The Work After the Model: Kuntal Malia on Turning Intelligence into Business Consequence

When Capability Has to Prove Its Business Value

Technology earns its place in business when it changes a decision, deepens customer understanding, improves a workflow or creates value the organisation could rarely deliver before. In consumer businesses, the test becomes sharper because customers respond to relevance, trust, convenience, confidence and fit before they respond to technical sophistication. For leaders working with data and intelligence, the real test begins after capability has been demonstrated: can the organisation turn it into better judgment, better customer experience and better operating discipline?

Kuntal Malia’s work gives the question unusual range. Her career has moved across large-scale digital platforms, consumer analytics, fashion-tech, entrepreneurship and physical retail, with experience across Yahoo, Shutterfly, ModCloth, StyleNook and now Metro Brands Limited. The value of the arc is strategic. She has seen intelligence systems from several demanding business positions: the modeller’s search for accuracy, the operator’s need for action, the founder’s struggle for customer relevance and the enterprise leader’s challenge of adoption at scale.

Her current role brings those questions into a larger institutional setting. Building a system is now the easier part of the mandate. The deeper test is whether the business has the data, workflow, trust, ownership and decision discipline required for intelligence to create value in daily operations.

Decision Latency and the Business Clock

An early moment at Shutterfly gave Kuntal one of the most durable business lessons of her career. She was working on a model and wanted more time before sharing the result. A business leader was waiting because a decision had to be made. She refined the work, improved the score and returned with stronger analytical output. The leader then asked what those extra days had actually delivered.

“Do you really think it was worth for me to wait the extra two weeks?”

The R-squared had improved by a few points. The business had lost time.

Call it decision latency: the cost a business pays when useful intelligence arrives after the decision window has started closing. Technical accuracy matters; the discipline is to know when additional refinement is still reducing business risk and when it has begun to cost the business time. The judgment call is rarely the same in every situation. It depends on the decision’s reversibility, downside risk, urgency and the level of confidence the business needs before acting.

“The business is often operating with zero information. If your information is at 30 or 45 percent, that also is useful for them to know.”

Kuntal’s point is strategic. The judgment call is not between accuracy and inaccuracy. It is recognising when further refinement improves the answer and when it mainly delays the decision. A partial answer, clearly caveated and delivered at the right moment, can improve judgment before uncertainty becomes paralysis.

Don’t wait until it becomes perfect.

Behind the sentence sits a demanding business discipline. Strong leaders know when evidence is sufficient, when uncertainty has been explained well enough and when delay has become the more expensive choice.

Where the Machine Earns Its Place

StyleNook gave Kuntal a founder’s view of the boundary between automation and human judgment. Online fashion already offered abundance. Customers could browse thousands of products and still struggle to find clothing suited to their body, preference, occasion and sense of self. The harder problem was relevance.

Kuntal and her team studied how stylists made decisions and found repeatable elements that could be systematised: explicit customer preferences, sizing and body shape. Algorithms could bring speed, consistency and affordability to those parts of the experience. Other parts required judgment. Fashion involves discovery, taste, inspiration and discretion. A customer may know the occasion while still being unsure about the right form, style or leap beyond her usual choices.

The business wisdom from StyleNook travels beyond fashion. Automation should be treated as a placement decision. Machine consistency is valuable where the decision is repeatable, measurable and scalable. Human judgment is valuable where the need is partly unstated, emotional, contextual or exploratory.

For leaders, the strategic task is to map judgment before mapping automation. Full automation can flatten value in experiences where interpretation matters. Heavy human review can slow a system without improving the quality of the decision. The sharper question is where each form of intelligence earns its place.

The Friction Has Moved Downstream

Kuntal’s early career began when the model appeared to be the centre of difficulty. At Yahoo, she worked on models for one of the largest online properties of its time, focused on questions that still define digital business: how to attract users, retain them and monetise better. Shutterfly deepened that analytical discipline, while ModCloth brought her closer to fashion, community and consumer behaviour.

“The effort that goes towards the AI model is maybe 10 to 15 percent. The data might be somewhere around 35 percent. The biggest piece is people, process, deployment and change management.”

Model performance matters. Value now depends on usable data, trusted workflows, clear ownership, stakeholder confidence and the ability of teams to absorb a new way of working. A technically strong model can underperform when deployment fails to match design, when data shifts after launch, when users avoid the system, or when the workflow was never ready to carry the change.

AI leadership increasingly includes operating redesign. Resistance from teams can reveal discomfort with change, valid friction in the workflow, or weakness in how the system has been designed for frontline use. Capability creates an opening. Operating discipline decides whether it becomes value.

Reading the Store

Kuntal’s move into enterprise retail created one of the strongest shifts in her thinking. Her earlier experience was largely digital, where customer behaviour leaves signals before the transaction. Digital businesses can study browsing paths, session time, comparison behaviour, abandoned carts and checkout friction. They can see where intent forms, weakens and breaks.

Physical retail often begins with a thinner structured record.

“In physical retail, when it comes to the data, the only thing that you have is the final transaction.”

A store is full of behaviour. Customers enter, pause, touch products, try items, ask questions, compare options, speak to companions and leave. The purchase records the ending. Hesitation, confusion, interest and near-purchase often disappear from the system.

For physical retail more broadly, the opportunity is to recover parts of the customer journey that have historically gone unrecorded. CCTV analytics, RFID signals, conversation insights, transaction data and timestamps can help retailers understand which products were tried but left behind, where customers asked questions, what created friction and how the physical journey shaped the final decision.

Reading the store is a different discipline from copying digital commerce. Digital produces cleaner behavioural trails. Physical retail contains signals that digital cannot fully replicate: touch, movement, social discussion, staff interaction and the presence of the product in a real environment. A store can become an intelligence system when those signals are captured with care and interpreted with purpose.

For retailers, the implication is strategic. Physical retail should avoid accepting permanent data inferiority against digital. The customer journey inside a store has always existed. The tools to observe and connect it at scale are becoming more accessible. Once movement, trial, staff interaction, product handling and final purchase are studied together, the store starts offering a fuller view of intent than the transaction alone could provide.

A richer view of the store changes more than reporting. Merchandising decisions become sharper when retailers know which products attract attention but fail to convert. Service improves when customer questions reveal confusion before a sale is lost. Store design becomes more disciplined when movement patterns show where attention gathers and where it drops. Inventory decisions gain another layer when trial data is connected with purchase data.

Reading the store also requires restraint. Behavioural intelligence becomes valuable when customers are treated with respect and the purpose of data collection remains tied to better experience, better service and better business decisions. Digital commerce taught businesses to study clicks. Physical retail may now teach them to study intent in motion.

The Pilot Without a Decision

Many organisations launch pilots with energy, speed and ambition before clarifying the decision the pilot is meant to inform. The demonstration may work. The result may look impressive. The leadership team may remain interested. Yet the organisation may still lack agreement on what improvement would matter, who owns the outcome, what decision will follow and what level of process change the business is prepared to make.

Your pilot ideally is not going to represent what will be at the real-world stage. It is meant to help you understand what capability you have and whether that achievement is in line with what you thought it would do.

A pilot becomes useful when the business has already defined what it wants to learn. A clear decision trigger gives experimentation institutional force. What must the pilot prove? Which business owner cares about the outcome? What level of improvement is enough to scale? What will the organisation stop doing if the pilot succeeds? Which process will change? Which metric will move?

Experimentation becomes a holding pattern when those answers remain unresolved.

Where Deployment Meets Institutional Resistance

The operational test begins before a model goes live, when the organisation decides who must trust it, who must use it, who will be affected by it and what work will change because of it.

Kuntal’s view of deployment is grounded in that reality. Models can be evaluated technically through checks, test sets, ongoing metrics and monitoring for data drift or concept drift. Those controls matter. They sit alongside the harder human and operating questions.

A model forced into a business can create resistance that looks like politics but may contain valid operating intelligence. People close to the work often see weaknesses that design teams miss. They may understand an exception pattern, a customer behaviour, a store-level constraint or a practical risk that the model has failed to absorb.

The leadership task is to diagnose resistance with precision. Resistance can reflect fear of change. It can reveal poor workflow fit. It can signal late stakeholder involvement. It can also show deployment design has failed to respect how the work actually happens.

Enterprise transformation becomes serious at this point. A strong model, a thoughtful roadmap and a credible use case can still fail if the organisation hesitates to change the workflow that gives the model its value. For Kuntal, people, process and deployment are where value is either created or lost.

Autonomy Requires Decision-Level Governance

The question of how much a system should be allowed to do on its own needs decision-level judgment. The right answer depends on the process, the outcome, the reversibility of error, the stage of maturity and the cost of a wrong decision.

Kuntal frames the issue through the difference between augmented and autonomous intelligence. In early stages, systems often need human intervention or review. As confidence improves and risks are better understood, some decisions can become more automated. The calibration has to be made decision by decision.

The credit card fraud example makes the trade-off clear. A system may block a genuine transaction because the risk of allowing a fraudulent one is considered higher than the cost of inconvenience. The customer may be frustrated, but the company has chosen autonomy because the downside risk justifies it.

A recommendation engine, a pricing decision, a hiring screen, a store-level operational suggestion and a fraud alert carry different levels of reversibility and human consequence. Governance becomes serious when leaders move from broad automation debates to more precise questions: where is autonomy justified, where is augmentation safer, and where does human judgment remain central?

Growth, Relevance and Distribution

Many AI programmes begin with efficiency: automate repetitive work, improve response time, reduce manual effort and lower cost. Kuntal recognises the value of efficiency, then pushes the conversation toward the larger growth question: what new capability, service or customer experience can the organisation now create?

Cost reduction is often the easiest use case to justify. Growth requires more imagination from leadership. A company can use automation to make an existing process cheaper, or it can use released capacity to create a new customer-facing capability.

Kuntal points to IKEA as an example of the second approach. Once automation absorbed parts of first-level customer support, the larger opportunity was to redeploy people into higher-value advisory work. In her telling, upskilling customer-service teams into design advisory roles allowed the company to support customers in more meaningful ways. Released capacity became a route to growth.

The business lesson becomes sharper through that lens. AI creates strategic value when leaders ask what the organisation can now do better, faster or differently because routine work has been absorbed. Efficiency may justify the first investment. Growth determines whether the investment changes the business.

Relevance is the filter Kuntal applies to AI enthusiasm. A tool can be impressive and still carry little consequence for a particular business. She gives a sharp example from the current generative AI wave: tools that produce presentations quickly.

“If you’re not in consulting, I don’t see how it actually impacts organisational productivity.”

Her point extends well beyond presentation tools. In a context such as consulting, where the deck is often the deliverable, faster presentation creation can matter directly. Outside that context, the capability may not meaningfully move the needle on organisational productivity if the company’s real performance levers sit elsewhere.

The same discipline applies to founders. The barrier to building has fallen. Applications, websites, tools and prototypes can be created faster than before. When building becomes easier for everyone, the scarce skill moves elsewhere.

As the barrier to build becomes lower, the need to figure out distribution becomes higher.

For founders and established consumer businesses, technical fluency has to be matched by market understanding. Will people use the product? Will they return? Is there a large enough market? Can the product hold attention, build trust and grow beyond early curiosity?

When everyone can build, advantage moves toward problem selection, distribution, trust and the ability to keep learning from the market.

Strategic Takeaways from Kuntal Malia’s Work

• Decision latency can be as costly as weak analysis. Useful intelligence creates value when it reaches the business while action is still possible.

• Automation earns its place through fit. Machine consistency works best where decisions are repeatable, measurable and scalable; human judgment matters where context, taste, emotion and ambiguity shape value.

• Retail stores can become behavioural systems. Physical retail gains strategic depth when it reads movement, trial, conversation, service interaction and intent with responsibility.

• Pilots need decision triggers. Experimentation without ownership, scale criteria and workflow implications creates motion without commitment.

• Deployment is an operating challenge before it is a technical milestone. Stakeholder trust, workflow fit and frontline intelligence must be built into the programme early.

• Autonomy is a decision-level calibration. Risk, reversibility, maturity and consequence determine where systems act alone and where human judgment holds.

• Growth requires more imagination than efficiency. The larger opportunity appears when leaders ask what new service, capability or customer experience becomes possible after routine work has been absorbed.

• Distribution becomes more valuable as building becomes easier. Future advantage will reward leaders who understand how to build, reach, retain and keep learning from the market.

The Work After the Model

Kuntal’s work points to a more mature view of business transformation. The model matters. The outcome depends on the system around it: the decision it improves, the data it requires, the workflow it enters, the people who must use it, the customer behaviour it helps interpret and the market it eventually has to reach.

Decision latency applies beyond one Shutterfly model. A company can keep refining the model, expanding the pilot, improving the deck and discussing the opportunity while the real questions remain untouched. Which decision improves? What customer is understood better? Which work is redesigned? What value is the business prepared to create?

Kuntal Malia’s career suggests that the next advantage will belong to leaders who can decide when evidence is enough, where judgment still matters, how the store should be read, when experimentation has become avoidance and why distribution may matter more as building becomes easier.

The work after the model is where capability becomes consequence.