07 · How I can help
Seven projects. Each starts from something on this site, and none needs AI to write a word.
The Ken's product problem isn't the journalism. It's the path between a reader and the journalism: finding a story, getting into the app, being recognised as a paying subscriber, renewing, and being visible where questions now get asked. That path is data and AI engineering, which is the work I do.
The projects on a two-by-two
Revenue effect against effort, in The Ken's house format. The positions are my estimates before seeing any internal data. Hover for the reasoning.
The thread through all seven
In 2021, about 98% of The Ken's readers were on 1- or 3-year plans, with no refunds. So the business is decided on one day a year per reader: the renewal date. Every project here either keeps that day from going wrong or brings new readers to it.
A renewal radar
Revenue fell 8% in FY25 with no visible change in list prices, and the last paid-subscriber number is from 2023. When nearly everyone is on annual plans, churn is invisible for 11 months and then arrives all at once. The radar spots readers who are fading 60–90 days before their renewal date, while there's still time to win them back.
- The evidence
- FY25 revenue −8.1% (Company). Seven reviews signal churn (cancelling, not renewing or uninstalling), and billing complaints rose from 2 to 8 between periods (Readers).
- What I'd build
- A score for each subscriber from reading behaviour already in Mixpanel: days active, stories finished, newsletter opens, podcast listens, app versus web, payment failures. Then a playbook by risk band: a personal "what you missed" email of stories on their topics, a gift credit, a corporate-seat offer to a reader whose employer is already a customer, or a call for high-value accounts.
- Success looks like
- Renewal rate in the targeted group against a held-out control group, so the effect is measured, not assumed.
- Where I've done this
- At Paytm I worked in credit risk on Postpaid while its volume grew from ₹30 Cr to ₹3,000 Cr a month in 18 months, scoring who would pay and who wouldn't. A missed renewal is the same kind of problem as a missed payment.
Find every paid reader who's locked out
Readers who paid and were then asked to pay again, downgraded, or shown "subscription not renewed" are the most avoidable churn there is. Seven reviews describe it between 2023 and 2026. Most got no public reply.
- The evidence
- "Paid but not recognised" is the third most common failure mode since 2023, and billing sentiment is −1.00 (Readers).
- What I'd build
- A daily job that matches payments (Razorpay or the web gateway, App Store, Play) against subscription status in WordPress. It lists every paid but not active account, every renewal charged but not applied, and every double charge. Subscriber Success gets the list each morning, before the reader writes in.
- Success looks like
- The number of mismatches found and fixed each week, and billing complaints in reviews falling to zero.
- Time
- About two weeks, with read access to the payment and subscription tables.
AI answer-engine visibility, and a crawler-by-crawler policy
In my test Perplexity cited 0 the-ken.com pages out of 65 sources, on topics The Ken has covered. When asked about The Ken directly, it gave the wrong prices and recommended a newsletter that ended a year ago, which The Ken's own pricing page still lists. robots.txt blocks citation crawlers and Bing, but leaves Google's and Meta's training crawlers allowed.
- The evidence
- The AI question: the six-question test, the brand answers, the robots.txt table.
- What I'd build
- A weekly panel of about 200 questions across ChatGPT, Perplexity, Gemini, Google AI Mode and Claude, tracking whether The Ken is cited, who is cited instead, and what is said about its products. Then a 60-day test of "block training, allow citation" on free pages, with llms.txt and correct structured data. The results also give The Ken evidence of its value if a licensing conversation comes up.
- Success looks like
- The Ken's citation share on its own beats, sign-ups referred from AI engines, and correct brand facts in AI answers.
- Where I've done this
- I built Vedlora, which tracks brand visibility across seven AI engines with confidence intervals, competitor gaps and the sources each engine trusts.
Search that understands time, people and companies
The March 2026 semantic search is a real step forward, and it handles topics well. In my test, "fintech companies that shut down before covid" returned mostly stories from early in the pandemic, plus one from 2022. Search result URLs also can't be shared or bookmarked (Product audit). Praveen's own observation is that high-intent searches convert, and that people search for their employer.
- What I'd build
- A query-understanding step before retrieval that pulls dates, authors, companies, sectors and story type out of the question and turns them into filters. Then hybrid ranking (keyword plus vector plus recency) and result cards that show the reporter's own paragraph that answers the question. Retrieval only, nothing generated. Plus a test set of about 300 real queries with judged answers, so every change is scored before release.
- Success looks like
- Accuracy on the test set, the share of searches that end in a story read, and the share of searches by non-subscribers that end in a sign-up.
- Where I've done this
- At Flipkart I used vector search to take down more than 50,000 malicious ads, domains and apps. At American Express I built an LLM-plus-retrieval pipeline that cut manual effort by 60%. ReadX, my news app, ingests and ranks 25,000 articles a day, with full-text search and category, country and region filters (see ReadX).
"Your company on The Ken": company pages and corporate leads
People search for the company they work at to decide whether The Ken is for them. The Explore menu's company list still leads with Byju's and Dunzo, and includes Hotstar and Facebook under old names. Ten years of reporting tagged by company is The Ken's most distinctive asset, and right now it's sorted mainly by date.
- What I'd build
- (a) A company graph from the archive: every company, its people, rivals and investors, with links back to the stories. (b) A public page per company showing headlines, a timeline and how often The Ken has covered it. These are good for Google, good for AI citation, and a natural place for an "unlock everything about your company" offer. (c) A corporate lead list for Sthaman's team: companies with many individual readers or free sign-ups but no team plan, found from email domains.
- Success looks like
- Sign-ups from company pages, and new corporate teams from the lead list.
- Why it fits
- Corporate seats are the steadiest revenue The Ken has, and the February 2026 corporate page redesign was built for exactly this buyer.
A weekly reader-and-brand watch
This site is a one-off version of it. The weekly version reads every new app review, comment, forum thread and mention, tags each by aspect, and sends a digest. It also tracks archive leakage, APKs and lookalike domains.
- The evidence
- Developer replies on Play fell from 42 in 2021 to 4 in 2026. 2,399 pages are on archive.today. The handle domain thekenweb.com belongs to someone else (Readers, Brand risk).
- What I'd build
- Run it on Doeity, my brand-intelligence product: mentions tagged by type and aspect, low-value mentions dimmed rather than hidden, alerts on spikes, and a weekly report. Billing complaints get routed to Subscriber Success the same day, with a draft reply ready for a person to edit and send.
- Success looks like
- Review reply rate, time to reply, store rating, and leak counts going down.
Audio for every story
Audio is the most persistent feature request: it appears in every period of the nine years of reviews. Audio is also where The Ken leads: Daybreak is #4 in Apple's India News chart, and four Ken shows are in the top 200. The Morning Context already includes text-to-speech in Premium.
- The question first
- Is a story read aloud by a synthetic voice compatible with "AI will not write stories"? The words would be the reporter's, unchanged. That is for the editors to decide, not product. If the answer is no, the alternative is human narration for the top stories each week, with podcast-grade playback for all of them: background play, lock-screen controls and resume position.
- Where I've done this
- ReadX's Listen Mode reads any story aloud in a natural voice, with one tap to switch between reading and listening (see ReadX).
ReadX: I've already built the reader's side of this
ReadX is a free news-reader app I built and run on iOS and Android. It ingests and ranks about 25,000 articles a day from 100+ sources across 10+ topic categories, and turns them into a personal feed. Many of the requests in The Ken's reviews are features ReadX already has in production.
| ReadX feature | What it does | Where it meets The Ken |
|---|---|---|
| Core reading | ||
| For You feed | Ranks headlines by each reader's own reading history. The more someone reads, the better the feed gets | The same reading signals drive The Ken's new recommendations, and the renewal radar (project 1) |
| Trending | Live rankings of the most-read stories | "Most read story in AI" is the example query on The Ken's own search box |
| Latest | Unfiltered headlines, sorted only by time | For readers who want everything The Ken published today, across 8 newsletters and daily stories |
| Discover | Daily cards: facts, book picks, career tips, brain teasers | A light daily habit, like The Ken's Visual Stories |
| Enhanced reading | ||
| Listen Mode | Any story read aloud in a natural voice, with one tap to switch between reading and listening | Audio is The Ken's most persistent request, and rivals' too (Rival readers, project 7) |
| Offline reading | Every story a reader opens is saved automatically for flights and dead zones | "Downloaded content never stays": offline sentiment is −0.75 in The Ken's reviews |
| Bookmarks | Save any story with one tap and come back to it later | Reading lists and resume position are a top-5 request in The Ken's reviews |
| Explain (Deep Dive) | A plain-English AI breakdown of any story: who, what, why it matters | Not for The Ken Its policy rules out AI-written text, and I'd keep it that way |
| Personalisation and control | ||
| Personalised setup | Readers pick topics and regions on first launch, and the feed keeps learning | The Ken's 30-day trial already builds a personalised page. This makes it smarter each day |
| Smart filtering | Block topics, keywords or sources a reader never wants to see | "So many newsletters that it's becoming very hard to keep up": let readers tune the flow |
| Country, region and category filters | Combine all three at once | The same idea as company and sector filters for search (project 4) |
| Daily digest push | An opt-in daily notification with the top stories from a reader's own feed | A daily reason to open the app, built around The Ken's 8:00 IST publishing time |
| Community and sharing | ||
| Community stories, follow readers, profiles | Readers post stories, follow each other and build a reading network | The Ken's subscriber comments are already a selling point, but the homepage shows only two |
| One-tap sharing | Share any story to WhatsApp or social media from inside the app | The same moment as The Ken's gift links, and where leak control matters (Brand risk) |
| Experience | ||
| Search | Full-text search across articles, topics and readers, updated in real time | The core of project 4 |
| Dark mode · swipe to read | A dark-first design for long sessions, and vertical swiping between full stories | Reading-experience sentiment is −0.41 in The Ken's reviews |
The hard parts are already solved
Ranking a large, fast-moving archive for one reader. Narrating it well. Keeping it working offline. Learning from every tap. These already work in production at ReadX, and they are the same problems behind The Ken's recommendations, audio and renewals.
The Ken is a paid product with a voice
ReadX is free and aggregates other publishers. The Ken sells its own reporting, and its AI policy is part of the brand. So the parts that carry over are ranking, audio, offline, filtering and search. The AI explanations stay at ReadX.
The first 90 days
Sequenced so something useful ships every two weeks, and the bigger projects start from data rather than guesses.
Stop the leaks
- Payment-to-subscription reconciliation running daily (project 2)
- The reader-and-brand watch live, plus a one-off review backlog cleared (project 6)
- A baseline AI-visibility panel, and the 100-copy archive.today check
- Renewal data audit: what Mixpanel and billing data can actually support
Score and test
- Renewal-risk model v1, with a held-out control group (project 1)
- Search test set of about 300 judged queries, and a query-understanding prototype (project 4)
- Crawler-policy test on free pages, if approved (project 3)
- NewsArticle and paywall markup extended to newsletter and podcast pages
Grow
- First renewal-playbook results against the control group
- Company graph v1 and 50 pilot company pages (project 5)
- Corporate lead list delivered to sales
- A recommendation on audio, with the editors (project 7)
Why me
Five years building analytics and AI systems for large Indian and American companies, plus 15+ products I've shipped myself. Several of them address The Ken's exact problems.
ReadX
A news app on iOS and Android that ingests, ranks and personalises 25,000 articles a day, with Listen Mode, offline reading, smart filters and a reader community. Feature by feature →
Vedlora
Tracks how ChatGPT, Perplexity, Gemini, Claude, Grok and Google AI describe a brand, with confidence intervals. Project 3 in a box.
Doeity
Brand intelligence: customer sentiment by aspect, competitors, and brand protection. This site follows its method.
India Today Group
Built a brand and sentiment intelligence platform used by the group's CMO. I've worked inside how an Indian media company looks at its audience.
Paytm · Flipkart
Credit-risk analytics through 100× growth in Paytm Postpaid volume. Vector-search brand-threat work at Flipkart behind 50,000+ takedowns.
American Express
An LLM and retrieval reporting pipeline across 10+ channels, with 60% less manual effort. As Manager, AI & Analytics, at EXL.
Ways to work together
Whichever fits the team. Each one starts with the same first step: two weeks with the data.
A two-week diagnostic
Read-only access to analytics and billing. Output: the locked-out-payers list, a renewal-risk baseline, an AI-visibility baseline, and a ranked plan. A small, fixed scope.
A quarter-long engagement
The 90-day plan above, working with Praveen, Iqbal and the product team, with fortnightly reviews and everything handed over: code, models, dashboards.
Join the team
As the person who owns data and applied AI across product and growth, working inside the AI policy, alongside the people who built semantic search.