to consumer-grade experience

Another critical lesson learned this year is that fit and finish matter. Whether for internal employee use or direct customer interaction, users compare agent experiences with leading consumer-grade AI like ChatGPT, so we have to strive for superior experiences.

When we first started using Agentforce, we were building agents for very specific needs - a wellbeing agent for benefits questions, a meeting agent for scheduling support, a career agent for questions about professional development. While these agents were helping employees offload repetitive tasks and focus on higher-level work, the employee experience was fragmented across dozens of agents.

This summer, we introduced the employee agent to bring together these agent experiences into one so employees no longer have to guess which tool to use or where to go. We also created a manager agent that brings together everything from employee survey results to quarterly review resources to team feedback. Employees now have one go-to agent to get all their workplace questions answered.

But it's not enough to create the agent. We know that for agents to be truly effective, they must be built right into the flow of work. Our agents aren't siloed in a separate application; they're seamlessly integrated into Slack, CRM, web, and email. These are the tools our employees and customers use every day. This approach has led to 86% of our employees using agents in Slack alone and 99% of our global workforce using our internal agents.

But it's not enough to create the agent. We know that for agents to be truly effective, they must be built right into the flow of work.

The work is never done. As data changes and behaviors evolve, agents require constant experimentation and refinement. We're testing and improving daily.

Closing out the year and looking forward

As Customer Zero, we do the work, find the bugs, and make the mistakes before our customers ever have to. We see it as our responsibility to experiment and iterate ruthlessly and blaze the trail to agentic transformation. We're building our experience into Agentforce to deliver measurable ROI and make deployment as easy as possible to ensure we build business value, not just cool technology.

One year in, we've seen remarkable impact and adoption - and we're just getting started. We're committed to guiding customers through this fundamental shift to become an agentic enterprise with confidence and proven results.

More information: