Beyond the AI Pilot: Transforming Customer Data into Personalized, Scalable Commerce

Business enthusiasm for AI has outpaced business results. Stanford’s 2026 AI Index reports that 88% of organizations used AI in 2025, but adoption of AI agents remained in the single digits across nearly all business functions. The gap between experimentation and value remains wide and is very visible and tangible in the customer experience. There are two key points to consider when trying to understand why this gap exists.

The first it is organizational maturity on artificial intelligence. Most companies are still in the “individual productivity” phase, building a scalable foundation of governance, data availability, and a portfolio of proven use cases. They still have a long way to go before they reach an “autonomous” phase where AI agents can operate within defined guardrails and people focus on oversight and exception handling. Reaching this stage of AI maturity is critical to attempting a “native AI operating model.” We’ll talk about it later.

The second it is the operational gap between data and AI-driven business decisions. Most organizations already have investments in data and systems. They have amassed customer data, adopted marketing automation, migrated to the cloud, and experimented with generative AI. Even so, customers continue to receive irrelevant advice, disconnected channels, and service interactions that ignore what the company already knows about them.

This is rarely a model problem. More often, AI simply exposes what was already broken: fragmented data, unclear ownership, disconnected workflows, technical debt, and business goals that were never translated into measurable use cases. The path forward doesn’t start with a broader AI initiative. It all starts with a few well-chosen customer decisions, a clear operating model, and an architecture that turns intelligence into action.

For leaders looking to move from isolated pilots to a repeatable system, some priorities matter more than others.

It starts with a decision, not a skill.

The first question is not “where can we implement generative AI?” It’s about “what customer experience decision do we need to improve?” An airline wants to give passengers better options during a disruption. A retailer wants to help a shopper choose the right product without increasing returns. A bank wants a service representative to resolve a complex issue without searching through systems. Each of these names a decision, a person and an outcome, which in turn indicates what data is required and how success should be measured. Starting here also helps leaders say no. If an application idea or agent request cannot be tied to a real decision and a measurable outcome, it does not deserve priority.

“Using AI to improve the experience” is too broad. It produces demonstrations, not operational capabilities.

When a consumer electronics retailer adds an “AI shopping assistant” today, that means a generic chatbot, judged by how many people use it. Reframed as a decision that helps a buyer choose the right laptop, it aims for higher conversion and fewer returns. It then becomes a tool that uses real-time specifications to better understand shopper needs, with success measured in sales, returns and satisfaction. Same idea, real result.

Treat unified data as a feature, not a database project.

A “360-degree customer view” is often described as a data centralization effort. In practice it is an organizational agreement on the identity, consensus, ownership, quality and meaning of key events. Does a browsing session belong to a known customer, a household, or an anonymous device? Which system is authoritative on consent? How quickly does a service interaction need to emerge in another channel? Without these answers, data centralization creates the appearance of untrustworthy integration. The most productive move is a shared data contract for each prioritized use case, covering the required fields and events, their authoritative sources, acceptable latency, permitted uses, and the quality team. The goal is not all the data. It’s the right data, reliable when a decision is made.

A car dealer wants to welcome a returning online shopper with the exact model he was configuring. Today this breaks down, because the configurator, the dealer CRM and the financial system each handle a different version of the customer and disagree on whether or not a three-week-old lease quote is valid. Reframed as a journey with clear data rules, it is established who owns what: the configurator is the source of truth for the build, the financials carry a defined consensus and deadline, and the inventory feed updates within minutes, so the buyer never sees a car just sold by the dealer. Much simpler than “unify all customer data” and the greeting actually works.

Design customization as a continuous decision-making system.

Many programs are still campaign drivers. Teams define a segment, prepare content, and send it on a schedule. Generative AI produces more variations but leaves the model unchanged. Context-sensitive experience works differently. It observes the customer’s current context, determines available actions, chooses one within business and political constraints, delivers it through the right channel, and then learns from the response. Fundamentally, individualizing each interaction is not the goal. Sometimes the right action is a standard message, a human step, or nothing at all. Effective personalization is not measured by generated variations. It is measured by whether the organization made a decision that was most relevant to the customer and whether it met the customer’s expectations in doing so. Confidence belongs to experience, not to a later compliance review.

A bank notices a customer repeatedly checking mortgage rates. Today, a campaign engine inserts them into a series of “first-time buyer” emails at the next scheduled send. Reframed as an ongoing decision, the system reads the moment: Is anyone considering purchases mid-purchase or simply curious about the months ahead? Unsure, you need to act with moderation, showing a helpful “do you want to explore what you can afford?” a timely push rather than a premature “get pre-approval today” that misreads where they are. If they put in the effort, it learns and adapts. If not, he holds back. Sometimes the most important decision is the quietest one, and this moderation is what preserves trust.

Resolve content and architecture constraints before accelerating.

Generative AI can create text and images at remarkable speed, but the speed applied to a broken content operation only shifts the bottleneck from creation to review and governance. The Content Marketing Institute found that 75% of enterprise marketers use AI, but only 15% integrate it into their daily workflows. Structured, reusable content—approved facts and components that can be assembled across channels—allows AI to adapt without inventing the underlying truth. The same discipline applies to the technology stack. Instead of redesigning everything under the banner of “composable,” map out the few paths tied to priority use cases, figure out where legacy systems create latency or block data access, and incrementally modernize around those constraints. Progress is not the number of technologies adopted. It’s the ability to change the customer journey safely and quickly without repeating the same integration work every time.

A retailer instructs the AI to write descriptions for 40,000 items. Today, this floods legal review, because claims and warranty terms must be checked on every item. Reframed, the fix is not a faster model but a governed source: a PIM that holds approved specifications, claims and warranty terms as reusable components from which the AI assembles itself, then adapts the tone without making up facts.

The same logic applies to the stack. Instead of reorganizing everything, the team just modernizes the old inventory system by blocking “in stock near you” customization in real time and leaves the stable systems aside. Progress is rapidly changing a journey, not adding tools.

Build governance into delivery and measure value as a portfolio.

Governance considered as the last stage comes too late, after choices that are difficult to reverse have already been made. Frameworks like the NIST AI Risk Management Framework are most useful when they model a use case from the start, covering privacy, bias, brand integrity, trust and traceability, with controls appropriate to the risk of the interaction. Value, meanwhile, is rarely a large number. It is distributed across customer outcomes such as satisfaction, task completion and retention, operational outcomes such as cycle, resolution and reuse times, business outcomes such as conversion, cost per service and lifetime value, and risk outcomes such as error rates, complaints and replacements. Everything should be measured against a baseline and net of the total cost of operating at scale.

An airline operates an AI agent to handle disruptions. Today, answering “is my flight delayed?” it is low stakes and fully automated. Rebooking a passenger or issuing travel credit above a predetermined amount leads to a human being, because one wrong call is costly. Reframed, value is measured by outcomes that matter: faster recovery during outages, rebooking that protects high-value loyalty members, and careful complaint control. Handled well, an interruption becomes a moment that preserves a flyer rather than loses one, and this is where the true value of life comes from.

The question of maturity.

The companies that create lasting value will not be the ones that conduct the most experiments. They will be the ones with a repeatable way to connect customer needs, reliable data, governed intelligence, reusable content and measurable results, along with the honesty of knowing where they are on that curve today. Most organizations are further behind than their pilot activity suggests.

An AI-native organization is one in which AI is woven into every layer of how it operates and competes. Rather than augmenting existing processes, the operating model itself is designed around artificial intelligence, with intelligence as the core business capability.

So, the most useful question for a leader is no longer “where can we add AI?” It’s about “what decisions can we improve, and what must be true for that improvement to scale responsibly?”