Modernizing Discovery and Promotion in Personalized Products

The best discovery systems do more than rank what a user is most likely to click next. They help people find value: the thing they came for, the thing they did not know they wanted, and the thing that is culturally or strategically relevant right now.

In mature digital products, personalization is rarely driven by one objective alone. It is usually a blend of four forces: what is personally relevant to the user, what is broadly popular, what is currently trending, and what the business strategically wants to amplify. The art is not choosing one of these signals. The art is knowing how much weight each should carry in a given context.

A new user may need more help from popularity and trend signals because the system does not yet know much about them. A deeply engaged user may benefit from stronger personal relevance because their behavior provides richer signals. During a seasonal moment, a cultural event, or a major business initiative, strategic and trending signals may deserve more influence. The best systems are dynamic. They adjust the balance based on user context, product context, business goals, and real-time performance.

Promotion should amplify discovery, not compensate for it

Promotion is most effective when it acts as an accelerant, not a crutch. Its job is to help strong or timely experiences find the right audience faster. It should be able to take a spark of user interest and give it oxygen.

But promotion should not be used as a substitute for healthy personalization, good metadata, strong product design, or high-quality content and experiences. If a discovery system consistently needs manual intervention to make important items visible, that is often a sign of a deeper problem. The ranking model may need refinement. The metadata may be incomplete. The user journey may be too rigid. The system may not be capturing the right intent signals.

A modern promotion system should serve as a strategic audience development tool. It should enhance the underlying discovery engine, not mask its weaknesses.

It should enhance the underlying discovery engine, not mask its weaknesses.

The four forces behind modern personalization

A useful way to think about discovery is as a blend of four signal families.

1. Personal signals reflect what is relevant to an individual user. These may come from explicit preferences, past behavior, affinity patterns, engagement depth, similarity to other users, or inferred intent. Personalization is strongest when the system has enough high-quality behavioral context, but it still needs room for exploration. A system that only gives people more of what they already know can become stale.
2. Popular signals reflect what is broadly resonating across the user base. Popularity is especially useful for new users, anonymous users, or moments when the system has limited confidence in personal relevance. Popularity also helps users stay connected to what others are engaging with, which can be valuable in products where shared cultural awareness matters.
3. Trending signals reflect what is gaining momentum now. This is different from popularity. Something can be popular in an evergreen way, while something else may be rising quickly because of seasonality, news, social conversation, a live moment, a launch, or a broader cultural pattern. Trending signals help a product feel current.
4. Strategic signals reflect business priorities. These may include new launches, seasonal campaigns, marketplace objectives, inventory considerations, lifecycle management, contractual obligations, brand priorities, or other goals that matter to the business. Strategic signals are legitimate and often necessary, but they must be balanced carefully against user value. When strategy overwhelms relevance, users feel it.

The healthiest systems do not treat these forces as fixed rules. They treat them as tunable inputs. The right blend changes by user, surface, session, lifecycle stage, and objective.

Human judgment still matters

A common mistake is to frame personalization as a competition between algorithms and human curation. In practice, the strongest systems use both.

Algorithms are good at scale, pattern recognition, optimization, and adapting quickly to user behavior. Human teams are good at understanding context, cultural relevance, brand nuance, editorial judgment, and business priorities that may not yet be visible in the data.

The goal is not to replace human expertise with automation. The goal is to encode human judgment in a way that can scale responsibly.

That means giving teams structured ways to express promotional intent without forcing them to manually control every placement or ranking decision.

For example, a team might indicate that a certain theme, category, product, topic, or experience is especially relevant during a defined period. The system can then consider that signal alongside personal relevance, popularity, and performance. In some cases, the business may need guaranteed exposure. In other cases, the better approach is simply to give the system a signal that something is timely or important and allow the ranking model to decide where it fits.

Some promotional needs are constraints. Others are signals. Confusing the two leads to poor user experiences and poor measurement.

Some promotional needs are constraints. Others are signals. Confusing the two leads to poor user experiences and poor measurement.

Promotion has real risks

Promotion can create value, but it also carries risk.

Too much promotion can create fatigue. Users may begin to ignore repeated placements, distrust the experience, or feel that the product is pushing an agenda rather than helping them find value. Over-promotion can also reduce perceived depth. When the same items dominate the experience, users may conclude that the product has less to offer than it actually does.

Promotion can also crowd out discovery. If a product continually amplifies the same high-priority items, it may suppress emerging, niche, or mid-tier items that could have found the right audience organically. This is especially problematic in ecosystems where long-term value depends on breadth, diversity, and serendipity.

There is also a measurement challenge. When something performs well during a promotional window, it can be difficult to separate true user demand from promotional lift. A spike in engagement may reflect genuine product-market fit, or it may simply reflect unusually high visibility. Without careful experimentation, holdouts, attribution, and post-promotion analysis, teams can overestimate the durable value of a campaign.

Good promotion systems need guardrails. They need impression management, frequency controls, cooldown logic, diversity protections, and measurement discipline. They should optimize not just for the performance of the promoted item, but for the health of the overall product experience.

From manual merchandising to intelligent orchestration

Many organizations start with manual placement. A team chooses what should appear in a prominent surface, schedules it, and moves on to the next campaign. That model can work at small scale, but it becomes fragile as products grow more personalized, more dynamic, and more context-dependent.

Modern digital products require intelligent orchestration. Instead of asking, “What should everyone see?” ask instead, “What is the best thing for this user, in this context, on this surface, at this moment, given both user value and business goals?”

That changes the role of merchandising and promotion. The work becomes less about manually filling slots and more about setting strategy, defining constraints, supplying high-quality signals, and measuring outcomes. The system handles more of the allocation, sequencing, and personalization. Human teams retain influence, but that influence becomes more structured and more accountable.

The work becomes less about manually filling slots and more about setting strategy, defining constraints, supplying high-quality signals, and measuring outcomes.

This is where promotion becomes part of a broader discovery ecosystem. It is not just about pushing individual items. It is about managing attention across the entire product.

Cold start requires exploration, not just promotion

New items, new users, and new categories all create cold start challenges. The system has limited information, but it still needs to make useful decisions.

The answer is not simply to promote everything new. That approach can waste attention and degrade trust. Instead, systems need structured exploration. They should identify likely audiences, test exposure in controlled ways, learn from early signals, and expand distribution when performance justifies it.

Early engagement patterns, similarity to known items, metadata quality, audience modeling, and contextual signals can all help the system make better initial decisions. As more data arrives, the system should update its confidence and adjust exposure accordingly.

The goal is to help promising items find their audience quickly without overwhelming users or distorting the broader ecosystem.

What good looks like

A strong discovery and promotion system should improve outcomes at multiple levels.

At the user level, it should reduce effort, increase relevance, and create moments of useful surprise and delight. At the item level, it should help valuable experiences find the right audience. At the product level, it should improve engagement, retention, diversity, and long-term trust. At the business level, it should support strategic priorities without compromising the user experience.

The most important measure is not whether a promoted item received more exposure. Exposure is an input. The better questions are:

• Did users respond?
• Did the system learn?
• Did the experience remain diverse and useful?
• Did the promotion create durable value beyond the campaign window?
• Did it improve the overall ecosystem, or merely shift attention from one place to another?

Personalization systems are often described as recommendation engines, but the best systems are decision systems. They decide how to allocate attention across personal relevance, collective behavior, cultural momentum, and business strategy.

Getting that balance right is one of the most important challenges in modern digital product design. Done well, it makes the experience feel both personal and alive. It helps users discover more value with less effort. It gives businesses a responsible way to shape demand. And it turns promotion from a manual placement exercise into a disciplined, adaptive, user-centered capability.

---

This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.