An organization first needs to understand what data we have, where it is, what the quality of it is, what the source of it is, and the permissibility of using it.
Jared AmaralData & Analytics Practice Development and Delivery @ VerndaleThe challenge of getting data ready for AI is not new. Reporting, analytics, marketing automation, and personalization have always depended on the same thing: data that is accessible, clean, well-understood, and permissible to use. Organizations have been working on this for years and, for the most part, haven't solved it. What AI changes is the speed at which weak data foundations produce consequences. An agent that acts on the wrong data doesn't just generate a bad report. It takes action at scale.
Jared Amaral leads Data & Analytics Practice Development and Delivery at Verndale, a digital experience and technology consultancy. He has over 20 years of experience leading enterprise SaaS initiatives across data strategy, customer data platforms, and risk management, including senior roles at Bain & Company, FordDirect, and Deloitte. He built FordDirect's first data strategy team and led its first enterprise data governance program.
"An organization first needs to understand what data we have, where it is, what the quality of it is, what the source of it is, and the permissibility of using it," Amaral says. "Data availability and suitability for use is not a new challenge unique to AI. It has always been the challenge."
Agents need context humans already have
Without human-provided context, AI breaks in concrete ways. A revenue field in one system may represent total enterprise revenue. In another, it covers a single business unit. "You need organizational agreement around what the data actually is," Amaral says. "Does this revenue field include only sales revenue, or does it include other sources? If people don't agree, the agent won't know the difference."
A personalization example shows the same risk at scale. An organization wants to generate personalized marketing messages based on purchase history. The campaign targets customers for repeat purchases. But somewhere in a call center system, records show that certain customers called in to express negative sentiment about the same product. "Trying to sell that product again to the same person would be a tone-deaf experience," Amaral says. "You want to be perhaps a little more conciliatory or offer a discount or win back the trust."
Consent is the wall most organizations have not audited
Amaral describes a multi-billion dollar prospect that had just signed advertising activation partnerships and planned to use its consumer data for broader reach. When they examined the data, the question was immediate: can we actually do this legally? "They said, we don't really know if we can do this," Amaral says. "They needed to evaluate how the data came to them, what rules and promises applied, and whether they had consent to use it the way they wanted."
He frames permissibility around four constraints: internal policies, laws and regulations, privacy statements, and trade partner contracts. "If we are acquiring data through a partner, does the contract permit me to use the data in the way that I want?" Amaral says. "Those four buckets set forth obligations and constraints that organizations need help navigating."
Start small and test before you scale
Executives want to move fast, and vendors demo in perfectly curated environments that make everything look seamless. Amaral pushes back on the instinct to go all-in. Production reality includes stale databases, inconsistent naming conventions, permissions scattered across systems, and data that has never been contextualized for machine consumption.
"You don't know what you don't know," Amaral says. "Start small in a meaningful way, evaluate, and improve iteratively. A test-and-learn mindset will reveal the blind spots, and then you can direct attention toward them." The alternative, deploying at scale without that foundation, produces output that looks acceptable on the surface but requires constant human review that cannot scale. "Having better quality data that's organized and contextualized and on which there is organizational alignment exceeds volumes of data for which there is none of those things," Amaral says.
The views and opinions expressed are those of Jared Amaral and do not represent the official policy or position of any organization.
