Redefining Customer Experience with AI, ETCIO
Every enterprise we sit down with wants to talk about agentic AI as an efficiency story: automate marketing, automate commerce, automate service. That’s a fair conversation, but it’s the easier half of the story. The harder question, the one most leader are still working through, is what happens when your customer stops being a person and becomes an agent shopping on their behalf. Two decades in customer experience have taught me that this pattern is different from every prior wave of transformation, and it’s the subject of this piece.
Demystifying the new experience: Agents on both sides of the transaction
That internal debate about automation misses the more basic shift underneath it. Earlier AI predicts, analyses, and generates. Agentic AI takes action. Machine learning segments the audience. Generative AI writes the campaign. Now the agent delivers it, tracks the response, and optimizes the next round on its own. Predict, generate, act. Once every function has an actor instead of just an advisor, a single all-knowing assistant stops being elegant. It becomes a bottleneck.
The same shift is happening on the customer’s side. Individual buyers increasingly let an agent handle discovery and comparison before a person ever looks at a result. B2B procurement functions are going further still, delegating the full evaluation and purchase cycle: the buyer sets intent and guardrails, and the agent compares vendors, checks terms, and completes the transaction on their behalf.
Omnichannel is quietly becoming agent-to-agent, our coordinated agents delivering experiences, meeting the customer’s agents evaluating and transacting. Agents are showing up on both sides of the transaction: the enterprises creating experiences and the customers consuming them. That is the world we now have to design for.
Fragmentation turns agents into fast, bad decision-makers
A chatbot in service. A recommendation engine in commerce. A campaign tool in marketing. None of them share what they know about the same human being.
Most enterprises still cannot resolve one customer into a single view. Identity is scattered across a CRM, a CDP (the customer data platform meant to unify behavioural and transactional data into one profile), a loyalty system, regional instances, and a spreadsheet someone won’t surrender. You cannot orchestrate an action across a journey you cannot see whole. Give an agent partial data and it makes bad decisions at machine speed, across thousands of customers at once.
Machine customers are already deciding whether you make the shortlist
Two things break at once when agents start transacting on a customer’s behalf.
First, you become invisible to machine customers. If your offer isn’t legible to an agent, clear terms, structured data, machine-readable pricing and fulfilment, it drops off the shortlist before a human ever sees it. A protocol layer is forming fast to standardize this: the Agentic Commerce Protocol, or ACP (Stripe, OpenAI), for discovery and checkout, the Universal Commerce Protocol, or UCP (Google, Shopify), for the full journey through returns and service, with the Model Context Protocol as the connective tissue between them. The new discipline is agent legibility, your offer has to be comparable to a machine, not just attractive to a person.
Second, your measurement goes dark. Sessions, bounce rate, and cost-per-click stop capturing what’s actually happening once agents transact, and legacy attribution starts at add-to-cart, several steps too late.
Trust and context become the architecture, not a feature bolted on top
The pattern enterprises are converging on now has a name: the agentic mesh, an orchestrated network of agents sharing context, tools, and data across four layers, agents that act, orchestration that routes and escalates, a shared context layer, and governance that audits and guards. The craft that makes it work is context engineering. Several CIOs we work with have reached the same conclusion independently: context, not model choice, decides enterprise AI success.
As agents gain autonomy, trust stops being a compliance checkbox and becomes the differentiator. Who owns an agent in production, and how do humans and agents share responsibility for an outcome? These are leadership questions, not technology ones. The playbook is concrete: transparent consent, granular permissions, action logs, and a verifiable agent identity that can be revoked the moment it’s compromised.
For India, this is sharper still. Orchestration multiplies inter-agent data, and under the DPDP Act, which governs consent and cross-border data movement, governed, sometimes on-premise architectures are becoming the defensible posture, not the conservative one. In banking, financial services, and insurance (BFSI), and in telecom, trust is the architecture.
One customer, many agents: The architecture that makes this work
None of this gets solved by buying more agents, and a mesh does no good without a customer record it can actually mesh around. The differentiator was never the agents themselves, anyone can spin those up.
The first pillar is the context layer, the same one the agentic mesh runs on, scaled from an internal architecture pattern into the foundation of the customer relationship itself. It means one consistent customer identity that every agent reads from and writes to, instead of five disconnected agents each guessing at a different version of the same person. Resolve one customer first, and many agents becomes an advantage instead of a source of contradiction.
The second is brand as inherited guardrails. Agents work best when they’re given goals and rules, not scripts, so voice, ethos, compliance, eligible offers, escalation triggers, and spend limits belong in the governance layer, inherited by every agent downstream. Speed without a shared rulebook just produces fast inconsistency. Human oversight isn’t a failsafe bolted on afterward. It’s a designed layer.
Retire the vanity metrics that go blind in this world and stand up agent-era measurement instead: agent-originated discovery, share of agent recommendations, outcome-based signals. That’s the fastest place to build an advantage.
The alignment gap between board, business unit, and floor decides whether agentic AI survives
Three altitudes show up in every customer conversation, and the gap between them is the real risk. At the boardroom level, the ambition is untethered: one AI assistant that owns the whole journey, agentic by next year. Business-unit leaders ask sharper questions, which use case actually moves a metric, who’s accountable when the model is wrong. The teams who will live with it feel quiet dread; they’ve been through transformations before and know the data isn’t as clean as the boardroom believes.
That distance, board, business unit, and floor, is the single biggest predictor of whether an AI initiative survives. The technology is rarely the constraint. The alignment is.
The way forward
That leaves the alignment gap. No architecture solves it. Only a way of working does.
The sequence we take clients through is deliberately unglamorous:
Start early, stay focused, prove value, then scale: move fast on a bounded, high-value problem, prove the outcome, and only then expand into the full journey. That’s not about starting small, it’s how you earn the right to scale into the transformation you’re building toward.
Get context-ready before you get agent-ready: resolve the single view of the customer and stand up the context layer first. Data readiness is the precondition, not a parallel workstream.
Build the spine, then the specialists: establish orchestration and governance, shared context, inherited guardrails, identity, and audit, before scaling agents across marketing, commerce, sales, and service.
Design the human tier and the machine-legibility layer in from day one, so oversight and discoverability are architectural decisions, not afterthoughts.
The enterprises that win this transition won’t be the ones with the boldest slide in the boardroom. They’ll be the ones that close the distance between board, business unit, and floor, then build in the right order, context first, spine second, specialists last.
The author is Srikanth Kamaraju, EVP & India Practice Leader, Bounteous.
Disclaimer: The views expressed are solely of the author and ETCIO does not necessarily subscribe to it. ETCIO shall not be responsible for any damage caused to any person/organization directly or indirectly.
- Published On Oct 3, 2026 at 08:00 AM IST
