At Deloitte Digital, we’ve supported over 30 Proofs of Concept for Agentforce—the conversational agent developed on the Salesforce platform. These projects, carried out across various industries, have helped us identify not only the true benefits but also the limitations and requirements of this new generation of intelligent assistants. 

Here’s a look at the key lessons learned.  

What Agentforce Does (and What It Doesn’t)  

Agentforce is not meant to replace human advisors. 
It’s not a black box to which customer relationships are blindly delegated. Instead, Agentforce integrates with existing tools to streamline operations, automate simple tasks, and assist agents in their interactions with customers. 

Here’s what it actually enables: 

 Automation of Simple or Complex Tasks  

Agentforce takes care of repetitive, time-consuming actions such as filling out forms, searching for information, and following up on standard requests. It frees agents from these tasks, allowing them to focus on human interaction and resolving cases that truly require their expertise. 

Real-Time Contextual Suggestions   

By analyzing the ongoing conversation, CRM data, and customer history, Agentforce suggests relevant responses or actions tailored to the situation. The human agent stays in control but saves valuable time thanks to this intelligent support. 

Structuring and Summarizing Interactions  

The tool is capable of automatically generating summaries, classifying requests, making recommendations, or guiding the next steps in processing—all based on predefined rules.

 Ability to Execute End-to-End Actions   

In well-defined scenarios, Agentforce can act autonomously—canceling a reservation, resending a document, or generating a simple request. Of course, all actions are governed by predefined security rules and escalation protocols.  

What We Observe in the Projects   

We tested Agentforce in real-world environments, with objectives that varied across industries. These field insights allowed us to better identify the most valuable use cases and the key conditions for success. 

Two Strong Trends Emerge:   

•  70% of the projects are focused on service teams, with a clear objective: automate simple cases, improve process reliability, and reduce the workload for agents.   
• 20% of use cases are client-facing, implemented in well-defined contexts where the user experience can be enhanced without introducing significant risk.  
  

Here are some representative use cases : 

• Technical Services Industry 
  The agent handles incident management: it identifies the user, automatically detects the location via a QR code, qualifies the issue, assesses human and financial risks, and directs the case to the appropriate teams.
  The result: fewer errors, faster processing times, and improved traceability.  
• Entertainment and Culture 
  The agent allows customers to cancel their reservations or receive their electronic tickets without any human intervention. It handles these requests in under 30 seconds, compared to 5 minutes previously. This represents a 30% reduction in processing costs for this type of interaction.  
• Automotive industry
  Agentforce automates appointment scheduling for customers within the distribution network, manages availability and confirmations, and integrates the request directly into business systems. This enhances the experience for both the customer and the agent.  
• Banking and credit 
  The agent analyzes the customer's request, assesses their eligibility, recommends suitable products, and can initiate the subscription process. It acts as an expert assistant, available at any time.  
• Nonprofit organizations
  It independently handles tasks such as address changes, membership renewals, or information requests, without adding workload to internal teams.

What you need to know before getting started  

Agentforce isn’t a one-click product. 
Its success depends on a structured approach and a few critical learnings. 

Test and Continuously Refine  

The traditional “build, test, deploy” approach isn’t enough here. You need to test, fine-tune, and test again—in a continuous loop. The goal isn’t 70% accuracy, but 90% or more. Reaching that standard takes time, expertise, and a rapid feedback loop. 

Clearly Define the Scope of Action   

The agent must know exactly what it can and cannot do. Clear scoping is essential: poorly defined or overly broad actions create uncertainty or lead to production errors. It’s better to start small, with a well-defined use case. 

Integrate with Existing Processes   

Agentforce isn’t a standalone solution. It must be embedded in business processes, follow internal rules, use the right data, and fit into the broader system. Integration with CRM, knowledge bases, and internal tools is critical. 

Support Adoption 

The agent may be technically ready—but remain unused if teams don’t trust it or don’t know how to use it. Training, change management, and clear communication are essential parts of the project.

Address Regulatory Requirements   

Since the EU AI Act came into effect, every project must be assessed based on its risk level, transparency obligations, and the safeguards embedded into the agent. This is now a non-negotiable part of any deployment. 

Build a Clear Business Case   

A business case isn’t just a gut feeling or a shiny tech promise. It must be backed by solid assumptions: 

• Which use cases will drive actual productivity gains? 
• What is the volume of data involved? 
• A successful Agentforce deployment is grounded in a realistic evaluation of ROI—and a plan to track it over time. 
  

In Conclusion: A True Operational Transformation Lever

Agentforce is not a vague promise. It’s a solid, field-tested solution that tangibly improves both agent efficiency and customer experience. 

But it requires proper implementation, a robust integration strategy, and a culture of continuous iteration. 

Across 32 projects, we’ve seen this clearly: when well-scoped, well-integrated, and well-supported, Agentforce offloads internal teams, improves reliability, and handles simple requests instantly—while leaving complex interactions to humans, where they matter most.

It’s an operational lever. A driver of efficiency. 
And a way to put humans back where they create the most value: in empathy, attention, and smart resolution. 

At the Salesforce World Tour, we dedicated our breakout session to this topic. This article extends that conversation—with conviction, and the weight of over 30 successful field implementations. 

And this is just the beginning. 

Have a project in mind or questions about integrating AI to enhance your service operations? Let’s talk.