Logged-In Internet Products Often Rely on Algorithmic Distribution

Originally published in The Times of India.

All major internet products — whether involved in social media, food ordering, quick commerce, mobility, OTT, D2C or speciality e-commerce — eventually evolve into algorithmically distributed platforms by adopting recommender models.

The biggest problem with logged-in internet products is the high customer acquisition cost (CAC). Therefore, consumption and retention need to increase to recover the investment made in CAC.

That’s where the user of recommender models comes in:

1. Get Users To The Platform Consistently

Users primarily visit a product because of its brand, value proposition, positioning, and recall. Those lacking these attributes often depend on search engines and social media for top-of-the-funnel traffic.

However, all internet platforms must eventually figure out how to bring acquired users back. This is typically achieved through push notifications, emails, or phone calls.

But there’s a delicate balance: too many messages can drive users away, while too few or irrelevant messages lead to indifference.

Recommender systems can address this challenge by tailoring communication, potentially boosting click-through rates by over 100% while reducing uninstalls or unsubscribes.

2. Grow On-Platform Consumption

Once users are on the platform, increasing consumption is crucial. This involves presenting products, services, or content that users are likely to engage with.

For example, social media platforms might aim to maximize time spent on site, while food ordering apps could focus on increasing order size. Even for blogs relying on ad revenue, boosting pages per session is key.

Recommender systems solve this problem at scale. For instance, a platform with 100 page views and a pages-per-session ratio of 2.4 has 41.66 initial session triggers and 58.33 from on-platform recirculation. A 100% increase in click-through rates can double recirculated views to 116.66, raising total page views to 158.33 — a 58.33% increase.

3. Improve Retention By Reducing Friction

Retention challenges are complex, with excessive ads and push notifications often driving users away. Recommender systems help mitigate these too.

Dynamic Ad Density: Would you ask someone for a favor when you first meet them? Or when they are dealing with a personal loss? In the real world, we adjust our approach based on the situation. Yet, online, we often present the same advertisement density to all users. Recommender models can help by dynamically adjusting advertisement density, increasing or decreasing it based on the user’s likelihood to churn.

Reduce Push Notifications: As mentioned earlier in this post, recommender models can boost the click-through rate (CTR) of push notifications by 100%. But what if reducing uninstalls is more critical than increasing sessions? In such cases, recommender systems can help reduce the total number of push notifications sent by 50% without decreasing the sessions generated by those notifications.

4. Enabling Revenue Diversification By Minimizing Cost of Revenue

All businesses eventually diversify into new revenue models, like Amazon’s shift from e-commerce to Amazon Prime subscriptions. However, introducing multiple monetization options on the same platform can lead to revenue cannibalization.

Hence, it’s essential to calculate the cost of revenue to maximize overall gains. For instance, if Story A generates 100 conversions from 100,000 paywall hits and Story B generates 50 conversions from 20,000 hits, Story B is more conversion-efficient.

Recommender systems can increase click-through rates and personalize content, achieving top-of-the-funnel targets for each revenue model while minimizing exposure costs.

5. Earning More Within The Same Engagement Levels

Recommender models trained without revenue data assume uniform profit margins across all consumed items, which is rarely accurate.

Different content categories and revenue models often have varying margins. For example, in media, advertisement rates differ by content type, while in food delivery, aggregators receive different commissions from each restaurant.

By optimizing recommender models to consider these margins, companies can increase earnings without necessarily raising consumption levels.

6. Increasing Revenue Through Price Differentiation

The right price for a product often depends on the customer’s context.

• Platforms like Gumroad allow creators to adjust prices based on purchasing power parity.
• Similarly, businesses might charge less for first-time customers or more for loyal customers who derive significant value from the product.
• In lending, understanding the trade-off between interest rates and default risks can boost commissions from banks.

Again, these pricing strategies can be enhanced through recommender systems.

7. Moving Supply From Fixed Costs To Revenue Share

Internet products can either operate as brand shops, selling in-house created content and products with associated fixed costs, or as marketplaces, like MSN, YouTube, Medium, or Amazon, offering third-party content and products on a revenue-sharing model.

For brand shops looking to evolve into marketplaces, building recommender models becomes critical without which distribution won’t be effective.

Conclusion

Struggling with low direct traffic, low loyalty, high costs, or limited monetization options? To address these challenges, adopt algorithmic distribution and transform your product into one that truly earns user loyalty in 2024.

–

Want to brainstorm? I am looking to host ThinkIns — one or two long brainstorming sessions on a well-defined topic with 3–5 domain experts in the field. Which topic? Could be any. DM me. Exquisite green tea or coffee is on me.

Want to republish it? This post was released under CC BY-ND — you can republish it as is with the following credit and backlinks: ‘Originally published by Ritvvij Parrikh on The Times of India. The author retains the copyright and any other ancillary rights to the post.