There’s understandably a lot of excitement around Salesforce’s Agentforce, a no-code builder designed to make it far easier to apply AI throughout your business.
As with any technology, however, it’s not about its potential - it’s about turning concept into actionable reality.
I’ve been digging into Agentforce (including building agents) and related documentation. Here’s what I’m advising our clients with AI and Agentforce, and I’ll use these terms interchangeably throughout this article:
This is worth pursuing - but you must ease into it
I don’t think buyers should be skeptical about AI and its applications. This absolutely is the way forward - and it’s a matter of identifying the right ways to apply it to your business.
And if you aren’t, your competitors certainly are. I expect in the coming years that AI enablement will be table stakes, but for the short term it offers a competitive advantage to those who adopt it well.
Innovative companies are probably doing a lot of internal legwork to make sure that they have the framework that can actually support AI models. Right now, they're likely talking about it without a ton of action - but that groundwork is being laid.
Some questions worth asking so you can build your own foundations:
How do I ease our company into this without inviting too much disruption?
What are the gains we want to make - and how can we identify if we’ve achieved it?
Who will build, maintain, and grow this - and do they have the bandwidth to do so? If not, what’s the opportunity cost?
Your infrastructure greatly matters
If your database is filled with bad data, incomplete data points, duplicate records, poorly-integrated technologies, or other issues that we run into when onboarding new clients - this needs to be handled first.
Because AI builds on your existing foundation and learns from your current data. If that data’s bad, then AI’s outputs will be just as poor.
Let’s take two examples:
If you’re in Sales and want a summary of leads you need to reach out to today, but your leads aren’t fully populated or a subset are the only ones being enriched from a data provider, your reports will be incomplete and the AI will train on a poorly-segmented dataset.
Time needs to be spent scrubbing the data and getting the database in good health - and that includes the surrounding processes and integrations that led to its current state.
If you’re in Support, having your various cases accurately tagged to the right products and with full information on customers asking for support are vital. Otherwise, AI will give you incomplete data to work with that may cause you more work.
Having updated help articles also matters - you don’t want Agentforce telling a customer to do something that worked on a previous iteration of your product, and was phased out! Ensuring your documentation is correct so the agent can train on it cannot be overstated.
Start with your Customer Support organization
This is one of the easiest areas to implement Agentforce and make quick gains - and they flow to other teams, which I’ll touch on in a moment.
In fact, some of these are in use now if you’ve chatted into a support queue at certain companies.
With AI, you can automate multiple interactions to keep your human agents focused on the more complex issues that truly demand their attention and expertise.
For instance, if a customer needs help with a password reset, connecting an integration, or a “how-to” question on using your product/service, they can be given information directly from relevant help articles. Some can even initiate a password reset now automatically.
One of my favorite examples I’ve run into is something a friend of mine, John, showed me - issuing a refund.
John used a meal delivery service, but a bag tore open and so a side item was ruined in transit. He chatted in, identified the product that had the issue, and was immediately offered a refund on that box equivalent to the side item’s value (around $10).
All automated and without needing to pull in someone to help!
Having experimented with different Agentforce builds, I’ve seen you can also set thresholds so the AI pulls in a person when a larger credit is required (i.e. “anything above $50”). There’s multiple guardrails that are easily put into place to keep things simple and safe.
The benefits are clear:
John took maybe two minutes to have his issue handled in a satisfactory manner and didn’t have to talk to a person when he didn’t want to
The meal delivery’s support team wasn’t pulled aside to handle this simple situation, instead getting to focus on higher-priority issues
Who should be involved?
Your agents should reflect your best humans - but in a scalable manner.
So, getting those people involved and understanding how they go about solving or finding the answers to customer related issues.
You also need someone that understands the data and how it flows - whether that's your Salesforce org and your object model or if you're using Data Cloud or another data warehouse.
You need people that understand the data, how it's connected, and how the agent is going to be able to access and bring some of that data to life for your customers and your employees. At minimum, that’s likely 4-5 roles involved to get the right.
Sales and marketing benefit
If a customer has a renewal coming up, having an Agentforce-enabled Support organization lets the AE or rep quickly understand how the account is doing based on Case and other account related records.
Or marketing can more easily identify customers likely to give a wonderful testimonial or be willing to join onstage for a speaking event or webinar.
These are simple steps, but Agentforce makes it more possible for those using Salesforce - and it drives more alignment because of the efficiency gains.
Everyone has questions
A couple of our clients have already earmarked Agentforce as a project that they want to tackle within the next 6-9 months. They've been waiting for the product and they're ready to implement, and I think they have good use cases and they have a lot of documentation ready.
So I think they're going to be good candidates, but they're definitely starting to ask questions such as: what other use cases can we leverage this technology for?
Lots of Salesforce Admins, especially those who’ve attended Dreamforce or a recent World Tour, are buzzing about Agentforce’s potential. It’s hard not to after getting your hands on the tool!
Leaders - and it depends on who owns the CRM, but it’s often Sales or IT - are also asking what gains can be made in the next 12 months as a business. They’re hearing from colleagues at other companies about the gains they’re making, and the C-level is obviously asking how they can get this for themselves.
From a budget perspective, it’s a priority - and CFO surveys back this up by showing a clear majority plan on investing more into AI.
One way I use AI now
We've been using AI internally for several months, and I really like how it saves effort with timesheets.
When DoubleTrack team members are working on a project, we're able to understand the different requests that they're getting from a client and where they're spending essentially each hour and what they're actually spending their time on.
We’re also able to quickly summarize this information for both internal resource management and for client status reports.
That's a very labor-intensive task when you have four or five people working on a single project and have to compile this information on your own.
Now, we still manually review it - and we’ve caught some errors. So the manual component is absolutely necessary … but it takes a lot of the grunt work of pulling data together off our plates.
This means we have more time to spend on improving our service offerings, communicating with clients, or exploring new products and services (such as Agentforce). The gains are very tangible!
Get started
First, I recommend really diving into what Salesforce already has. There’s help documentation and good ideas to help you identify use cases for your business.
Then, identify an owner for building your first agent - ideally around a customer service process such as automatic escalation or answering common chat questions.
Look at the data that flows into this and see what changes and updates are needed to the agent trains on the best data.
Then, get building - and be sure to test, then test some more!
And of course, if you want to move quickly, we’re here to help - simply get in touch today.