From Quick Service to High-Consideration Retail, How Pranshu Rastogi Matches AI to the Way People Buy

Pizza and mattresses

A pizza order and a mattress purchase make almost opposite demands on a customer service function. One is measured in minutes and repeats every few weeks. The other happens once in several years and involves a decision the customer cannot easily undo.

Rastogi spent more than eight years running customer experience at Jubilant FoodWorks, across Domino’s Pizza, Dunkin’ Donuts, Hong’s Kitchen, Ek Dum and Popeyes in India, Sri Lanka and Bangladesh. In April he became chief experience officer at The Sleep Company. What travels between the two jobs is a test he applies before any deployment.

“The biggest thing I am carrying forward is the belief that AI should solve customer friction, not become a technology project. Whether someone is ordering a pizza or buying a mattress, customers are trying to achieve an outcome with the least effort possible.”

The buying journey is where the two part company. Quick service generates a high volume of requests that look alike.

“In QSR, most customer interactions are transactional. Customers want speed, status updates, refunds, order modifications, or issue resolution. The majority of these use cases are repetitive and highly automatable.”

Mattresses generate fewer requests, each carrying more weight.

“A mattress purchase is very different. It is a high-consideration category with a longer decision cycle. Customers have questions about comfort, body type suitability, materials, warranty, financing, and return policies. Here, AI cannot simply act as a transaction engine. It has to act more like a knowledgeable advisor.”

Which changes what a good outcome looks like.

“So what I am leaving behind is the assumption that higher automation always means better CX. In a category like mattresses, I would rather have AI help customers make better decisions and assist agents with richer context than aggressively deflect conversations.”

Deflection is the standard scorecard of a support function running on automation. It counts the conversations that never reached a person, and it rewards keeping that number high. For a customer who is still deciding, the count stops describing anything useful.

When fewer calls is not good news

Rastogi has a 40 percent reduction in voice queries on his record, alongside an annual saving of 1.8 million dollars. It is the kind of figure that enters a board deck without much scrutiny. It can also mean two entirely different things, and he is direct about that.

“A reduction in contact volume by itself is not evidence of success.”

The alternative reading is that customers tried the automated channel, got nowhere and stopped trying. Separating the two takes more than one number.

“We looked at multiple indicators together. First-contact resolution, repeat contact rates, customer satisfaction, escalation rates, transfer rates, abandonment patterns, and post-interaction feedback all helped us understand what was actually happening.”

The failure mode has a recognizable shape.

“If voice calls decline but repeat contacts increase, you have not solved the problem. You have merely moved it somewhere else.”

So the team went looking for displacement instead of waiting for it to surface in a quarterly review.

“We specifically tracked whether customers who used automated channels returned within a defined time window for the same issue. We also monitored bot abandonment versus successful resolution journeys.”

“The most useful metric was repeat contact reduction. If customers stopped contacting us because their issue was genuinely resolved, repeat interactions dropped significantly. That gave us confidence that the reduction in voice volume was healthy rather than hidden dissatisfaction.”

Repeat contact works as a check because it resists gaming. A customer whose problem was handled has no reason to come back. A customer who gave up on the bot and found another route reappears in the data within days, on some other channel, and the volume drop stops looking like a saving.

The credit problem

His public record also carries a customer satisfaction improvement from 3.2 to 3.8. Gains of that kind rarely arrive alone. New technology tends to land alongside new processes, retrained staff and a leadership team paying closer attention than usual. Assigning the result to any one of those is a problem most CX functions handle by not handling it.

“I generally avoid giving AI sole credit for any CX improvement.”

“Most successful transformations come from a combination of process redesign, better training, improved governance, operational discipline, and technology.”

“What AI often does is amplify the effectiveness of a good operating model.”

Amplify carries an implication worth sitting with. A system that magnifies whatever it is placed on top of will magnify a weak process as readily as a strong one.

Isolating the contribution came down to comparison.

“We tried to isolate AI impact by comparing similar customer journeys before and after deployment, monitoring adoption patterns, and measuring outcomes where AI was directly involved.”

It also produced a result the team had not expected.

“There were also situations where the data challenged our assumptions. In one case, a new AI capability improved response speed dramatically but had almost no impact on customer satisfaction. When we investigated, we found that customers cared more about resolution quality than response speed.”

Response time is among the most heavily instrumented metrics in any contact center, partly because it is easy to measure and easy to move. Here it moved, and customer satisfaction barely followed.

“That was a reminder that customers judge experiences differently than operators do.”

What the agents were telling him

Rastogi has run AI-assisted operations across more than 600 people. At that headcount, whether a tool works is visible in the behavior of the floor well before it shows up in a dashboard.

“Agents are usually the first people to tell you whether your AI is actually useful.”

“Whenever agents ignore recommendations, bypass workflows, or create workarounds, I treat that as valuable feedback rather than resistance.”

That reading is less common than it sounds. Workarounds usually get escalated as a change management problem, and the fix gets aimed at the agent.

“We observed this particularly in recommendation engines and knowledge-assist tools. Agents would sometimes disregard AI suggestions because they knew from experience that the recommendation was incomplete or lacked context.”

Experienced agents were overriding the system because they held context it lacked, which makes the override a form of quality control.

“The issue was rarely technology adoption. The issue was trust.”

“What we learned was that AI needs explainability. Agents must understand why the recommendation is being made. They also need confidence that the system understands the customer’s actual context.”

“The most successful implementations were those where AI acted as a co-pilot rather than a supervisor.”

Five minutes with a vendor

A large share of what is sold as AI for customer experience is automation that predates the label. Rastogi has one question for filtering it.

“Show me a customer problem that became materially better because of your solution, and tell me what metric moved.”

The answers sort themselves quickly.

“Serious vendors talk about outcomes, implementation realities, adoption challenges, and lessons learned.”

“Trend-following vendors usually spend most of the meeting discussing models, architectures, and buzzwords.”

The question does a second job. Answering it requires the vendor to have been present for a deployment that reached production and stayed there, which is a different experience from having built a demo.

“I am less interested in how sophisticated the AI is and more interested in whether it improved customer effort, resolution quality, agent productivity, or operational efficiency.”

From efficiency to decision support

Asked what he believes now that he did not believe three years ago, Rastogi describes a change in what he thinks the technology is for.

“Three years ago, I viewed AI primarily as an efficiency tool.”

“Today, I see it as a decision-support tool.”

“The biggest shift is not that AI can answer customer questions. It is that AI can help organizations understand customers at a scale that humans never could.”

Scale is the operative claim. A support function generates a continuous record of what customers struggle with, and most of it has historically gone unread beyond the individual ticket.

“The organizations that win will not necessarily be the ones with the most automation. They will be the ones that use AI to learn faster from customer behavior and continuously improve experiences.”

His closing view places AI alongside the infrastructure decisions that came before it.

“It is becoming a foundational capability, much like CRM systems or analytics platforms became in previous eras.”

“The leaders who succeed will not be those who automate the most. They will be those who apply AI with the most discipline.”

“Start with a real customer problem. Measure outcomes relentlessly. Keep humans accountable. And remember that customer trust takes years to build and minutes to lose.”