INMA Study Tour on Artificial Intelligence - Ritvvij Parrikh INMA Study Tour on Artificial Intelligence | Ritvvij Parrikh Humane ClubMade with Humane Club

INMA Study Tour on Artificial Intelligence

2 NOV Thursday 08:00 Asia/Kolkata
Duration :
1 day(s) 10 hour(s)
Group 81
Block Pattern: Regular

The International News Media Association hosted the 2023 INMA Study Tour on Artificial Intelligence, where 25 news leaders from 10 newsrooms attended. Madhavi Sekhri, INMA’s South Asia Division Head, generously invited me to moderate the two days. Here’s my roundup from the two days.


Meet the Cohort

The meeting was well-attended with 27 participants from 11 different newsrooms. We had representation from Product, Technology, Editorial, and Business departments.

  1. From Jagran New Media, we welcomed Gaurav Arora, the COO, along with Subhashish Dutta, who serves as Associate Editor.
  2. The team from Deccan Herald included their CEO, Shankar Sitaraman, and Pulkit Gupta, the Editorial Head.
  3. Hindustan Times sent a robust delegation led by Aviral Mathur, the Chief Data Scientist, followed by Abhishek Sharma, the CPO; Amit Verma, the CTO; and Varun Bagga, the Head of CDP.
  4. ABP News also had a significant presence with Subhamoy C. holding the position of CTO, Anindya B. as AVP, Arindam B. as CPO, and additional team members including Bishwajit K, and Debmalya D.
  5. From Eenadu, we had Brihathi Kiron Cherukuri, the Director, and Khayum Basha, the Chief AdOps.
  6. Amar Ujala was represented by Vibhas Sane, the Associate Editor, and Vikas Shekhawat, the AGM Product.
  7. The team from Malayala Manorama included Santhosh Jacob, the Content Coordinator; Venkatesan Mohanam, an Architect; and Tom Antony, an Engineer.
  8. Representing Sakshi Media Group were Anurag Agrawal, the CEO, and Kancharla Yadagiri Reddy, the Deputy Editor.
  9. Anastasia Geydarova was the sole representative from Sputnik News.
  10. From The Hindu, we had Manoj Kumar Venkatara, who works in Growth Product, and Nazir Ahamed, a Senior Product member.
  11. Lastly, from The Times of India, we had Chandrima Banerjee, our assistant editor news, and me.

Each attendee brought a unique perspective from their respective positions, contributing to a diverse and enriching exchange of ideas.


Warmup Session

In the bus, I covered some material from my JournalismAI lecture.

2.1 Grounding Ourselves in Reality

Let’s start by grounding ourselves in reality, particularly when it comes to the news industry’s business model and revenue challenges.

  • The average revenue per user (ARPU) in the news business is quite low because while we serve millions of users, each contributes only a minuscule amount to the overall revenue.
  • Furthermore, news has become a commodity and most newsrooms have not built competitive differentiation to be able to justify charging for subscription.

If you study the fundamentals of the news business, it is evident that the industry has either relinquished or been pushed out of many key business functions to AI-driven marketplaces, particularly in the domains of Supply (Search and Social) and Ad Engines.

Copied

We’ve also spent hard-earned time, money, effort, and careers retooling to play BigTech’s Hype – Incentivize – Mass adoption – Commoditize game, typically for short-term cash but assured long-term further commoditization. These growth hacks are tactics that don’t add much to strategic differentiation eventually.

Copied

Additionally, the Environment Is Changing:

  • Historically, computer science has provided storage, search, and data analysis capabilities. The expertise of understanding, applying, and generating content was almost exclusively within the domain of editorial teams. However, with the emergence of large language models, this dynamic is evolving. AI technologies are advancing up the value chain, venturing into territories that were once the preserve of editorial departments.
  • Historically, news companies had to pay a form of ‘tax’ for discovery (SEO, Social) and monetization (Ad Networks). Increasingly, this will be supplemented by an ‘editorial tax’ in the form of ChatGPT or Bard API costs.
  • Moreover, how our users interact with and access knowledge is changing. Chat interfaces are slowly replacing search. However, there is yet to emerge a clear strategy for monetizing chat interfaces.
  • When this change in interface is coupled with the third-party cookie deprecation, it is likely that Search, Ad Networks and thus news publishers will lose revenue. For example, Stackoverflow — a media product to help developers — is constantly losing users due to GitHub Copilot — an AI LLM tool that answers developer queries.

Uncertainty looms large.

2.2 So Why Could You Invest in AI?

  • Big Bet: Bet big in AI if you’ve a strategy to increase Average Revenue Per User (ARPU) or to reclaim strategic control with AI. However, can your competitors copy the same strategy? Do you have the cash to make it cost-prohibitive for competitors to follow you? If not, then in a small Total Addressable Market (TAM), not everyone can enjoy an increase in ARPU.
  • Table Stakes: You could invest in automation. However, it’s important to realize that these investments may ultimately become as commonplace as purchasing Process Technology Improvements (PTI), Automatic Number Identification (ANI), DataWrapper, and incorporating Accelerated Mobile Pages (AMP) and Search Engine Optimization (SEO). Such technologies, though once innovative, are now standard practice and do not offer a competitive edge.
  • Efficiency Bets: Lastly, ‘Efficiency Bets’ may seem alluring by potentially increasing the supply of content in Search and Social marketplaces to boost programmatic earnings. Nevertheless, this approach is unlikely to provide a sustainable competitive advantage. AI-driven marketplaces tend to balance itself out. For instance, Search and Social platforms may begin to down-rank AI-generated content, or the effective Cost Per Thousand impressions (eCPMs) could diminish over time. Therefore, it’s essential to monitor the unit economics on a weekly basis to ensure that the investment continues to yield the expected returns without diminishing in value.

May be yes, may be no.

Monetization. Remember, we operate in AI-driven marketplaces. Hence, I suspect that the marketplaces will correct themselves to possibly down-rank AI-generated content. Below are the early signs that I see:

  • OpenAI launched a new system to test if content has been written by AI.
  • Google, Schema.org, and IPTC — the organizations that control SEO — are coming up with changes to the schema that we’ll have to report if the content was AI-generated. Eventually, this will in turn feed into SEO algorithms and down-rank AI-generated content.
  • One platform — Medium — has decided to de-amplify AI-generated content.

Production Unit Economics. The cost of labor in India is low and the eCPMs in some regional languages can be as low as Rs. 10. These publishers will have to do their own unit economics calculations to see if there is a point to spending on OpenAI API costs and then getting humans to refine and fact-check it.

Maybe there’s an opportunity to blend human creativity with the output of AI to create something valuable.

Copied

2.3 What might not qualify as a Strategic Bet?

Yes, news media is a low-margin business, but then so are big retail companies like DMart and Walmart. The difference is the retail businesses continue to own demand-supply variables that control their business, and hence every strategic investment can help drive down costs. In contrast, we’ve lost control to AI-driven marketplaces (Search, Ads) that auto-adjust every time we try to change something.

The Peanut Butter Manifesto was written by an executive at Yahoo during Marissa Mayer’s tenure as the Yahoo CEO.

The Peanut Butter Principle refers to the strategy, or rather lack of one, where you do a little bit of everything, almost as if spreading a thin layer of peanut butter evenly across bread.

Generally, the peanut butter principle does not work because your cash/investment/focus is spread across many bets, eliminating the option of building deep competitive differentiation in any one thing.

Copied

“Replace Journalists with LLMs! “

That’s an expectation that isn’t based in reality. Here’s why 👇

2.4 What might qualify as a Strategic Bet? Here are three examples:

Google is a company that earns money by running targeted advertisements. A key variable in their business is to be able to uniquely identify users. Hence, they invested in building and giving away for free high-utility tools — Gmail, GDocs, Android — that users will need in their daily life and wouldn’t mind logging into.

Copied

Twitter, under Elon Musk, has made some wise investments to become Compute-Heavy.

  • They’ve replaced a manpower-heavy safety team with personalization and community notes (AI-assisted crowd-sourced fact-checking).
  • They are forcing users to subscribe. It might not be only for cash. Subscribers end up revealing their identity and get classified as digitally transacting users. Brands pay a premium to advertise to these folks.
  • Finally, they’ve killed off the Twitter API, thereby protecting the data inside from being used by LLMs.
Copied

One of the smartest strategic investment strategies I’ve seen so far is from the JSW Group. Instead of diversifying into general IT outsourcing, they are building on their leverages and have invested in three new businesses:

  • An online marketplace e-commerce for Steel, Cement, Paint. After the price of land, these three are the highest expenditure items in any construction.
  • They’ve set up a business that provides end-to-end vetted construction as a service. So, in case one wants to get a house manufactured and does not want to rely on local vendors, they can contract JSW.
  • Finally, the company is applying for a banking license.
Copied

2.5 Thinking Outcome First

In each of the five, you’ll end up spending cash, but the outcomes can vary.

  • You might not earn tangible cash from it. For example, many news companies tried investing in NYT Style immersive journalism in a half-hearted fashion but failed to monetize it.
  • Some table-stakes investments like AMP, SEO-optimization, etc., will become must-haves but won’t give you strategic cash. Similarly, bad implementations of AI, like off-the-shelf personalization of one-off widgets, won’t move the needle.
  • Then you could do deals with BigTech firms to speed up implementation in their marketplaces. However, it will only pay short-term cash. For example, a BigTech firm might pay you cash to create content for a new marketplace they are trying to seed. You will earn a good margin from it; however, in the long run, the BigTech firm will stop subsidizing once the marketplace is seeded.
  • Other times, news media companies end up investing in assets that they do not have the leverage to monetize. For example, building costly CDPs and DMPs but lacking the sales force to
  • Finally, there are waves like subscriptions which have an extremely low TAM in India. Hence, there will be only few winners in it.

2.6 How should you bucket everything you hear in these two days?

Bucket everything you hear into three categories:

  • Automate. Use AI to replace your employees’ labor. This will eventually become table stakes in the industry, similar to the spends in PTI, ANI, etc. However, the first movers will end up burning more cash to make the AI just work! The laggards will be able to directly replace manpower with AI.
  • Augment. Use AI to replace your employees’ labor and judgment. It will be extremely valuable but tough to prove attributable, sustained revenue from these investments.
  • Transform. These are your strategic bets that may or may not work, but if they do, they could give you a long-term edge.

2.7 What should your decisions be?

Bucket everything you hear into four decisions:

  • Hold: Ask upper management to keep observing this space! Be on the lookout for breakthroughs.
  • Assess: Ask middle management to study it deeply with tangible details.
  • Trial: Spend a small amount to pilot it with the goal of proving unit economics.
  • Adopt: Spend a big amount to scale up the unit economics.

2.8 Should you Build or Buy?

I’d recommend picking one of two extreme paths—compute-heavy or compute-light—while acknowledging that adopting AI is fraught with challenges.

Buy

Buying AI is great for efficiency and table-stake bets. However, if you use it for a strategic bet, then consider:

  • Your competitors can buy the same technology, thus eliminating all gains.
  • If you tightly couple a strategic bet with a vendor, you’ll end up spending a ton of cash integrating with them and might not have the cash and bandwidth left to exit a bad deal.

Build

Building AI in-house can be a nightmare, too. Apart from the fact that it can get extremely costly, you might not have the staff and systems to get it right:

  • News media companies operate technology that tends to have high error margins. For example, if your pollution tracker is off by 20%, it isn’t going to directly impact your bottom line. However, AI mandates setting up highly accurate data pipelines. News media companies tend to struggle with collecting data.
  • News media companies tend to serve millions of users but through CDNs. Hence, they have limited muscle memory of building and operating compute-heavy infrastructures!
  • Finally, all your stakeholders (bosses?) are used to operating deterministic software. In contrast, AI is probabilistic. Try explaining to your editor why certain stories are getting recommended to him and whether they should.

2.9 How are the two days organized?

On day 1, we will visit Big Tech companies that provide cloud infrastructures. Use these if you want to invest in strategic bets. Utilizing them will require you to spend a CAPEX (leveraging these meaningfully is a CAPEX at news media scale).

On day 2, we will have 10 small technology companies that will provide off-the-shelf tools requiring minimal integration. Use these as an OPEX for table stakes and efficiency bets until they are ROI-positive.


Day 1

We visited three BigTech companies—Microsoft, Google, and Amazon.

At Microsoft India
At Google India
With the AWS Team

Microsoft: For a long time, Google has dominated the search, programmatic advertising, and browser market (Chrome). Since the launch of OpenAI’s Large Language Models and Microsoft’s strategic investment in the company, Microsoft’s search engine (Bing) and browser (Edge) have been growing its share in these markets. Additionally, Microsoft acquired Xandr, a digital advertising marketplace similar to Google Ads. With this, is Microsoft increasingly becoming a viable alternative?

Are you looking to build custom AI solutions in-house? In this situation, it is likely, you’ll end up building on the Google Cloud or Amazon Web Services.

Google Cloud. All digital publishers use Google Analytics and most end up using BigQuery for analyzing their Google Analytics data. Beyond this, Google Cloud helps publishers in creating personalized, India-first news content, offering advertisers new revenue streams through data-driven insights and cross-promotion with retail.

Amazon Web Services. With cost savings and quick scaling, AWS helps publishers build competitive edge by offering solutions in the following three fields:

  • Monetization: Publishers can automate ad performance, manage digital identities, and use predictive analytics for revenue growth.
  • Content Management: A centralized cloud infrastructure streamlines workflows and content distribution.
  • Personalized Experiences: AI/ML innovations offer deep audience insights and optimize engagement.

I will write more about the material covered on Day 1 in the next 7 days.

Day 1 ended with a dinner get-together at the Delhi Gymkhana.


Day 2

On Day 2, we met 10 vendors at IIC.

4.1 Editorial

We had two vendors pitch: JustBaat.ai and Personate.ai. My assessment is that if you are not already, then go ahead and try out the technology.

Opportunity exists for those newsrooms that lack the “video DNA”. This technology can help bridge that gap and enable you to create videos at scale. It can also help you break away from the traditional reliance on celebrity anchors.

There is significant earning potential in video content since video eCPMs are higher compared to text. By tapping into video SEO and taking advantage of Google’s audio indexing, one can significantly boost visibility. Moreover, building brand IPs opens up opportunities for sponsorships.

Cost-saving is another critical factor to consider.

  • Producing videos, especially at scale, is costly due to expenses such as shooting time, anchor salaries, and editing. This technology can help automate most of this.
  • Furthermore, the capability to multi-publish textual stories in video format and to mass-create videos from automated data feeds, like air quality reports, stock market updates, and COVID-19 statistics, allows to drastically increase supply of videos.

However, risks should not be overlooked.

  • The current state of the technology leaves much to be desired as the anchors, along with their expressions and gestures, can often appear artificial, though it’s important to note that advancements are being made rapidly.
  • There’s a possibility that the content might be perceived as gimmicky if not executed well.
  • Another concern is AI-driven marketplaces adjusting: as more publishers begin to use AI to mass-produce video content, there could be an oversupply, which might lead to a significant drop in video eCPMs. Therefore, it’s crucial to keep a close eye on expenditures and earnings from programmatic videos on a weekly basis to stay ahead of the curve.

Personally, I want to get my own synthetic avatar made and experiment deeply with this technology.

Copied

Next, we had Manoj Prabhakaran from Chennai-based TrueInfo Labs provide consulting services to implement newsroom efficiency solutions on top of large language models. They offer a varied array of solutions. This will help newsrooms pick and choose what is right for them and get it implemented.

Why AI in CMS? In the past few months, integrating Generative AI into CMS has become table stakes! OpenAI’s ChatGPT APIs have made this a variable cost. Hence, we invited two CMSes.

WordPress

I deviate, but I want to spend a bit of time on why WordPress. In the last 10 years, I must have worked on multiple consulting projects to build CMS. Over the years, I’ve built a healthy appreciation for WordPress.

  • Unless you have massive scale, there is limited direct financial ROI that newsrooms have to gain from building and investing in custom-built CMS. Some newsrooms are realizing it now. For example, VOX shut down its CMS and migrated to WordPress.
  • Newsrooms got distracted building supply-side efficiency technology like CMS instead of revenue-generating advertisement targeting solutions.

Representing the WordPress community was Rahul Bansal from rtCamp. They are one of the sharpest WordPress agencies from Pune.

Here’s what Rahul shared:

  • WordPress is a Lego-like modular system, and there are plugins for everything.
  • Altis DXP provides editorial workflow assistance with ChatGPT.
  • RankMath SEO, an Indian company, provides solutions to mass SEO-optimize content using LLMs.
  • Elementor GUI plugin provides background removal solutions right inside of WordPress using LLMs.
  • WPML integrates with different translation models.
  • Akismet provides AI-driven content moderation.
  • FalconAI generates outlines of articles using AI.
  • etc.

Quintype

One of the most successful custom-built CMS in the market from India is Quintype. Over 200 publishers across the world use them! Chirdeep Shetty, their CEO, walks us through their AI integrations. According to him, LLMs are best left to the OpenAIs and Googles of the world, and the focus of Quintype is to get the UX right as per the workflow of the newsroom. Instead of me trying to explain what they do, I’d recommend watching the video below.

Copied

I co-founded and ran Pykih, a profitable information design studio for five years before folding it. I can say with some authority that Gurman is probably India’s most talented journalist turned information designer! She has now founded ReVisual Labs, her own Information Design Studio. She can help you build competitive differentiation with immersives!

Opportunity: In an industry as commoditized as news, Gurman’s work can help you build competitive differentiation, especially for subscriptions.

  • Investing in the creation of long shelf-life visual properties could be a strategic move.
  • These properties hold the potential to capture audience attention during High Traffic Events or even foster a habit of daily engagement, thus aiming to enhance daily and monthly active users (DAU/MAU).
  • The measure of success in this endeavor is subscription conversions and an uptick in DAU/MAU.

However, this approach is not without its risks.

  • You might not be able to prove profit on a per-story basis. Hence, information design might not work for short-term, tactical stories.
  • Moreover, integrating immersive stories into your current content management system (CMS) may prove to be complex and problematic.
  • It’s also important to consider that immersive stories typically yield little in terms of search engine optimization (SEO) benefits, which can be a significant drawback for online visibility.

Thus, carefully consider if immersives works for your audience and if it does then invest big in it to get that competitive differentiation.

Copied

Finally, we had Anshul Pandey from Kalakari.ai. Anshul is one of the brightest NLP-based data scientists I’ve known for almost a decade now. He founded and continues to run the New York-based Accern. He is now in India and is setting up a number of different AI-based firms. Kalakari is using LLMs to speed up content creation.

4.2 Product

In today’s world, digital news companies cannot survive without products like Mixpanel, Clevertap, and Web Engage. These tools are critical to record clickstream events at scale, warehouse them, analyze them, and when required, segment them to send out targeted communication via SMS, Email, In-Apps, and Push Notifications.

There were two other vendors — the Langoor team presenting Quilt.ai and Conscent.ai. By this time, I was extremely exhausted and took a break. I will add their details in the coming week. If you were there at the study tour and want to help me write this part out, please DM. 🙂

4.3 Business

We saved the most insightful session for last. Here are insights from Sandeep Amar, the ex-CEO of Indian Express and now the founder of PDlab.me.

4.3.1 Monetization with Advertisement

Sandeep Amar’s slide
Sandeep Amar’s slide

Programmatic and direct digital eCPMs are facing significant challenges, especially when it comes to the Hindi market where eCPMs are as low as Rs. 10.

  • This scenario becomes even more problematic when considering the revenue distribution model, where ad networks, including giants like Google, claim as much as 50% of the top line.
  • Google primarily operates on a performance-based buying mechanism, utilizing cost per click (CPC), whereas media purchases typically hinge on impression-based models (CPM). This dichotomy leads Google to reverse calculate CPC to CPM when making payouts to media, which invariably affects the bottom line.
  • Additionally, the situation is compounded by the involvement of Supply-Side Platforms (SSP) and Demand-Side Platforms (DSP), which are in the business of constantly trading ads. These platforms further erode the margins, taking away a significant portion of the revenue in the process.

Based on the discussion within the cohort, we’ve come up with some rough estimates regarding India’s digital advertising market.

  • It is valued at approximately Rs. 30,000 crore, which translates to about $4 billion.
  • A substantial majority of this market, about 85% or Rs. 25,000 crore, is dominated by digital behemoths Google and Meta.
  • In contrast, news publishers in the digital space have a much smaller slice of the pie. Their total digital advertisement revenue potential through direct sales stands at around Rs. 1000 crore, with a more optimistic estimate reaching up to Rs. 2000 crore.
  • This estimation aligns with insights from Ashwin Padmanabhan of GroupM.

Changing scenario: With third-party cookie deprecation, Google’s revenue is expected to drop, and so is that of news publishers.

The direct sales pie is degrowing.

Copied

Ad Networks are AI-driven marketplaces owned by BigTech. These AI models are optimized to maximize revenue for the BigTech company. These models keep pushing the boundary of what is the lowest eCPM they can pay out. So if Hindi eCPMs are at Rs. 10, they will keep pushing the floor to Rs. 9.90 across a couple of months.

Hence, it is critical for news publishers to invest in AI to reverse push the boundary of the floor price up until the sell-through drops. By playing on buy and sell, you could see a 25% increase in programmatic revenue. Revenue from advertisements is seasonal. Hence, of course, do this during high-demand months (Oct-Nov-Dec) and play it conservatively from Jan 1.

Copied

4.3.2 Distribution to Invest In

Users spend 10x to 20x more time on social media compared to the time they spend on news media. Given that, it is recommended to invest in both native (on-platform) and social footprint and sell both as part of direct ads.

  • Direct sales. Micro-celebrities can charge brands anywhere from Rs. 30,000 to a lakh for a social media post. Top cricketers charge a couple of crores for a post on their social media handles. This is especially true for videos.
  • Subscriptions. Referral traffic from Twitter tends to give much higher subscription conversions compared to traffic from other sources like Facebook.
Copied

4.3.3 Content Strategy that Aids with Advertisement

If brand marketing or display advertising were only a matter of recency and frequency at the lowest possible cost, then brands would advertise on low-cost conspiratorial or explicit sites. However, this would negatively impact the brand of the business.

On the upside, brands want the credibility of the publisher to transfer to them. At a minimum, by advertising with credible publishers, brands avoid unnecessary public backlash. This is why businesses prefer advertising with credible news products over any random blog. This is also the reason why celebrities get ads but not everyone else. However, on news sites too, brands would not want their advertisements next to sad news — crime, rape, death, war, etc.

Below are some of the factors that contribute to premiumness:

  • Medium’s Exclusiveness: An advertisement on Super Bowl night is more valuable than on another media providing a similar set of impressions.
  • Quality of Content: Determines if the media is considered premium by audiences. For example, the stories on The Ken are much more well-researched and nuanced than those on say, a local entertainment gossip news website.
  • Platform: Mobile apps tend to have higher eCPM than websites.
  • Adherence to Cultural Norms
  • Fact-Checked
  • Content Strategy: Make news more meaningful, help people make money, save money, become fit, choose the right gadget, plan travel better, know about the future of the country!
Copied

I will wrap up now. These two days were extremely rewarding, especially the conversations with other cohort members.