Agentforce is transforming how work gets done, from triaging cases to querying records and engaging customers. Even as adoption accelerates, the question remains: How much time do agents actually save us? If a task takes a human five minutes and now an agent can do it in seconds, does that mean we’ve saved five minutes?
The answer isn’t so straightforward, because that logic assumes that the task was clearly defined, frictionless, and self-contained. In practice, human workflows don’t look like that. So our Research & Insights team at Salesforce set out to measure the benefits of using Agentforce by understanding the work humans do today and how agents can reduce that effort.
Our research uncovered a clearer, more realistic way to measure the return on investment (ROI) in Agentforce.
Here’s what we’ll cover:
Mapping agent use cases by task complexity and frequency
Where agents deliver value – or don’t
How to apply our evaluation framework
How to design agents with intention
Mapping agent use cases by task complexity and frequency
Our research team surveyed more than 2,700 internal sellers to understand:
- How sellers work today and identify tasks where agents can significantly save time.
- Benchmarks for the time spent on tasks without agent assistance and the frequency sellers perform tasks.
- The most challenging aspects of seller tasks, elements that sellers believe require more human judgment or automation, and how agents currently support their work (or don’t).
It was important to draw a distinction between technical model benchmarks and the reality of how people work. Based on our findings, we built a framework to evaluate agent actions against three key dimensions:
- Task complexity: How difficult is it to complete the task today without an agent? For example, we looked at the number of steps, the amount of necessary human judgment, and data requirements.
- Action maturity: How reliably and at what level of quality does the agent complete the workflow today? And how ready are our systems and processes to support agent automation of this work?
- Frequency: How often do users perform this task?
By mapping agent use cases across these axes, we can forecast realistic time savings and spot where agents can provide the greatest potential time savings. It can also indicate where further process review or system integration is necessary to achieve maximum ROI.
Where agents deliver value – or don’t
To identify tasks with high potential ROI, we benchmarked seller tasks across a range of CRM use cases, from account lookup to forecasting and sales coaching.
Here’s what we found:
Effortful, frequent tasks deliver the greatest ROI: When agents support actions that are both demanding and repetitive (for example, writing sales outreach emails), the time savings are tangible and add up over time and the number of users.
Highly complex tasks still require significant human judgment and time: Workflows that often involve nuance, cross-functional alignment, and strategic decisions are better suited for agent augmentation (for example, supporting parts of the task), not full automation.
Low-complexity tasks are easy to automate, but may be low-impact: While it might be simple to automate tasks like completing an opt-out form, it may not yield meaningful time savings if it isn’t a frequent occurrence. Their real value comes when the task is embedded in scalable design patterns that can be reused across multiple actions or use cases.
How to apply our evaluation framework
To apply the framework for yourself, you first need to determine the complexity of the task you want agents to automate, along with action maturity. If you aren’t deeply familiar with the task yourself, consult a high performer. Learn from proficient users about their perception of task complexity and maturity of current solutions.
Task complexity
To determine task complexity, start with the following:
- Count the steps: How many distinct steps are required to complete this task?
- Check task interdependencies: Does this task rely on or impact other tasks?
- Assess sources and navigation complexity: Does the task require data from and/or switching between multiple systems?
- Assess human judgment: Does completing the task involve required human expertise or decision-making?
The higher the value for each of the steps above, the more complex the task is.
Action maturity
After you understand task complexity, you can evaluate current action maturity by:
- Reviewing existing capabilities: Is there current functionality that handles or partially handles this task?
- Assessing action reliability: How consistent and accurate are those actions today?
- Identifying data sources and readiness: How many sources are required, and which ones? Are the sources complete, accurate, and up-to-date?
- Checking integration status: Are these sources already integrated with existing actions or this agent?
Greater comprehensiveness and reliability in the steps above reflect higher action maturity.
After you understand task complexity and maturity you can identify potential ROI at the task level. Then multiply that task by the frequency it’s performed to get cumulative time savings. If this task has limited potential return, consider whether other tasks or different components of the same workflow could yield more productivity gains.
How to design agents with intention
There’s still plenty we all need to figure out when it comes to working with agents. But we can confidently say that automating tasks just because they’re “simple” often won’t result in noticeable productivity increases. However, it is possible to design agents for measurable time savings if you’re thoughtful about your approach.
Grounding design in a deep understanding of the real work informs how you can intentionally reduce human effort. That time savings then can be reinvested into higher-value activities. With meaningful productivity gains from using agents, humans can then focus on higher-impact work that drives key business outcomes like growth, customer satisfaction, and innovation.
