Agentforce allows organizations to build and manage autonomous agents for tasks across various business departments – in fact, there’s an “Agent in every app”. There’s the potential to build Sales Agents, Service Agents, Marketing Agents, Commerce Agents, and Platform Agents. Agentforce for Service and Sales is expected to be generally available by the end of October 2024.
As it’s such a new technology – and having been hyped up at Dreamforce – there are plenty of questions to be answered. We’ll dig into some of these, sharing what we know so far and where the question marks still lie.
1. How Does Agentforce Differ From Other Salesforce AI Solutions?
On the surface, Agentforce may seem like the next iteration of chatbots. However, this is too simple of an assumption to make.
In the progression of artificial intelligence, with agents, we’re entering the third wave. This is following ‘predictive’ and ‘copilots’.
Salesforce has been in the predictive AI game since 2016; the year that Einstein was introduced. The growing collection of features over the years has included Einstein Lead Scoring, Pipeline Inspection, Einstein Bots, and more.
Copilots (wave two) arrived with Einstein 1 (formerly Einstein GPT). These were described as your ‘sidekick’, following users wherever they needed to ask a question, and copilots would generate a response.
The biggest difference between Copilots and Agents is that Agents are autonomous. While Copilots/chatbots can handle a variety of tasks (e.g. generating emails, responding to FAQs), they rely on the inputs/outputs being stringently configured by someone. Backed by the Atlas Reasoning Engine, Agents are autonomous as they can retrieve the right data at the right time (via Data Cloud which underpins the Salesforce platform), and therefore are able to make smarter decisions.
Plus, they are easier to access across multiple channels than a classic Copilot/chatbot – in other words, not strictly an employee-facing tool nor a chatbot embedded on only one channel.
A good way to summarize this evolution is to look at service as an example – from Live Agent, to Einstein Bots, Einstein Copilot for Service, and now Service Agents (see how these work here).
2. Who Gets Access to Agentforce?
Customers with Enterprise, Unlimited, Performance, Einstein 1 (formerly Unlimited Plus) editions will be eligible for Agentforce. As we will cover later, Agentforce operates on a consumption-based pricing model, centered around conversations.
Purchasing conversations will give you other entitlements, such as a certain number of Einstein requests (also explained later in this guide) and Data Cloud credits.
Having Data Cloud configured is a prerequisite – that being the data streams and data model objects. There are talks of a freemium consumption model, but what exactly this means is to be confirmed.
3. What Level of Customization is Available Within Salesforce?
You may have heard that building Agents is a clicks-not-code exercise. This is thanks to Salesforce being thoughtful in their design and providing you with the essential building blocks.
Topics and topic actions are foundational in that they prescribe to the Agent what it can do. These help the agent route to the best action(s) it should take in a given scenario.
Actions are actually flows (or prompts, apex, Mulesoft APIs) behind the scenes (which you will see shortly) which are then assigned to topics. This creates a grouping of actions that are then assigned to an agent.
When creating a new action, you select a reference action type (Apex, Flow, a prompt, or Mulesoft API). Then, reference the library of processes/APIs that have already been created.
There are standard prompts available like sales email generation, service email generation, service reply creation, and knowledge search.
In terms of industries, there are 100+ industry-specific standard prompts that you can use to command agents. Scan the QR code below to browse the Industry AI Use Case Library.
4. What Metrics Can Measure Agentforce’s Impact on Productivity?
The goal of Agentforce is to alleviate the overhead of the work human users traditionally had to carry out, freeing them up to focus on higher-intensity tasks. ‘Productivity’ is a tough term to place a finger on – it’s often up to interpretation from one business to the next. Here are some examples:
- Number of tasks completed by agents – e.g. resolved cases.
- Number of tasks handed to humans: Not necessarily a negative if this metric is high. Instead, this could be interpreted as agents having taken the starting legwork away from a human user.
- Time savings (Speed to resolution/Average handle time).
You may already be familiar with Omni Supervisor, originally a feature tied to Service Cloud for managers to oversee teams of customer service agents. Now, Omni Supervisor is being repurposed for agents. You can see some of the metrics above featured on the Omni Supervisor dashboard below:
5. What Counts as a Conversation?
Agentforce pricing starts at $2 per conversation. But, what exactly is a conversation?
A conversation has a start and an end. The starting point is when an Agent responds for the first time. A conversation can end in a few ways:
- Transfer: The query is transferred to a human user.
- End: The customer ends the interaction.
- Time: The query goes over 24 hours since being initiated.
Each conversation consumes a certain amount of Data Cloud credits and Einstein Requests – which leads us to ask…
6. What’s an Einstein Request?
This is any request that goes through the LLM and is consumed when it reaches the gateway.
As far as we know, these requests are consumed in line with how many words are passed back in order to respond to a query. You can read more and stay updated on any changes with the Salesforce Help article.
7. How Will Pricing Scale With Consumption?
At $2 per conversation, there will also be volume discounts on offer. The extent to which discounts will be applied is to be confirmed, but one can assume that plenty of pricing negotiations will be taking place.
Salesforce professionals have already been crunching the numbers to get an indicative price per conversation that would make it worthwhile for them based on the number of human users, how many cases they are expected to resolve (e.g. per hour), and how much the human users earn.
Salesforce predicts that customers will see huge ROI. Agentforce agents are more cost-effective than back-and-forth conversation with a human agent in freeing them up to deal with more nuanced queries.
8. Could Agentforce Impact Service Cloud Licenses?
If a business is to invest in Agentforce and conversation credits, will this lead to fewer Service Cloud licenses being sold? Service Cloud works on a per-seat basis, which is more predictable for organizations to budget for than consumption-based pricing.
There are two sides to this argument, which boils down to a) the nature of the requests a business receives, and b) the needs in other parts of the business that are lacking resources. If an organization has a complex product/service, or a high-value/long sales cycle, agents will not replace human users but instead will alleviate any low-level workload. Perhaps organizations will re-skill employees into more nuanced roles, such as customer success which needs a human touch.
Then again, perhaps even if fewer Service Cloud licenses are sold, the revenue gained from conversations balances this out. Of course, this is not something Salesforce expects, as they continue to innovate heavily on their core products (Sales Cloud, Service Cloud, etc.) and are enacting ‘land and expand’ initiatives, such as Salesforce Foundations.
9. Does Agentforce Replace the Einstein 1 Platform?
The Einstein 1 platform (the revised name for Einstein GPT) was introduced to us at Dreamforce ‘23. Einstein 1 was a relaunch of the Salesforce platform, guarded by the Einstein Trust Layer that gives your organization peace of mind when it comes to data privacy and security. Under this umbrella, Copilot Studio was the name given to a suite of tools: Prompt Builder, Skills Builder, and Model Builder.
Agentforce now takes this a step further. Agentforce has surpassed its predecessor (Copilot Studio), to include Agent Builder alongside Model Builder and Prompt Builder tools mentioned above.
So, does Agentforce replace Einstein 1? The simplest answer is yes in terms of brand name – but with the addition of the Agent Builder within the suite of tools.
10. Does Agentforce Replace Prompt Builder and Model Builder?
Prompt Builder and Model Builder were first introduced to us as part of Copilot Studio. As mentioned, Agentforce Studio has surpassed Copilot Studio in terms of brand name. However, Prompt Builder and Model Builder are still very much ‘alive and kicking’.
In fact, Agentforce agents rely heavily on both Prompt Builder and Model Builder – any current investments you’ve made into both will accelerate your Agentforce deployment.
Summary
Through these questions, you may have sensed there’s a mix between what’s new and what’s been a rebranding exercise. The key takeaway is that Agentforce is geared up to allow us to enter the next wave of AI (graduating from predictive and copilots, to autonomous) – there’s no doubt about this with the Atlas Reasoning Engine innovation. However, the foundational elements that we’ve heard about before are very important – that’s Prompt Builder, the Einstein Trust Layer, Data Cloud, and (of course) good data quality.
While Agentforce pricing still may seem up in the air, hopefully, the advice given here has shed some light on the situation.
What will be interesting to see is Salesforce professionals getting their hands on the Agentforce technology, and see the use cases that are prioritized, the implementation and testing, and the impact of Agentforce.
