Using AI To Maximize Sales | Automate with Agentforce

The challenge

As a Salesforce partner, I have the privilege of working with a wide range of companies across many industries. Some are small and scrappy, others are global household names. And while their challenges vary, one theme is universal: too much work, and never enough time or people.

In early 2025 I was asked to explore potential use cases for Agentforce at a manufacturing customer. Like any good consultant, I put the technology aside and focused first on the business pain points. Which tasks were repetitive? What was falling through the cracks simply because there wasn’t enough time? And where were limited resources putting the business at risk?

That’s when I uncovered a familiar challenge: “High Volume, Low Value” deals. The sales team couldn’t keep up with the flood of incoming opportunities, and smaller deals were being lost—not because they weren’t worth pursuing, but because there simply wasn’t enough time or people.

Instead of accepting lost revenue as a cost of doing business, we imagined a different approach: a digital assistant that steps in to keep those deals alive. One that handles the repetitive work but keeps the sales team firmly in control. One that frees them up to focus on the high-value deals that truly need their expertise.

Sure, we could’ve gone with classic process automation or a simple mail merge, but that would have felt impersonal. What this customer wanted was the perception of a human touch, even if AI was doing the heavy lifting behind the scenes.

With less than 1 hour of configuration, Agentforce made that a reality.

The solution

Saving revenue with just one hour of configuration might sound ambitious, but that’s the beauty of a simple setup. At the heart of it: finding the right deals and sending thoughtful, relevant emails before they go cold.

What does that mean in practice? The quick, but powerful, agent was built with two key pieces.

• A flow to identify high-volume, low-value deals that need attention
• A prompt template to craft the perfect follow-up email, using meaningful context from the opportunity

That’s it! Two pieces of configuration, wrapped up in a friendly agent. We were in business in less time than it takes to pick your next Netflix watch.

With a simple under-the-hood solution, the user experience is just as easy and elegant. It begins with the salesperson loading up the Agentforce panel and asking for their “gold deals.” After a few seconds of thinking, the flow runs and surfaces the most relevant deals.

When a salesperson clicks into one of the opportunities, Agentforce takes them straight to the record page. From there, they open the email composer. Normally, they might spend 10 minutes reviewing the opportunity and another 10 drafting the perfect message. But with Agentforce, they simply click Draft with Einstein, choose the premade template, and voilà—a thoughtful, context-rich email that reads like a human wrote it.

This small solution packs a big punch, cutting a 20-25 minute task down to just 30 seconds. Most importantly, it’s still human-driven. The user chooses which deal to work on and reviews the email before it goes out. This isn’t a bot churning through the pipeline—it keeps the human touch where it matters.

The flow that powers the first part of this workflow is made up of two key elements. First, it finds the record IDs of relevant opportunities. Then, it passes those IDs to the agent.

While you could instruct Agentforce to search for deals that meet specific criteria, this simple, elegant flow ensures consistency. It reliably finds opportunities that fall into the right category, belong to the user, and sit within a defined date range.

Large language models (LLMs) are inherently non-deterministic, generating unique outcomes every time even with the same input, but finding specific records has to be repeatable.

We have our flow and it works like a charm, but for Agentforce to make any use of it, it has to be plugged into an Agent Action. This is as simple as creating a new Agent Action in the Agentforce Assets Setup menu, selecting our flow, and entering the following details.

Agent Action Instructions are used to help Agentforce identify what the action does and when to invoke it. The use case for this action is specific, so we’re able to make the instruction short and to the point.

The Prompt Instructions in the Output section of the action tell Agentforce what kind of data to expect. In this case, our output is a list of opportunity record IDs. Under normal circumstances, this would output a list of 18 character IDs that make no sense to anyone. The second part of the instruction tells Agentforce to pass this list into another action called Query Records. This out-of-the-box action can search Salesforce and present the list as a series of clickable links. It’s the function that makes our action look human readable.

The last piece to make this work is to create an Agentforce topic and add the action to it. It may feel like a lot of work, but once you get some practice in, everything you’ve seen so far can be achieved in less than 15 minutes. In short, Agent Builder takes care of the heavy lifting, leaving admins free to focus on delivering value, not wrestling with configuration.

While the action to find the opportunities is repeatable and precise, this is in complete contrast to what’s needed for the sales email. The email has to feel natural, with enough variation to avoid sounding robotic or repetitive. That’s where we lean into the strengths of the LLM and dial it up to 10.

By giving the prompt template some simple instructions, we’re able to generate an email that’s professional, clear, and “good enough.” But with a little extra effort, we can go far beyond good enough.

Through trial and error, we can enhance the instructions to fine-tune the tone and inject context into the final message. The following is the exact prompt we used to keep the email consistent in tone while including vital context from the Next Step and Description fields.

You're Account Executive {!$Input:Sender.Name} from {!$Input:Sender.CompanyName}. You need to check in with {!$Input:Opportunity.SDO_Sales_Primary_Contact__r.FirstName} to see if the {!$Input:Opportunity.Name} quote is still valid. Today's date is {!$Flow:Get_Today.Prompt} and the current close date is {!$Input:Opportunity.CloseDate}. If the close date is in the past, ask the customer for an updated timeframe.

Create an email that is addressed to {!$Input:Opportunity.SDO_Sales_Primary_Contact__r.FirstName} and asks for the above information.

Start the message with "Hi {!$Input:Opportunity.SDO_Sales_Primary_Contact__r.FirstName}".
Do not explicitly write today's date.
When referring to close dates, only use month and year.
The value of the deal is {!$Input:Opportunity.Amount}.
Use simple friendly language. Avoid words like "inquire" or "assist".
Use contractions where appropriate, for example "I'm" instead of "I am" or "I'd" instead of "I would"
Offer a follow up call if help is needed.
Do not make any reference to any meetings as none have taken place.
Do not use the phrase "prompt feedback" or indicate that the recipient needs to be quick in their reply.
Use any details found in the Opportunity Next Steps. These details have not already been discussed with the recipient. Opportunity Next Steps: {!$Input:Opportunity.NextStep}.
Use any details found in the Opportunity Description but assume that the sender and recipient have already discussed this prior. Opportunity Description: {!$Input:Opportunity.Description}.

And it bears repeating: All of this took 1 hour to configure, including the many iterations of the above email prompt template instructions.

Key benefits and lessons learned

The most immediate benefit of this implementation is the significant reduction in time required to generate outreach emails. But beyond time savings, the contextual, relevant messaging that Agentforce produces has the potential to directly impact revenue. Even if just one deal is closed for every 20 emails sent, the return is clear. What starts as a helpful, new tool quickly proves its value, transforming from a smart enhancement into a self-funding solution.

Just as important is the fact that this foundation is free of technical debt. That means it’s easy to maintain, simple to evolve, and doesn’t create future headaches for admins or the wider team. Admins and business super users can confidently adapt the logic, all without needing to involve a developer. The result is a lightweight, scalable solution that can grow alongside the business without ever becoming a burden.

Takeaways for admins

When exploring solutions like this, the most important thing isn’t the tool—it’s the problem. It can be tempting to dive straight into configuration, but the real value comes from stepping back and focusing on the business need. The strongest solutions usually come from 90% thinking and just 10% doing.

A great question to ask your stakeholders is, “If you hired an assistant today, what job would you give them?” The answers to that question often reveal the repetitive tasks, the forgotten follow-ups, and the hidden friction that slows the business down.

Once you’ve clearly defined the problem, the solution often reveals itself—and more often than not, it’s simpler than expected.

There’s tremendous power in simple solutions. They’re easier to build, easier to maintain, and easier for the business to adopt. A solution that gets used is a solution that wins.

Resources

• Trailhead: Get Started as a Salesforce Business Analyst
• Trailhead: Build an AI Agent with Agentforce
• Trailhead: Get Started with Prompts and Prompt Builder
• Trailhead: Build Flows with Flow Builder

Want to see more good stuff? Subscribe to our channel!