Sharad Khandelwal and the complaint nobody read

A company had a theory about its website. Customers, it believed, were having a miserable time there. Sharad Khandelwal later described what happened when the feedback was examined: people liked the site’s simplicity and ease of use. The trouble was elsewhere. It is a pleasingly awkward result. A business can spend a considerable sum solving a problem its customers never asked it to solve.

Khandelwal has made that sort of disagreement his working material. As co-founder and CEO of SentiSum, he deals in the things people write when a delivery disappoints them, a payment fails or an answer leaves them with another question. The sentences are often untidy. They rarely arrive wearing the vocabulary of a strategy presentation. They are useful precisely because the customer has something specific to say.

His proposition begins with a change of attention. The organisation already has conversations with customers. Someone has already paid to receive them, answer them and store them. What happens to the explanation after the ticket is closed? That question has carried him from building an analytics product to asking who, inside a business, has the responsibility to do something about its findings.

From trading systems to untidy sentences

The career before SentiSum was a technical one. Khandelwal worked as a software engineer at Infosys from 2004 to 2006, then as a senior technology associate at Sapient. His next stop was University College London, where he studied Software Systems Engineering in 2009-2010. He has also described his studies there in terms of machine learning.

He went into equities algorithmic trading at Merrill Lynch, followed by an associate director role at RBC Capital Markets from 2011 to 2015. Finance gave him experience inside large organisations. Software gave him a way to build. In his account of starting SentiSum, he connects those two strands: the ability to make a first product, and familiarity with companies that accumulate data while leaving it separated across departments.

The transition appears in his career record as askSenti in 2015-2016. He describes building SentiSum in 2016. Those dates mark an early company and a product taking shape, rather than a single ceremonial moment when everything suddenly worked. He wrote the initial minimum viable product himself. Later versions came from the product and engineering teams, whose work he has been careful to credit.

There is a practical advantage in a founder who can make the first version. An idea can become something a customer can argue with. That matters here because the eventual problem is full of inconvenient detail: different channels, different labels, several subjects in one message. A spreadsheet can keep a count. Understanding what belongs together asks considerably more of the person, or the software, doing the reading.

The firewall had other ideas

One early collaboration put that work against real data. SentiSum approached Just Eat with an offer to summarise and analyse customer reviews. The partnership gave the startup access to production data and feedback that helped shape product features. Khandelwal and Just Eat’s technology head, Matt Gillam, presented the collaboration at Food Tech Week London in 2016.

The obstacles were wonderfully prosaic. There were questions about contracts and data protection. There was an application running on a port that Just Eat’s firewall blocked. There were differing priorities. Just Eat arranged a same-day meeting with its legal team; a follow-up discussion resolved the technical snag. Both sides worked out shared priorities.

The episode supplies a useful corrective to the polished founder origin story. Even a product designed to understand language must get through a contract, a network and somebody else’s timetable. The first customers help make the software real. They also reveal all the things surrounding the software that have to work before anyone can use it.

That practical emphasis continued. In his later account of SentiSum’s rebrand, Khandelwal organised the company’s values around simplicity, speed and smartness. He also thanked early customers for their feedback. The interface and the relationship were both part of the product’s development. A difficult system that produces a clever answer still has to find a willing reader.

A score cannot repair a website

By November 2024, Khandelwal was discussing the limits of surveys and Net Promoter Score on CX Insider. NPS asks about a customer’s willingness to recommend a business. He questioned what companies could actually do with the resulting number. The website anecdote makes the point neatly: an internal explanation can survive until someone looks closely at what customers have said.

His writing carries that argument into operational detail. In The Silent Majority, he describes an anonymised retailer whose fan and air-conditioning calls rose thirteenfold in a week. Hot weather explained the demand. Reading the conversations exposed an additional problem: the website showed stock available in stores whose shelves were empty. Customers made the trip, found nothing and contacted the business.

He presents that case as an example of going beyond a visible spike to the mechanism underneath it. Weather cannot be repaired by the retailer. Stock information can. The distinction gives a team a different conversation to have, with a different person responsible for the next step.

In his retail writing, he returns to the gaps between the website, the store and the app. Customers experience the results of policies, payment processes and delivery promises. They rarely care which department designed each part. His work takes that ordinary customer perspective and asks the company to examine the whole sequence rather than the fragment visible from one desk.

The bot needs a performance review

The arrival of more customer-service automation gives Khandelwal another set of questions. His guide to evaluating AI support analytics asks whether the system can inspect conversations handled by an AI agent. A business that reviews only the tickets reaching a human leaves a portion of its service unexamined.

The guide also asks buyers to test detail, latency, ease of use and explainability. A broad billing category offers a weak starting point for a product manager. A specific failure offers something that can be investigated. An answer should lead back to the conversations supporting it. An executive should be able to ask a question without commissioning a new report.

These demands make for a less comfortable sales demonstration. They ask what the software actually knows, how quickly it knows it, and how someone checks the answer. The glossy screen has to earn its place. A useful demonstration is one in which the buyer can follow the explanation all the way back to the customer’s words.

The distinction also protects the meaning of a completed conversation. A bot can stop speaking. A customer can stop replying. Neither event, on its own, tells the business whether the underlying problem has been resolved. Khandelwal’s insistence on examining automated interactions keeps that question open instead of letting the end of the chat supply the verdict.

An answer looking for an owner

His more recent argument moves from interpretation to responsibility. On LinkedIn, he describes findings that reach a presentation, receive agreement and then disappear into a backlog. His proposed remedy is practical: decide who owns each kind of finding before starting an AI pilot. Put the information into the tools where people work. Count the problems acted on.

“A dashboard that nobody opens is the same as an answer nobody heard.”

Sharad Khandelwal

In his 2025 essay on retention, he asks whether customer-experience, product and marketing teams can see the same emerging problems without arranging a meeting. Speed, in that framing, includes the time it takes people to share an understanding. The finding must travel across the business while someone can still use it.

The people around him have changed, too. Vivek Ganotra joined as co-founder in June 2024, bringing experience at companies including Salesforce, THG, BAT and Accenture. SentiSum’s partnership with Denmark-based MakesYouLocal connects the analytics work with cross-border e-commerce support. Khandelwal has described that partnership as a way to learn from knowledge across industries.

His hiring language has been less genteel than the average company biography. A marketing recruitment post invoked the resilience of a cockroach and said the company had grown without deep pockets or giant teams. It is an unusual mascot for ambition, but an effective description of the persistence he wanted applicants to recognise. Category leadership remained the aim; survival had earned a place in the vocabulary.

Leave room for the long walk

Alongside that ambition sits an ordinary account of life between meetings: cooking, eating, a long walk and some television. Asked about the rules he lives by, Khandelwal offered “be kind and pursue happiness”. His explanation placed freedom and happiness among the reasons for building a successful company.

That detail fits a profile concerned with the person doing the work. The career is technical. The company’s subject is human language. Its daily challenge is getting institutions to take that language seriously. The founder’s own account leaves space for a life beyond the institution he is building.

The website that customers liked is a good place to return. It reminds a business how easily an assumption can acquire a budget. Khandelwal has spent years working on a way to examine the sentences that might interrupt it. The customer has already explained something. Somewhere inside the company, there needs to be a person who reads the explanation and can change what happens next.