Why the future of commercial data depends on helping AI agents find the right information, not just more information.
I’ve spent much of my career helping organizations improve the quality of their customer data. Whether the problem involved duplicate records, incomplete profiles, outdated contact information, or inconsistent company details, the objective was usually the same: make the data more accurate so the business could make better decisions.
But as AI agents begin taking on more responsibility for those decisions, I think we’re approaching a fundamental shift in how organizations will evaluate, acquire, and use data.
Accurate data isn’t always decision-ready data
Imagine a Salesforce administrator saying their CRM is 98% accurate. The accounts are properly matched, duplicates have been resolved, and contact names, phone numbers, and email addresses have been verified. That’s an impressive accomplishment, but I’d ask a different question: Does that information tell your sales team who they should be talking to, why they should be talking to them, and whether now is the right time?
Having an accurate email address tells me I can contact someone. It doesn’t tell me whether I should.
Knowing someone’s job title doesn’t necessarily tell me whether they’re a buyer, a decision-maker, or an influencer. Knowing that a company demonstrated interest in a particular technology doesn’t tell me whether that interest exists today or whether it was something they were researching a year ago. The information can be completely accurate and still lack the context that determines whether a sales opportunity succeeds or fails.
Traditionally, we’ve evaluated data quality through dimensions such as accuracy, completeness, consistency, uniqueness, and timeliness. Those measures remain essential, but they don’t necessarily tell us whether the information is sufficient for a particular business decision. A customer record can meet every defined quality standard and still leave the business without the information it needs to act intelligently.
I think that distinction between clean data and decision-ready data becomes much more important when the consumer of the information is an autonomous AI agent.
When agents need to find their own answers
Consider an AI agent responsible for evaluating an inbound lead and determining the appropriate next action. It has access to a verified contact record, accurate company information, and an indication that the organization demonstrated buying intent eight months ago. The information might be accurate, but is it enough to determine whether the company is actively evaluating solutions today? Is the contact still involved in the buying process? Have the organization’s priorities changed?
A human salesperson might recognize those gaps and research the account before deciding what to do. An AI agent should be capable of doing something similar. It can identify what information is missing, determine which available tools or external sources might provide it, retrieve additional context, and evaluate whether it now has enough information to proceed.
The important distinction is that the agent doesn’t need every piece of information to exist in the CRM before the workflow begins. It needs to understand what the decision requires and have the ability to obtain the information necessary to support it.
This is where I believe the relationship between AI and commercial data providers becomes particularly interesting.
The right data depends on the decision
Throughout my career, I’ve worked with data providers that offer different types of company and contact intelligence. One thing I’ve learned is that two providers can supply accurate information about the same organization while representing that organization in very different ways.
Dun & Bradstreet and ZoomInfo offer a useful example. D&B has well-established corporate linkage capabilities, including legal ownership relationships and global corporate family structures. ZoomInfo provides company hierarchy information as well, but its broader intelligence platform is strongly oriented toward sales and go-to-market use cases, including organizational structures, key decision makers, buying signals, and technology adoption.
The distinction isn’t that one provider has accurate hierarchies and the other doesn’t. It’s that the information available, the way relationships are represented, and the additional context surrounding those relationships may differ. An agent needs to understand those differences and select the source and attributes most appropriate for the decision it’s making.
Imagine an AI agent trying to determine whether two companies belong to the same corporate family for a compliance-related process. It may need a legally oriented ownership hierarchy. Another agent might be trying to understand how a sales organization should manage related accounts, assign territories, or identify expansion opportunities. That agent may benefit from a different representation of the same organization.
Both agents need accurate information, but they don’t necessarily need the same information.
I believe agents should be capable of recognizing these differences and selecting the appropriate source based on the problem they’re trying to solve. That requires more than simply exposing a collection of APIs or tools. The agent needs meaningful metadata explaining what each provider offers, where its information is strongest, how current it is, what limitations apply, and when one source may be more appropriate than another.
It also needs to consider the cost and consequences of obtaining additional information. If an approved provider can answer a question with sufficient evidence at a reasonable cost, there may be little reason to consult with several others. But if the results are ambiguous, or the decision carries significant consequences, seeking corroboration from another source may be justified.
Autonomy requires guardrails
This is an important distinction because I don’t believe organizations should give agents unrestricted access to every available data provider and simply hope they make good decisions.
Business owners, data quality professionals, and AI product teams need to collaborate on the objectives, instructions, tools, and guardrails that govern agent behavior. The business understands the decision and the consequences of getting it wrong. Data quality experts understand the strengths, limitations, and reliability of the underlying information. AI product teams understand how to make those capabilities available through tools and workflows while establishing appropriate operational controls.
Together, they can define which providers are approved, what information is required for particular decisions, when additional evidence should be acquired, and when the agent must defer to a human. Within those boundaries, I believe agents should have considerable flexibility to determine how best to accomplish their objectives.
An agent evaluating a sales opportunity might begin with information already available in the CRM, recognize that a buying-intent signal is outdated, and retrieve more recent information from an approved provider. It might discover that the original contact is no longer involved in the purchasing process and use another source to identify the current decision-maker. If the evidence supports proceeding, the agent could recommend an action or initiate an approved workflow. If the information remains ambiguous, it should recognize that limitation rather than manufacture certainty.
That last point matters. An agent’s confidence in its own conclusion is not proof that the conclusion is correct. The reliability of its decision depends on the quality, provenance, relevance, and consistency of the evidence it uses. Organizations need to evaluate agents not only on whether they complete tasks, but also on whether they select appropriate sources, identify conflicting information, respect their constraints, and recognize when they don’t have enough information to act.
A new competitive landscape for data providers
I think this creates an important opportunity for the commercial data industry, but also a new competitive challenge.
Historically, commercial data providers have competed to become an organization’s preferred source of company, contact, and market intelligence. Buyers evaluated coverage, accuracy, depth, integrations, licensing models, and price. Once a provider was selected, its information was integrated into established business processes, with people deciding when and how to use it.
We’re already seeing this model begin to evolve. Dun & Bradstreet has announced agent-oriented integrations involving MCP and AI platforms, and ZoomInfo has also introduced MCP capabilities for accessing company, contact, and go-to-market intelligence. These developments suggest that commercial data providers increasingly recognize AI agents as another important way their information will be consumed.
Making data available isn’t enough
But making information accessible through the Model Context Protocol, or MCP, is only the beginning. The greater challenge is helping agents understand when to use a particular source, whether its information is appropriate for the task, and whether the evidence returned is sufficient to support a decision.
Commercial data providers won’t compete solely on the information they possess. Within the ecosystems where they are approved, they’ll also compete on how effectively agents can discover, understand, evaluate, and consume that information. A provider might have excellent data, but if its tools are poorly described, its coverage is unclear, or its responses don’t provide enough context to evaluate reliability, an agent may struggle to use it effectively. Clear capability descriptions, meaningful metadata, provenance, and transparent usage costs could become increasingly important differentiators.
This doesn’t mean enterprise purchasing decisions disappear or that agents should operate in an unrestricted marketplace where they can independently purchase information from any available provider. I believe that would introduce significant cybersecurity, privacy, and governance risks. Organizations will continue to negotiate contracts, establish approved providers, manage costs, and define security and compliance requirements. Within those approved ecosystems, however, I expect agents to have increasing autonomy to determine which source is most appropriate for a particular task.
The distinction is important. An organization might authorize an agent to access several commercial data providers, each with different strengths and capabilities. The agent should be able to evaluate those options and select the most appropriate source based on the information it needs, the reliability of the source, and the cost of obtaining the data. But that autonomy should operate within clearly defined boundaries, with appropriate controls over access, usage, and the handling of sensitive information. Those boundaries cannot depend solely on instructions given to the agent. Approved-provider lists, access permissions, usage limits, and restrictions on sensitive data must also be enforced by the systems and tools the agent uses.
Even within an approved ecosystem, the information returned by an external provider shouldn’t automatically be treated as trustworthy. Agents need to evaluate the quality, relevance, and provenance of that information while maintaining a clear separation between external data and the instructions governing their behavior. The objective isn’t unrestricted autonomy. It’s giving agents enough flexibility to make intelligent decisions without sacrificing the security, governance, and accountability organizations require.
Data quality becomes part of the agentic workflow
I think this also creates an interesting opportunity for the data quality industry. Historically, much of our attention has been focused on the quality of information stored in systems like Salesforce. We’ve built tools to cleanse records, resolve duplicates, validate fields, enrich profiles, and monitor data quality over time. Those capabilities remain essential, but as AI agents begin retrieving information from multiple sources and using it to make decisions, the scope of data quality needs to expand.
Organizations will increasingly need to evaluate information as it moves between external providers, AI agents, and business applications. Two approved providers might return different values for the same attribute or represent corporate relationships differently. An agent needs a way to understand those differences, determine whether they reflect genuine contradictions or simply different business perspectives, and decide which information is appropriate for the task.
That doesn’t mean every discrepancy needs to be resolved into a single universal answer. In some cases, both sources may be correct within their intended contexts. The challenge is ensuring that the agent understands those contexts and that the information it ultimately uses is appropriate for the decision being made.
I believe this is where data quality becomes an integral part of agentic workflows rather than a separate activity performed before or after them. That shift will require data quality capabilities to become more accessible to AI agents, allowing them to evaluate, validate, match, and enrich information as part of the decisions they’re making. Validation, matching, enrichment, source evaluation, and governance become capabilities that agents can invoke throughout a process as they encounter new information or uncertainty.
The goal isn’t to eliminate human judgment. It’s to incorporate more of that judgment into the design of the system, where business experts, data professionals, and AI teams establish the objectives, boundaries, and evaluation criteria within which agents operate.
I’ve spent much of my career helping organizations make their data more trustworthy. I still believe that foundation is essential. But as AI agents make more decisions and take more actions on behalf of businesses, I think the question we ask will change. It won’t simply be, “Is our data accurate?” It will be, “Does our AI have the right information to make this decision, and does it know what to do when it doesn’t?”
I believe the organizations that learn to answer that question will realize the greatest value from autonomous AI. And the commercial data providers and data quality platforms that help agents reliably answer it will have an important role in shaping what comes next.
