Capability and understanding grow best together.

AI is now part of the platforms many of us use every day. Assistants answer questions inside analytics workspaces. Agents draft journeys, suggest audiences, and surface changes in performance. Customers meet AI in service conversations and personalized content.

With each new capability, I ask a few questions. How was this created? What data is it drawing on? When should I rely on it, and when should I check it? Where does my judgment fit? In my view, those questions are where enablement begins.

Those of us who design data and CX programs tend to think in layers: collection, identity, profiles, activation, measurement. My perspective is that enablement deserves a layer of its own, because so much of how AI behaves comes down to what people define, configure, and review.

Confidence is built

In Microsoft and LinkedIn’s 2024 Work Trend Index, AI power users were 35% more likely to get role-specific AI training. In BCG’s AI at Work 2025 survey, 79% of people with more than five hours of training were regular AI users, compared with 67% of those with less.

In CX specifically, Adobe’s 2026 AI and Digital Trends report, a survey of 3,000 executives and practitioners, found that 45% of organizations have sufficient AI training programs and 44% believe employees are comfortable using AI in their roles. 65% have responsible-use guidelines for generative AI; 37% have them for agentic AI.

These surveys show association, not cause. I read them as a map of where the industry is building now.

What would change if your team’s questions about AI became the starting point for enablement?

AI works from what we define

AI increasingly works from the structures practitioners design. In the Adobe stack, Agent Orchestrator coordinates Audience Agent, Journey Agent in Journey Optimizer, Data Insights Agent in CJA, and Experimentation Agent. Adobe describes them as drawing on Experience Platform data and customer metadata, with plans visible so users can step in.

That makes configuration work a form of AI context:

• Foundation. XDM field descriptions, identity namespaces, and merge policies shape how an agent understands a customer.
• Audiences. Clear naming, descriptions, and a clean inventory shape what Audience Agent surfaces.
• Journeys. Standardized entry criteria, content, and conditions become the patterns Journey Agent works within.
• Analysis. Data view curation, from component descriptions to hidden components, shapes what Data Insights Agent returns.

If an agent read your implementation today, what would your schemas, audiences, and data views teach it?

Governance travels with the data. Data usage labels, marketing actions, and consent policies evaluate together at activation, and attribute-based access control extends labels to fields and audiences. As agents take on more audience and journey work, in my experience that framework depends on people understanding what each label protects, and what an agent-built audience inherits.

I’ve made a version of this point before, about taxonomies and about person-to-account mapping. Structure gives a new capability somewhere to go. Agents are no different.

Behind the counter at Summit

In my Adobe Summit 2026 session, Agentic AI in CJA (S905), I used a coffee shop analogy. AI Assistant is the barista, interpreting what you ask for. Data Insights Agent works behind the counter, preparing the analysis from what you’ve already configured. The drink depends on what’s stocked: your schema, data view, metrics, and dimensions. When an answer surprises you, refine the prompt or revisit the data structure. Context is everything.

These tools don’t change your data or apply attribution or forecasting for you. They build the starting point, so you can switch to algorithmic attribution, project a trend with a confidence range, or add intelligent captions. The judgment stays with you. As I put it on stage:

It’s not removing the analyst. It’s removing the friction.

Responsible use has two audiences

In CX, responsible AI reaches the customer too. Adobe’s 2026 report also surveyed 4,000 customers. About one in five say they can reliably detect AI in their interactions, and 45% are comfortable with an AI agent working directly with a brand’s human representative. Organizations ranked clear AI disclosure (68%) and easy escalation to a human (61%) among the top trust factors.

Both depend on people. Disclosure is clear when the team knows where AI shows up. Escalation works when the person receiving it knows what the agent already did. From my perspective, customer confidence in AI is built, in part, by the confidence of the team behind it.

Where does AI show up in your customer’s experience, and who on your team can explain how it got there?

Learning in the flow of the work

This part is personal. Traditional math didn’t click for me in school, and today I work in analytics. The numbers didn’t change. The context did. Once numbers told human stories, like a customer finding what they needed, the work clicked.

It wasn’t numbers that unlocked my brain. It was meaning.

Much of that learning happened inside the platforms, on my own time. I’ve written that about 70% of my analytics knowledge was self-taught. AI can extend that scaffolding. In my Summit demos, each AI Assistant answer opened in Analysis Workspace with one click, so someone newer can see which metrics and dimensions answered the question and keep building. An answer you can inspect becomes a lesson as well as a result.

A framework: the Virtuous Cycle

I frame data work as a Virtuous Cycle: Learn → Think → Make → Measure. It applies to new AI capabilities too.

• Learn. Know what the capability draws on: schemas, identities, audiences, data views, labels, and consent. The AI layer amplifies whatever is underneath it.
• Think. Decide where human judgment sits and what customers will be told.
• Make. Curate the context: write descriptions, retire duplicates, confirm labels and roles. Pilot with a defined group.
• Measure. Track adoption, confidence, quality, and customer impact. In a past role, I ran Target and Journey Optimizer with a universal person-level holdout for a consistent cross-channel baseline, a design that suits AI-assisted personalization. As I said at Summit: document where you are, then document where you go.

Where in the cycle will your next AI capability benefit from enablement first?

What to measure

Recommended metrics by stage, with published reference points where they exist:

Metric	Cycle stage	Reference point
AI training and upskilling coverage	Learn	45% report sufficient programs (Adobe, 2026)
Practitioner comfort with AI in their role	Learn	44% believe employees are comfortable (Adobe, 2026)
Responsible-use guidelines	Think	65% for generative AI; 37% for agentic AI (Adobe, 2026)
Descriptions on fields, audiences, and components	Make	Set your own baseline
Regular use, by role	Make	79% with more than five hours of training vs. 67% with less (BCG, 2025)
Human review rate on AI-assisted outputs	Make	Set your own baseline
Time to first insight	Measure	My S905 demos: under two minutes vs. an estimated 2–5 minutes built manually (demo data; varies by skill)
AI measurement framework in place	Measure	44% for generative AI; 31% for agentic AI (Adobe, 2026)

Benchmarks give context. Your own baseline shows how enablement is landing.

Where to start

1. Who curates your AI’s context? The owners of schemas, identities, audiences, data views, and labels, or their equivalents in your stack, are your first enablement audience.
2. What will customers be told? Define where AI disclosure and human escalation live, and make sure those teams understand how the agent reaches its recommendations.
3. What’s your baseline? Measure two metrics from the table, plus a holdout for AI-driven experiences, before your next launch.

AI capabilities will keep evolving. Enablement designed into the stack gives teams, and their customers, a steady place to stand.

Sources: Adobe, “2026 AI and Digital Trends Report,” February 2026 (3,000 CX executives and practitioners; 4,000 customers). Microsoft and LinkedIn, 2024 Work Trend Index, May 2024. BCG, “AI at Work 2025,” June 2025 (10,600+ workers). Katie Klein, S905, Adobe Summit 2026. Adobe Experience League: Agent Orchestrator; automatic policy enforcement; attribute-based access control.

What would your CX architecture look like with enablement drawn in as a layer of its own?

---

If you have questions, or just want to talk analytics, data strategy, and anything in-between, reach out to me at Katie@DataOnTrend.com.