Why use Process Configuration Mining output to drive Agentification?
Agentification transforms business processes by introducing AI agents to perform tasks where they deliver speed, scalability, and better decision-making.
However, identifying where AI agents can add value in an Org cluttered with legacy automations and undocumented processes is extremely difficult.
Process Configuration Mining solves this problem by auto-generating an accurate business process diagram (in UPN notation) directly from live Salesforce configuration.This gives a structured, verifiable map of what is happening in the Org today — enabling you to:
Spot tasks that could benefit from AI pattern recognition, prediction, or recommendation.
Distinguish between tasks suitable for simple automation vs. those ideal for AI agents.
Quickly prioritize AI opportunities without guesswork or massive manual analysis.
When to use Process Configuration Mining diagrams for Agentification analysis?
Use the Mining output when:
You are struggling to come up with good Agent use-cases for your organization.
You need to identify realistic, high-impact starting points for introducing AI.
You need to demonstrate to senior stakeholders how AI Agents will help the business in context of the existing, custom ways of working.
Prerequisites
Enterprise plan with Process Configuration Mining enabled
How to use Process Configuration Mining diagrams to identify Agentification opportunities
Step 1: Generate a Process Configuration Diagram
Begin by generating business process diagrams for the important business objects that drive most of your business value or Salesforce adoption.
Step 2: Review and Amend the Process with Stakeholders
Process Configuration Mining can only account for what is built inside Salesforce. It scans automations, validation rules, user interface settings, permissions etc. But it cannot account for steps which are manual or done outside of the platform.
Example: An opportunity process consists of stage transitions (e.g. BDR qualified -> Sales qualified -> Business case -> Procurement -> Closed won). This, alongside any validation rules and record triggered automations firing on status changes will be accounted for through Process Configuration Mining.
However, it might be the case that the sales team, during / after moving Opportunity to 'Business case' have multiple calls with the prospect on platform like Zoom. And the recordings and transcripts are stored outside of the Salesforce platform, yet are big part of the process. This will not be detected by Process Configuration Mining.
It is therefore important to 'fill in' the missing pieces and context in a workshop.
Best practices for validation:
Engage the right stakeholders: Identify key individuals from different teams who understand and execute the selected process. This might include process owners, business analysts, and department heads.
Use Elements’ collaboration tools: Share the diagram directly in Elements, allowing stakeholders to add comments and sticky notes on specific process steps. Encourage asynchronous feedback, enabling more participation.
Ask targeted questions: If you are presenting the process diagram during a workshop or a call, ask questions that lead to actionable insights:
“Where in this process do people need to make a judgment call or check something that isn't tracked in the system?”
“What happens outside the system before this step is triggered—or after it ends?”
“Who else is involved behind the scenes in making this step happen—even if they’re not shown in the system?”
Capture any identified manual steps and steps happening outside of the platform.
Step 3: Analyze Verbs (Yes, really)
Automatic Agentificaton opportunity detection is currently in development and will become part of the product experience in Spring of 2025.
Every step in the UPN diagram is described with an active verb phrase. For instance 'Create a case' or 'Develop business proposal' etc.
Read about Universal Process Notation here.
This is very valuable because AI is not meant for every use-case. Some processes are better served with deterministic automations (e.g. flows, apex classes etc.), while others are ripe for Agentification. And we can spot those opportunities by simply scanning the business process for appropriate verbs.
Verbs Indicating Opportunities for Deterministic Automation
These actions follow a clear, rule-based logic. The inputs, decision rules, and expected outcomes are well-known and repeatable. No probabilistic reasoning is required, making them ideal for declarative (Flow) or programmatic (Apex) automation.
Verb | Why It's Deterministic? |
Calculate | Inputs follow formulas or static transformations (e.g., total = price × quantity). Precision is expected. |
Update | System fields are changed based on known logic (e.g., update Stage to "Closed Won" when Opportunity is 100%). |
Assign | Allocation is based on clear criteria (e.g., region = EMEA → assign to EMEA queue). |
Notify | Triggers are simple (e.g., notify when case status = "Escalated"). It’s a conditional rule, not inference. |
Route | Similar to assignment, routing decisions follow a static decision tree or logic table. |
Validate | Checks are made against known, unchanging rules (e.g., field must not be blank; date must be in the future). |
Action: Mark these for Flow or Apex automation — not AI agents
Verbs Indicating AI Agentification (Probabilistic Automation)
These actions require probabilistic reasoning, pattern recognition, interpretation or learning from data. Rules are fuzzy or too complex to be hard-coded. Human judgment is often used today. AI agents can scale and enhance this judgment.
Verb | Why It Needs AI |
Analyze / Verify | Involves pattern detection across large or unstructured datasets (e.g., verifying documents, spotting inconsistencies in logs). |
Predict | Forecasting involves trends, probabilities, and uncertainty (e.g., will this deal close? Will the user churn?). |
Recommend | Suggesting next steps depends on context, past behaviors, conversations, and preferences — not static rules. |
Classify | Often used with open-ended input (e.g., categorize a support ticket by intent). Requires models trained on labeled examples or making judgments against semantic criteria. |
Detect | Anomaly or signal detection (e.g., fraud, unusual config changes) cannot be reliably hard-coded due to changing patterns. |
Investigate | Involves exploratory reasoning, cross-referencing multiple data points, and making sense of incomplete or conflicting inputs. It's not just about applying rules, but figuring out what the rules might be in a given situation. |
Generate / Write | Content creation demands coherence, structure, domain awareness, and contextual nuance. Whether it's generating user stories, email drafts, test cases, or documentation, the task requires synthesis of multiple inputs, understanding intent, and producing fluent output. |
Action: Highlight these as prime Agentification targets.
Step 4: Assess Activity Context and Outcomes
For each AI-candidate activity, assess:
Input: Does it require large datasets or pattern recognition?
Outcome: Is it probabilistic (uncertain but optimizable) or deterministic (always same result)?
If input and outcome align with complex analysis or prediction, that points to strong AI opportunity.
Step 5: Prioritize Opportunities
Prioritize AI opportunities by evaluating:
Frequency: How often is the step performed?
Impact: How valuable is improving this task?
Feasibility: Is the necessary data available and structured?
First Target: High-volume, medium-complexity steps involving pattern analysis or classification.
Step 6: Annotate and Plan
Inside Elements:
Use sticky notes on candidate steps to record AI opportunity insights.
Group related activities into logical agent roles.
Create linked business requirements tagged "Agentification" for structured planning.
Tip: Consolidate simple decision steps into one agent where possible to maximize ROI.
Summary
Process Configuration Mining diagrams provide the critical, structured visibility needed for Agentification by:
Turning undocumented Org behavior into actionable process maps.
Revealing clear AI opportunities based on real-world activities and outcomes.
Supporting confident, prioritized, and measurable AI agent design.
Without this clarity, Agentification is slow and uncertain. With it, you can rapidly build scalable, intelligent automation at the heart of your operations that bring instant ROI.