Agents need good data (governance)

Step 1 – data governance not data cleanup

Salesforce’s report “The One Factor That Defines AI Agent Success” cites data as the key issue. AI is only as good as the data you feed it. Bad data is their junk food. Nourish them with clean, complete, connected data.

In this new era of AI and agents, customer data and metadata are the new gold for the enterprise.

Rahul Auradkar, EVP and GM for Data Cloud at Salesforce

Salesforce’s latest State of IT report found that IT decision-makers cited several barriers to using generative AI. 2 of the top 5 are data-related issues:

• Generative AI will introduce new security threats to their data (71%)
• Their organization lacks a unified data strategy (59%). 

But data is not just held in databases. It is application metadata. It is unstructured data in emails and documents. And it needs to be written in a way that AI can easily understand. AI is very literal, so it doesn’t understand nuances, the context a human would have, or the assumptions they would make. It can also be video and audio.

CoastCloud, a Salesforce SI, in their 2025 AI report, shows that organizations see data is a blocker to moving forward with AI.

Another factor is the level of Salesforce org complexity, the high levels of technical debt, and the lack of metadata documentation. The Elements.cloud Change Intelligence Series report showed that 51% of custom objects are never used, 43% of custom fields are never populated on average across all objects, and this climbs to 88% for core objects. The average number of fields on the opportunity page layout is 175.

When you say we need to clean up the data in the Stage field on Opportunity, can you tell me which one? We have 4 stage fields: 2 are picklists, one is a multipicklist, and the other is free text.

You will never be ready

But organizations need to realize that data will never be “good enough” and therefore they may never get started. This is a huge mistake. Organizations need to start building up their experience building agents and building new skills if they are going to stay ahead of their competitors.

The focus should not be on data cleanup, which is where most organizations seem to be focusing. We believe that this is not the right place to start. First, you need to understand the data flows, the lifecycle of the data, who can access it and where, and how it is governed. As orgs are so poorly understood and documented, getting this information is a time-consuming manual activity. But only once you have the information can you identify the data that matters, put the governance in place to keep it clean, and prioritize the cleanup. 

There is no point cleaning the pond if you still have polluted streams, including underground springs that you don’t even realize exist, running into it

What is Data Governance

Validity is an Elements.cloud customer. They are the most trusted name in customer data quality. Tens of thousands of organizations rely on Validity solutions to improve customer engagement and manage their customer data.  Here is their definition of Data Governance:

Data governance is a system for managing a company’s data throughout its lifecycle. It refers to set principles and processes that determine who can do what with information, under what circumstances, using specific methods, and when. Data governance also helps establish accountabilities and decision rights.

Some of the questions you need to ask are: 

What’s the lifecycle  

• how and where is it created
• How does it get changed, enhanced
• Where does it go to die (if at all)?

Security profile

• Who has access and at what level; CRUD, export, import
• How do they get access and at what point in the lifecycle

Regulatory compliance

• security; SOC2 or ISO27001
• Data Protection; GDPR, HIPAA, CCPA
• other industry standards

GDPR: The General Data Protection Regulation (GDPR) is a law that protects the personal data of European Union citizens. It was adopted in 2016 and went into effect in 2018. 

HIPAA: Health Insurance Portability and Accountability Act, a federal law enacted in 1996, primarily focuses on protecting the privacy and security of individuals’ health information.

CCPA: The California Consumer Privacy Act is a state law that provides California residents with greater control over their personal data by giving them the right to know, delete, opt-out of sale or sharing, and have their information corrected. 

Understanding your data

To answer most of the questions requires a detailed understanding of how your systems are configured, and for many customers, the heart of it is Salesforce and possibly Data Cloud. The bad news is that there probably isn’t any up-to-date documentation available. The good news is that the org metadata can be used to recreate, most if not all, of the documentation you need for data governance and clean-up prioritization.

Below is a diagram drawn automatically by Elements.cloud based on org metadata. In the app, you can zoom in, see the metadata attached to each box, drag on sticky notes, comment, and edit. 

It is the “opportunity process from a sales perspective”.  That was the natural language prompt used to generate the diagram.  What is interesting is the L shape of the diagram. Most diagrams are this shape because there are often multiple ways that the object is created. In this diagram, we can see that there are 5 ways that users can create an opportunity.  Is that really what was designed and expected? 

If we zoom into the first activity box, we can see who can access it and where it is visible. But also attached to the activity box, we can see the metadata that configured it, including the permissions. We can dig in to see what level of access they have. Again, was it the intention that they should all be able to create an opportunity?

You can then drill into the fields on the page layout to see the data population. You can then decide which ones can be removed from page layouts to simplify the UI, which ones need validation rules to improve data quality, and what permissions to remove.

These diagrams are created by Elements.cloud’s Process Configuration Mining capabilities.  There is also an equivalent capability that builds ERD / Data model diagrams based on a natural language prompt.

Process Configuration Mining provides data governance insights

Process Configuration Mining can take the 100,000s of metadata items in the enterprise applications – databases, fields, security permissions, automations, validation rules, metadata dependencies, and complexity analysis. It uses AI to take an analyst’s request for a specific view of the system, and then draw a very detailed process diagram and data model, putting all the metadata in context.

This is changing how we think about documentation. Instead of creating documentation and leaving it on the shelf to gather dust, you create it on demand when you need it, from a specific perspective.

“Process Configuration Mining” is different (and more powerful) than Process Mining. 

• Process Mining tells you what paths users have been taking through application screens based on tracking users over a set period of time.

• Process Configuration Mining tells you every possible path that different user types could take based on the app configuration.

The difference is that Process Configuration Mining is the map of the city with all the dark alleys and cut-throughs. Process Mining shows the most popular routes over a set period of time, provided there are sensors everywhere.

It is a current, accurate view of how the org works. 

• Not how you all thought it worked. 
• Not how the documentation (if you have any) claims it works. 
• Not how people remembered it works. 
• Not how the design documentation says it works. 

This is HOW IT ACTUALLY WORKS.

Here is a 1-minute summary

For a deeper dive to really understand how it works and the implications

Final Word

AI presents a valuable opportunity to revisit and strengthen foundational data quality and governance practices. By proactively addressing data issues, engaging stakeholders, and implementing AI in a measured and iterative way, organizations can ensure that AI initiatives deliver meaningful insights and drive business growth, rather than creating confusion or being based on unreliable information. 

The benefits of focusing on data quality extend beyond AI adoption, improving overall data-driven decision-making across the organization. Process Configuration Mining offers rapid, valuable insights to be able to prioritize the work.

Find out more and see what secrets Process Configuration Mining can reveal in your org.

Image: steve-johnson-VVxQEpum11g-unsplash