Every Customer Operations leader I meet today is running an AI pilot. In leadership meetings, use of AI in running day-to-day operations across various functions dominates the discussions. CEOs want higher productivity and lower operating costs, and functional heads and operating leads are under tremendous pressure to implement GenAI and Agentic AI in their work to achieve this. Yet, behind closed doors, many leaders admit an uncomfortable truth – the business cases aren’t holding up.
Across telecom, hospitality and other industries, I have watched CX leaders launch impressive sounding initiatives: a GenAI chatbot, a human sounding voice bot, an “agentic” coaching tool, voice analytics with real-time agent assist; only to see most of them quietly shelved within a year, or worse, left running with almost no one able to explain what business impact they actually made.
In Terminator 2: Judgement Day, Skynet does not fail because the technology is inadequate, it fails because it is given complete authority against a narrow, literal objective with no one accountable for how that objective plays out once the system is live. The failure pattern is uncomfortably familiar to anyone who has piloted a real AI initiative in customer operations – the technology runs as designed, it is the design itself which is not thought through properly and optimized.
With 25-plus years leading customer operations as a CXO across telecom and hospitality industries and deploying AI initiatives and transforming businesses, I have observed that most AI initiatives do not deliver business results for reasons which have very little to do with the technology or the vendor. They don’t deliver because the basic groundwork is not done properly and the initiatives don’t have clearly defined ownership. I am summarizing below my views on what leaders need to consider to make AI projects successful and deliver the right business outcomes.
1. Re-engineer the journey before you AI-enable it
If there is one mistake I have seen repeatedly over the years, it is treating AI as a software implementation on top of an existing operating model. A company takes its current IVR, its current contact center workflows, its current customer journeys, and simply moves them to the AI platform they are implementing. The basic objectives and the journey or workflow hasn’t changed. The data inputs are still fragmented across CRM, billing and backend
systems. Escalation logics are still the old ones. AI simply makes a broken process run faster, rather than fixing it and delivering substantial business results.
When we built the e-Care platform for a large digital-first telecom service provider, the AI and self-service layer wasn’t created at the end – it was built-in from day one as the backbone of the inbound and outbound journey designs across all channels. The sequence of “design or re-engineer the journey first” and then AI enable it is the single biggest differentiator between initiatives that scale and ones that stay stuck as perpetual pilots.
2. Focus on delivering business outcomes, not intermediate operational metrics
Most AI initiatives in customer operations are still judged by basic operational metrics – bot containment rate, deflection rate, no. of conversations automated, etc. These numbers look good in functional decks but mean very little to the senior leadership and the PCL.
When we deployed AI-led voice analytics and Agentic-AI based agent coaching across a large contact center, we didn’t stop at calls audited and analyzed. We tied the initiative to three hard outcomes – reducing agent call volume, correcting agent errors and most importantly, driving meaningful on-call cross-sell and upsell. The last outcome is what got the contact center reclassified internally from a cost center to a revenue contributor impacting the top line of the organization. Make sure your AI business case is focusing on organizational impact – revenue generation, renewals C retention, cost optimization, etc., instead of just functional operational metrics.
3. Treat Data & Information completeness and accuracy as a pre-requisite, not an afterthought
Clear and correct data and information needs to be ready before you touch any AI modelling or analytics or train a model. A recommendation engine is only as good as the customer data feeding it, and a self-serve AI assistant is only as good as the operational APIs it can actually use to link up with the booking system, entitlement engines, property management system, etc. Skip the groundwork and you just get an impressive looking pilot which never scales to give any tangible business improvement.
In the hospitality business, before we could meaningfully lift the member holiday conversion and booking NPS, we first had to build a personalized recommendation layer: AI-enabled but integrated deep into the booking and inventory systems. Similarly, the AI support platform we built for the telecom service provider improved resolution rates substantially within months not because the model got smarter overnight, but because we kept feeding it real edge cases and feedback until it could genuinely close the loop, not just deflect the queries.
4. Put one owner in charge, not a team or a group
Last but perhaps the most important, most AI initiatives fail because nobody owns the outcome end-to-end. Data science owns the model. IT owns the platform. Operations owns the Agents. Marketing owns the customer experience. When something goes wrong, or when results are not as expected, everyone can point to someone else. This is where many programs quietly lose momentum. The initiatives I have seen work had one thing in common – a single leader accountable for the full journey, with the authority to pull data, technology and operations teams together under one umbrella.
So what does it really take?
To summarize the above into a short playbook for CXOs, CTOs and Heads of Customer Experience approaching their next AI initiative, it would be the above four points: re-engineer the journey before you AI-enable it; focus on delivering business outcomes instead of intermediate operational metrics; treat data and information completeness and accuracy as a pre-requisite; and put one owner in charge, not a team or group.
None of this is fancy – it is the same discipline that separated good transformation programs from failed ones long before “AI” entered the discussions: strategy, operating model design and execution rigor, now applied to a powerful set of new tools. The organizations that will win the next phase of customer experience won’t be the ones with the most advanced models. They will be the ones who did the unglamorous work of redesigning underlying journeys before AI-enabling them. Technology will continue to evolve. The discipline required to make it work hasn’t changed much over the years.
Authored by Anupam Srivastava
(Anupam Srivastava is a former CXO with 25+ years of leadership experience across Telecom, Hospitality, IT and Managed Services, including senior roles at Reliance Jio, Tata Teleservices, Mahindra Holidays and Price Waterhouse. He now advises organizations on Customer Operations Transformation, Enterprise AI & Digital Enablement and Revenue Growth)