The customer funnel has collapsed:

Author

Kapil Ahuja is a Partner Director and CTO for Digital Experience at Nagarro. He has more than 20 years of experience in architecture, engineering, and product leadership across digital platforms, AI-led architectures, and large-scale enterprise systems.

There comes a time in the customer journey  when they cease to browse product categories – the old way of interacting – and  use  a conversational interface instead  to simply say  what they want to achieve. For instance, a young customer who is in his first job wants to know what  their money can do for the future.  They  are unlikely to  click through  loans, insurance, deposits  and allied categories to approach financial planning. Instead, using a conversational interface, they will simply type, “I want to buy a house in five years”. The AI engine then recommends the right combination of products.

Many industries (like travel, shopping, and entertainment) have already shifted from product-centric experiences to goal-centric ones. However, many enterprises within these industries continue  to measure success using  old metrics such as the number of loan applications, insurance page  visits, or, for that matter, which product category got the most clicks. 

For decades, CMOs and their digital  marketing teams have  created a separate customer acquisition funnel for each product. Each one with its own conversion metric, optimized  in isolation by a team that  owned  it. And  adding to the complexity, these categories  fragment  into many  variants, again, each with its own consideration path, eligibility  rules  and application form.
The architecture rests on an unwritten assumption that  the  customer knows what they want – which specific product or product categories - but that’s rarely the case. The customer arrives  with an intent, not an order form, so that assumption  no longer holds.

Yet another instance. A traveller tells an airline to get his family to Bali in less than four thousand dollars, business class one way, because his mother can’t sit in economy class for twelve hours. The  challenges  to do with fares, seat inventory, and loyalty redemption, will have to  be  solved together as one, and not as fragments. In many instances, the intent falls straight through the cracks between the verticals, though  each of which was  perfectly optimized on its own.

You  would  have already seen it. Increasingly, customers begin their journey inside an AI assistant rather than  the enterprise’s website, researching and narrowing their options  first  through tools like ChatGPT and Perplexity. Adobe tracked AI-referred traffic to U.S. retail sites growing 138% year over year in May 2026 alone (Adobe Analytics, via Digital Commerce 360, June 2026). Parts of the top-of-funnel discovery are moving off the channels enterprises own. The classical mechanics of cross-sell and up-sell assumed  enterprises owned the moment of discovery. In reality, now, they are starting to lose it.

So, the funnels have not so much been disrupted as collapsed into each other, and the org chart that mirrored them no longer matches the customer. 

Two officers, one shift
The collapse lands on two desks, and it lands differently on each.

The CMO loses the funnel, and here’s how. Every campaign, every attribution model, every cross-sell trigger was instrumented around a single-product journey with a measurable conversion point. When the customer arrives with a cross-product goal stated in natural language, there is no funnel to attribute against and no clear single point of reference to convert. While the playbook still runs, it just stops describing reality. These systems, while they continue to operate, no longer reflect how customers actually behave.

The CIO inherits a different problem: two stacks and measurable RoI for only one of them, the old system. They keep the old journey alive because it is not archaic yet, and also stand up a conversational stack alongside because the market demands one. While running both systems, the new one cannot yet show its return. And in addition, there’s another challenge. The demo works beautifully but breaks down in a production environment because a language model hallucinates, and the underlying architecture was never built for the load or the edge cases.

MIT’s Project NANDA found that around 95% of enterprise generative AI pilots had no measurable profit-and-loss impact (MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, via Fortune). In 2024, Gartner projected that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025 (Gartner press release, July 2024). Consumer expectations are now outpacing the measurement infrastructure beneath them.

These aren’t exaggerations. CXOs  we talk to believe in  the  technology, but they aren’t  as clear about the use cases. It takes significant effort and cost to run an agentic solution, for which they are unable to show revenue attribution until they are already invested. This is not how the world built businesses, at least not until now.

The three floors of IX

Call the whole stack IX, short for Intelligent Experiences. It is the umbrella for all of these customer-facing systems, whatever industry they run in.

Most of the industry responds to all of this by building a chatbot and calling it transformation. That is the ground floor. It helps to see the whole building.

Think of IX the way Hamleys in London thinks about its own floors. On the ground floor, you walk in and browse. A few floors up, kids build lightsabers and get their faces painted, while a drone show runs in the background. Every floor is a different experience, and not only about different toys. IX works the same way: each floor offers a different experience for the customer. Agentic is not one of the floors. It is the staff walking the floor and doing things for you instead of pointing at a shelf, and it can show up anywhere required.

Agentic walks into any of these three floors. On the ground floor, it just books the one thing you asked for. On the top floor, it is the one executing a plan that spans across a car loan and a hospital referral, or untangles a family’s flights when one leg cancels, with a human still in control of every step.

Most enterprises are still standing on the ground floor with agentic bolted on as a faster way to do the same narrow thing. The gap between that and the top floor is the real CX problem, and a better chatbot does not close it.

The new rules of design

Climbing the floors changes what  good  design even means. Three rules replace the old ones. Enterprises should  design  for how the customer thinks, not  how  their products are filed. The interaction is in natural language now, which means the system has to read intent, hold it, and ease cognitive load when the goal is complex, rather than handing the customer a menu and wishing them luck.

It has to synthesize intent across diverse modes of communication. A customer will speak, then type, then point at a chart, then sketch. The system has to assemble one coherent intent from all of it, not treat each input as a fresh start. Design for trust, which sometimes means intentionally designing in friction. Verification steps, transparency about what the agent is doing, and clear handoffs to a human  at the moments that carry weight. In lending, in surgery scheduling, in anything where a mistake costs real money or real trust, or a life, the right amount of friction is a feature, not a defect.

The Experience Graph: surfaces are nodes, intent flows across edges, and cognitive load determines the weight.

Great experiences are not built by reducing screens, but by finding the shortest path from intent to outcome and relentlessly removing weight from every node along the way.

CCL: the three loads under every node

That weight on a node is not one number. It is the accumulation of three loads on the customer, and so we call it Cumulative Cognitive Load (CCL).

Cognitive load is how hard they have to think. Creative load is how much they are being asked to compose or decide. Logistical load is how much doing and co-ordination the task demands.

A single node can be light on one and crushing on another. For instance, a loan comparison asks little coordination and a lot of thinking; a document upload asks the reverse. Reading which of the three is rising, at the moment, is what tells the system how to reshape the surface: drop a step, generate the summary, hand off to a human, instead of reacting to one blurred number. Track which load is climbing, and how fast. Speed is the trigger - a load rising quickly means reshape the surface now, before the customer feels it.

The build has to match the experience

The orchestrated floor holds a goal that changes shape as life changes: a car loan today, a wedding-timed hospital plan next year, a canceled flight next week, etc. Enterprises cannot freeze that into a spec written six months before the customer showed up with the goal. The software behind these floors has to be built the same way it serves the customer - from intent, not from a document written before anyone knew what the goal would be. Call it what it is: “Intent-Driven Software Development”. A spec-first team is always one document behind the customer’s actual goal.

The direction runs both ways. Every floor above teaches the enterprise something about building intent-first: the friction that stalls a customer’s cross-product goal, knowledge trapped in people, decision logic that doesn’t travel, data that doesn’t connect, is the same friction that stalls an engineering team trying to build from intent instead of spec. Build these floors well enough, and you have not just built a better CX, you have field-tested the discipline your delivery teams need.

Intent-first software isn’t proven yet, not at this level of stakes. A recommendation engine failing quietly is one problem. However, a system sequencing someone’s retirement plan, or rebooking a mother mid-flight, is a different experience entirely. We are asking teams to trade a broken discipline, spec-first, for one still finding its footing. We believe it is the right trade. Of course, we do not have three years of failure data yet to prove it.

The barrier is organizational friction. Knowledge is tacit in individuals and does not scale. Decision logic is inconsistent from one process to the next, so automation has nothing stable to stand on. Data is fragmented across systems with broken connections between them. Intent ignores all of it. The customer’s goal crosses every line-of-business wall, and those walls simply do not exist inside a conversation. When intelligence cannot flow across teams, systems, and decisions, AI investment does not yield a return. That is why the pilots stall. Clear that friction and the models keep rising to meet the floors you have opened for them. Leave it, and no model, however good it gets, saves the pilot.

This is a phase transition

The temptation is to treat all of this as a feature upgrade: add a chat box, ship a few agentic widgets, report progress. That misreads what has happened. This is a phase transition of how customers reach anything that used to live behind its own funnel.

The enterprises that win the next decade will not be the ones with the most polished chatbot. They will be the ones who deliberately climb the floors, design with intent rather than for their own org chart, and remove the friction between their people, processes, and systems so that intelligence can actually flow. The funnels are already gone. The only question left is whether engineering teams will rebuild around the customer’s intent or keep optimizing the rubble.

The spec sitting in your backlog right now- how many months old is it? That is how far behind your customer you already are.