How AI Spots Undersold Products and Automatically Boosts Their Visibility (And Revenue) 📈

By Dr. Julie Jones, PhD in Artificial Intelligence

If you have ever managed an e-commerce catalog, you have likely noticed a quiet but expensive problem: not every product gets an equal share of attention. Some SKUs sit in the right category, have decent margins, and even positive reviews, yet they barely move. Meanwhile, a smaller set of hero products receives most of the traffic, most of the recommendations, and most of the revenue.

This is not just a merchandising issue. It is a visibility problem. And it is one that modern AI systems are increasingly good at solving.

The good news is that you do not need a data science team to fix it. You do not need to manually audit hundreds of listings every week. What you do need is a system that can observe your catalog, understand which products are underperforming relative to their potential, and then take action by adjusting placement, promotion, and recommendation logic in real time.

That is exactly what AI can do.

The Hidden Cost of Uneven Product Visibility

In most online stores, traffic is not distributed evenly. A classic pattern shows that a small number of products capture a large share of sales. In retail analytics, this is often described using a Pareto-like distribution: roughly 20% of products may drive 80% of revenue.

The problem is that "potential" is not the same as "performance."

A product can be well-priced, well-photographed, and genuinely in demand, yet still be undersold simply because shoppers never see it. It might appear too low in search results. It might be missing from key recommendation slots. It might not be cross-sold at the right moments.

In other words, the product is not failing. The distribution system is failing to represent it fairly.

This creates a compounding cost:

• Lost revenue: Undersold products do not generate the sales they could have.

• Inefficient inventory: Slow-moving SKUs tie up capital and warehouse space.

• Weaker customer experience: Shoppers see a narrow set of options and may leave to competitors.

• Poor data feedback loop: Low visibility means fewer interactions, which makes it harder to understand true demand.

The result is a self-reinforcing cycle: products that are seen get more data, more tuning, and more visibility, while products that are not seen stay under the radar.

AI can break that cycle.

How AI Detects Undersold Products

At its core, detecting undersold products is a problem of comparison. The system needs to ask: Given what we know about this product, what level of sales or engagement would we expect? And how does actual performance compare?

That sounds simple, but in practice it requires several layers of analysis.

1. Baseline Expectation Modeling

The first step is to estimate a reasonable baseline for each product. This is not just "average sales across the catalog." A good model considers:

• Category-level demand

• Price point relative to category

• Seasonality

• Product age (new products need a ramp-up period)

• Listing quality (images, titles, descriptions)

• Historical performance of similar products

With this baseline, the system can identify products that are performing below expectations. A product with a $40 price point in a high-demand category, with a well-optimized listing, but only 12 sales per week, is a strong candidate for visibility boosting.

2. Visibility-Outcome Correlation

Not all products are underperforming because of visibility. Some are underperforming because of price, quality, or fit. So the system also needs to estimate how much of the underperformance is attributable to visibility versus other factors.

This is typically done with a feature-based model that correlates:

• Impression count

• Click-through rate

• Add-to-cart rate

• Conversion rate

• Recommendation slot frequency

• Search position

If a product has a strong conversion rate but low impressions, the likely bottleneck is visibility. If it has good impressions but low clicks, the issue may be listing quality or image appeal. If it has good clicks but low conversions, the issue may be price or product-page experience.

This diagnostic layer matters because it tells the system what to fix, not just that something is wrong.

3. Opportunity Scoring

Once the system knows which products are undersold and why, it can assign an opportunity score. This score estimates the expected revenue uplift if visibility is improved.

A simple formulation might look like:

Opportunity = (Expected_Sales − Actual_Sales) × Average_Margin

Where Expected_Sales is the baseline from the expectation model. The system can then rank products by opportunity and focus its boosting actions on those with the highest expected return.

How AI Automatically Boosts Visibility

Detection is only half the job. The other half is action. And this is where automation becomes powerful.

Dynamic Search Ranking

Search is the single biggest traffic driver for most e-commerce sites. A product's position in search results determines how many shoppers see it.

An AI-driven ranking system can adjust product scores in real time. Instead of a static ranking formula, the model dynamically weights factors like:

• Relevance to the query

• Expected conversion probability

• Margin contribution

• Inventory level

• Product freshness

• Cross-sell potential

When the system detects that a product is undersold but has strong expected performance, it can nudge its search rank upward. The product appears higher in results for relevant queries, which increases impressions, which increases sales, which generates more data, which refines the model.

This creates a positive feedback loop.

Intelligent Recommendation Slots

Recommendation engines are the second major visibility channel. "Customers also bought," "Frequently bought together," and "You may also like" slots are high-intent placements where visibility directly drives revenue.

An AI recommendation system can do more than simply find similar products. It can optimize recommendations for business goals:

• Revenue maximization: Recommend products with the highest expected revenue per slot.

• Margin maximization: Recommend higher-margin products where possible.

• Inventory balancing: Push recommendations toward products that need to move.

• Discovery boosting: Give more recommendation slots to undersold but promising products.

The system learns which products perform well in which recommendation contexts. A product that converts well in "frequently bought together" but poorly in "you may also like" gets placed accordingly.

Targeted Promotional Allocation

Not every product needs a discount to move. But for some, a small, well-placed promotion is the most efficient lever.

An AI system can analyze:

• Price sensitivity by product and customer segment

• Historical promotion response rates

• Inventory aging

• Margin headroom

Then it can decide which products to promote, at what depth, for how long, and to which customer segments. This is far more efficient than blanket "20% off everything" sales, which erode margin and train customers to wait for discounts.

Listing Optimization

Sometimes the bottleneck is not placement but presentation. An AI system can analyze listing quality signals:

• Title keyword coverage

• Image quality and count

• Description completeness

• Attribute completeness

• Review sentiment and volume

Products with weak listings get flagged for optimization. In some systems, the AI can even generate improved titles and descriptions, suggest image ordering, or recommend attribute completions.

Measuring the Impact

A visibility-boosting system is only as good as its ability to measure what it changes. The key metrics are:

A healthy system should show revenue becoming more evenly distributed across the catalog over time, without sacrificing overall conversion quality.

A Practical Example

Consider a mid-size outdoor gear retailer with 2,400 active SKUs.

Before AI-driven visibility optimization:

• Top 10% of products generate 68% of revenue

• 340 products have fewer than 5 sales per week

• Average margin across catalog: 32%

After 8 weeks of AI-driven visibility boosting:

• Top 10% of products generate 54% of revenue

• Products with fewer than 5 sales per week: 112

• Average margin across catalog: 34%

• Total weekly revenue up 11%

The revenue concentration dropped, meaning more products are now contributing. The margin improved slightly because the system boosted visibility for higher-margin products and avoided over-discounting.

The total revenue increase was not just from selling more of the same products. It came from making the catalog work as a system, where each product gets a fair and data-driven share of visibility.

Why This Matters More Than Ever

Catalogs are getting larger. Customers are more selective. And the cost of customer attention is rising. In that environment, a system that treats every product as a potential revenue source, and allocates visibility accordingly, has a structural advantage.

AI does not replace merchandising judgment. It augments it. The human team still decides on brand positioning, pricing strategy, and customer experience. The AI system handles the continuous, granular, data-intensive work of making sure the right product is in front of the right shopper at the right time.

That combination is where the revenue comes from.

The Bottom Line

Undersold products are not a sign of weak demand. They are a sign of imperfect distribution. And distribution is a problem that AI is uniquely equipped to solve at scale.

By modeling expected performance, diagnosing visibility gaps, scoring opportunities, and automatically adjusting search, recommendations, and promotions, an AI-driven system turns the catalog from a static shelf into a dynamic revenue engine.

The products are already there. The demand is already there. What was missing was the mechanism to connect them.

Now there is one. 💡

Dr. Julie Jones