Most enterprises wear a modern facade: sleek website, fresh logos, dazzling AI demos. But beneath the veneer, the digital revolution is stalled. The foundations remain unfinished. Customer data spans three disconnected generations, access rules are unclear, and workflows rely on CSVs and heroics.
Adding agentic AI to that environment doesn’t fix the fundamentals; it amplifies existing problems and increases risk.
This isn’t a niche problem. McKinsey estimates that 78% of organizations are using AI in at least one business function. A PwC survey found that 88% of companies are boosting their AI budgets for agentic AI adoption. The footprint is vast and growing.
As a global CMO, I’m pro-speed. Yet speed without structure creates brand, security and privacy risk. Before scaling AI, we need to finish the digital revolution: clean data, modernized pipelines and disciplined identity governance.
What AI Really Accelerates
AI accelerates broken processes and amplifies errors such as duplicate SKUs, pushing noise and risks faster. If permissions are vague, agents will connect to sources they shouldn’t. None of these are model problems. They’re hygiene problems: data quality, access clarity and change discipline.
Three risks that grow with AI adoption:
• Incorrect Connections: automations reading from the wrong tenant because no one mapped the lineage.
• Scope Creep: “temporary” elevated permissions becoming permanent because there’s no expiry or owner.
• Invisible Decisions: content or offers shipped without a trail of inputs, approvals and claims.
Identity governance isn’t paperwork—it’s the engineered control of who can do what, where and how long. It’s the AI-era enterprise’s operating system. It decides not only who can act, but how consent and constraints travel across systems.
Done right, it shifts from role-based access control’s static permissions to attribute-based access control’s contextual rules: enabling scale without blind spots.
What ‘Ready’ Looks Like (A Marketer’s Version)
You don’t need a moonshot to be ready; you need a clean, predictable surface for AI to run on, including:
1. Unified Customer References: One identifier per customer, reconciled across systems. Not perfect, but predictable. If you can’t trace a record across ad, site, commerce and care, you can’t promise relevance or privacy.
2. An Access Ledger: A continuously refreshed list of people and nonhuman actors (bots, service accounts), systems they can touch and precise actions they’re allowed to take. Every entry has an owner, purpose and end date.
3. Consent And Claims That Travel: Customer preferences and legal claims attach to the data and the asset, not the app. When AI assembles a new combination (like copy + offer + audience), those constraints are enforced automatically.
4. Provenance By Default: Every asset, decision and data transformation carries its recipe: inputs, approvals and sources. When a regulator calls, you answer with facts, not Slack archaeology.
None of this slows marketing. It unblocks it because the safest path becomes the fastest one to ship.
A Different Partnership Model
This foundation requires a tighter partnership, with reimagined roles.
Marketing brings the outcomes and the brand safety standard. Engineering makes the pipelines predictable. Security designs least-privilege patterns and the off switch. Privacy ensures lawful basis and purpose limitation. The job isn’t to win turf; it’s to shorten the time from idea to safe impact.
Here’s the decision principle I use for anything material to customers, data or brand: Marketing proposes the outcome and scope. Security and privacy shape the limits. We all sign the plan, including expiries for exceptions and the evidence we’ll capture. It’s simple, fast and repeatable.
To achieve this, two rituals make a huge difference without bloating calendars:
• Pre-Flight: Before any automated workflow changes channel-facing output, confirm three things: the source list, the access scope and the rollback. If any are fuzzy, it waits.
• After-Action: When something ships or fails, capture what changed, what was learned and what to templatize for next time. Small loops equate to compounded learning.
A 12-Week Sprint To Finish The Digital Revolution
This 12-week sprint gives leadership teams a path to finish the digital transformation foundation before scaling AI:
• Weeks 1–2: Inventory reality. List systems that publish or message customers. For each: who can publish, who can approve, what data it reads and who owns the permissions. Expect surprises. Record them.
• Weeks 3–4: Fix the worst entitlements. Identify the top 10 over-privileged identities (human and machine). Reduce them to the minimum needed and set expiries. Name an owner for each. This cuts the blast radius.
• Weeks 5–6: Make consent portable. Ensure consent and preferences are accessible the same way from every channel. If a customer opts out in a service interaction, it must stick in ads and email. Build the adapter; stop pretending a privacy link equals compliance.
• Weeks 7–8: Attach provenance. Require a basic recipe for anything that publishes externally: source data, claims, approvals. Start with templates; refine later. If provenance feels heavy, you’re doing too much by hand.
• Weeks 9–10: Pilot a safe automation. Pick one workflow. Run it inside your new constraints. Measure cycle time, error rate and the share of work that didn’t need a human redo.
• Weeks 11–12: Retire a manual checkpoint. Use the gains to remove at least one manual step you no longer need, proving that governance equates to speed.
The Upside Of Finishing The Revolution
Cleaning data and clarifying access might not trend on social, but it moves numbers that matter: faster cycle times, fewer reversals, lower cost to serve and higher trust.
It also earns marketing a durable seat at the table. When the brand can show that every leap forward rides on a finished foundation, the organization stops debating if to scale AI and starts asking where to point it next.
Most companies have left their digital revolution unfinished. AI won’t finish it. It will only punish the gaps. The safe path forward is to finish the foundations now: clean, govern and label data with identity at the core. Do that, and AI runs on rails instead of luck.
