You return a product. Two days later, an ad for that exact item follows you across three websites. You’ve been a loyal customer for six years. Your favorite retailer emails you a “welcome” offer meant for first-time buyers. You start planning a family vacation, and by the third site in your research, you’re typing the same details into the same fields for the fourth time.
None of these is a technology failure in the traditional sense. The brand simply didn’t remember you. That failure makes every “personalized” feature a guess.
This is the central tension in AI-driven CX: we’re pouring enormous energy into smarter models, faster decisioning engines, and sophisticated content generation, but the foundational layer underneath it—identity—remains neglected.
What “remembering” actually requires
For a customer to feel remembered, a system must resolve who they are across time, devices, and touchpoints, and make that information available fast enough to act on. That’s harder than it sounds, because “identity” spans several overlapping layers:
- Deterministic, 1:1 identity — a logged-in user, a known email, a loyalty ID
- Device and probabilistic identity — signals that suggest continuity without certainty
- Behavioral and intent signals — clickstream, browsing patterns, in-store interactions, media touchpoints
- Agent identity — an increasingly urgent category, as brands now need to distinguish humans, bots, and AI agents acting on a human’s behalf
Most organizations resolve one or two of these layers well, creating the personalization pain points we outlined above.
The problem is governance, not technology
The technology to solve identity resolution is no longer the limiting factor for most brands. Organizational ownership is.
Identity efforts often lack clear accountability, creating a hub-and-spoke (or worse, fully decentralized) model with no governance. That structure works for basic segmentation but collapses when a resolved identity must travel across web, app, email, media, and in-store for real omnichannel orchestration.
The fix isn’t one-size-fits-all. A B2B business with a small, high-value audience needs near-perfect identity accuracy. A high-volume ecommerce retailer needs speed and immediacy across a larger, lower-margin base. The right audience strategy, data model, and level of investment look fundamentally different in each case. The consistent principle: focus on a concentrated, well-governed subset of cohorts tied to real business value and expand only as the opportunity justifies.
The deceptively hard question: new customer or existing?
There’s a foundational question that sounds trivial and rarely is: is the person in front of you new or existing?
Some organizations define “existing” purely by login state. If you’re not signed in, you’re treated as new. That’s simple to implement, but it creates problems when loyal customers browse anonymously or on a new device.
Others use probabilistic matching to infer continuity from device and behavioral signals, treating known customers as known even before authentication. That recovers more “false new” cases, but risks treating actual new prospects as existing customers, or worse, exposing one customer’s context to someone else.
At times, an organization’s own systems define customer statuses differently. Marketing and email might use different thresholds for what makes someone “lapsed,” leading to one system suppressing win-back while the other sends it. That wastes spend and creates confusing customer experiences.
“New versus existing” isn’t a technical checkbox, it’s a governance decision about how much confidence you require before you act, and how many systems must agree before that decision is trusted downstream.
The overlooked imbalance: paid media remembers, but your own website often doesn’t
Paid media disproportionately consumes identity signals—search, display, TV, email—to improve acquisition. Owned channels, meanwhile, are often the last to use that identity signal, leading to a generic, disjointed on-site experience.
Matching media and owned messaging to the customer’s actual stage—awareness, consideration, research, decision—is where identity resolution earns its value.
Where most organizations actually are
Most personalization programs still rely on content-level A/B and split testing. More mature ones are shifting toward algorithmic decisioning. Methods like multi-armed bandits and next-best-action models adjust in near real time as identity and behavioral data update. That requires infrastructure that can resolve identity and return it fast enough to act on before the next page loads.
The organizations seeing real returns are picking a contained, high-value use case, proving the model, and expanding deliberately. Done well, identity resolution lifts account signups, visit frequency, average order value, lead quality, and customer lifetime value. It also cuts the cost of re-acquiring customers you already have. This isn't theoretical. We've helped a leading financial services brand overcome the "forgetting problem" by resolving identity on the homepage, driving a 400% increase in account signups. We're now extending those identity signals across the entire journey.
Where we’re headed: when identity belongs to the agent, not the interface
In five to ten years, we expect the most disruptive shift won’t be smarter personalization engines. It’ll be that your customers won’t always be human.
As AI agents increasingly shop on a person’s behalf, “identity” needs to recognize an agent acting with someone’s authority, verifying intent, permissions, and preferences that may never pass through a traditional web interface. Organizations that treat identity as a governed, flexible capability now, rather than a fixed set of cookies and login events, will be better positioned when the “customer” is software.
Where to start
You don’t need to solve everything at once. If you’re earlier in this journey:
- Name an owner. Identity and audience strategy need a single accountable team
- Pick one high-value, contained use case and prove the resolution-to-experience loop end-to-end before expanding scope
- Audit the media-to-owned handoff to ensure that your identity signals actually reach your on-site or in-app experience
- Let AI accelerate pattern recognition but keep a strategist in the loop on cohort definition
- Start tracking human, bot, and agent traffic separately, even informally
The distinction will only become more consequential. Personalization has never been a content problem or a technology problem at its core. It’s a memory problem. Solve for genuinely remembering your customers, and AI-driven personalization becomes dramatically easier to deliver.
