Size Recommendation Algorithms: A Technical Deep Dive

Most size recommendation tools fail because they start with the wrong question. They ask, "What size does this shopper usually buy?" That sounds useful until you remember that every brand fits differently, every product category behaves differently, and "usually" is a weak substitute for actual measurement.

The better question is sharper: how does this shopper's body compare with this garment's size chart?

That is the question VTS is built to answer. The shopper enters height, weight, and gender. They take two full-body photos, front and side, in fitted clothing. Computer vision extracts more than 20 body measurements. The system matches those measurements against product-specific size data. The shopper receives a recommended size with a fit heatmap that shows where the garment is expected to feel snug, perfect, tight, or loose.

This is not a quiz. This is not "customers like you bought medium." This is body scanning applied to the actual fit problem on a Shopify product page.

The first generation: static size charts

Static size charts are the baseline. They list garment or body measurements and ask the shopper to decide. Chest, waist, hips, inseam, shoulders, sleeve length, and similar dimensions may all appear in the table.

The chart is not useless. It is just incomplete. It assumes the shopper has a measuring tape. It assumes the shopper knows exactly where to measure. It assumes the shopper measures consistently. It assumes the chart is accurate. It assumes the shopper understands ease, cut, stretch, and garment intent. That is a lot of assumptions for a moment that should be simple.

Static size charts also tend to flatten nuance. A shopper can be between sizes in one measurement and solidly inside another. A dress may fit at the waist but feel tight at the hips. A jacket may fit the chest but pull at the shoulders. A pair of jeans may match waist measurement and still disappoint because rise, hip, and inseam interact.

The static chart provides data. It does not provide a decision.

The second generation: rules and lookup tables

Many recommendation engines improve on static charts by adding rules. If height is within a range and weight is within a range, recommend a size. If the shopper bought medium before, recommend medium again. If reviews say "runs small," bump up one size.

Rules can be useful for edge cases, but they break down when brands, categories, and bodies vary. A lookup table is only as good as the inputs it receives. If the shopper's self-reported data is rough, the output is rough. If the product cut is unusual, the output gets weaker. If the shopper's body distribution does not match the average represented by height and weight, the recommendation can miss.

This is why height and weight alone are not enough. Two shoppers can share height and weight and have very different chest, waist, hip, shoulder, and inseam measurements. A size algorithm that ignores that difference is not really measuring fit. It is estimating.

The VTS approach still uses structured inputs, but it does not stop there. Height, weight, and gender help frame the scan. The two photos give the system the body information it needs to make the recommendation personal.

The third generation: body scanning

Body scanning changes the quality of the input. Instead of asking a shopper to describe their body, VTS uses computer vision to extract measurements from two full-body photos. The target is more than 20 measurements, including fit-critical dimensions like chest, waist, hips, inseam, shoulder width, and arm length.

The shopper experience matters here. The process has to be quick, clear, and comfortable. VTS asks the shopper to stand about 1.5 meters from the phone camera and take front and side photos in fitted clothing. The scan should take seconds, not minutes. The result should feel like useful guidance, not a medical exam.

Once measurements are extracted, the system can compare the shopper with the product size chart. That comparison is where the fit recommendation becomes concrete. The engine is no longer relying only on a shopper's memory, a generic review, or a brand-level guess. It is comparing real body data to real garment data.

This is why VTS describes AI Size Finder as the smartest fit engine on Shopify. The differentiator is not simply that it uses AI. The differentiator is that it uses AI to remove guesswork from the measurement step and then explains the recommendation visually.

Matching the body to the garment

A size recommendation is only as good as the garment data it uses. If a product's size chart is wrong, incomplete, or inconsistent, every recommendation engine has a harder job. VTS solves part of this by allowing merchants to upload size charts or let AI auto-detect from product data during setup.

The matching process has to understand that fit is not a single number. A top, a pair of jeans, a dress, and outerwear all have different fit priorities. For one product, shoulder width may be critical. For another, hip and waist interaction may matter more. For a third, length may drive shopper satisfaction.

That is why the output should not be only "buy large." A shopper deserves to know why. If the engine recommends large because medium is tight across the hips, the shopper should see that. If the fit is expected to be perfect at the waist but snug at the shoulder, that nuance matters.

The fit heatmap is the interface layer for this. It translates the algorithm's comparison into a shopper-friendly visual. Green is easy to understand. Yellow is easy to understand. Red is easy to understand. Instead of dumping measurements onto the product page, VTS turns them into fit confidence.

Confidence scoring and recommendation language

Recommendation language has to be careful. A confident recommendation can increase conversion, but overconfident language can create disappointment if the product does not match expectations. The best system gives a clear answer while showing the evidence.

VTS uses fit heatmaps and confidence scoring to do that. The recommendation is not naked. It is supported by a visual explanation. That helps shoppers trust the answer because they can see the logic behind it.

This is a major improvement over vague language like "usually fits true to size." That phrase is common because it is easy. It is also weak because it hides variation. True to size for whom? Based on what measurements? Compared with what product data? Under what body shape?

A strong recommendation should be personal, product-specific, and explainable. That is the bar AI fit technology needs to clear. If the shopper cannot understand the recommendation, they are still guessing. If the merchant cannot measure the impact, the tool is still a black box.

Virtual try-on solves a neighboring problem

Size recommendation answers fit. Virtual try-on answers appearance. The distinction matters because shoppers need both.

A technically correct size can still fail if the shopper does not like how the garment looks on their body. A product can match measurements and still feel wrong because of silhouette, styling, or personal taste. That is why VTS pairs AI Size Finder with Virtual Try-On.

The shopper can generate a photorealistic image of themselves wearing the product. Not an avatar. Not a cartoon. Their actual body in the merchant's actual clothes. This gives the shopper a visual answer to the question that size charts cannot answer: do I like this on me?

The two systems reinforce each other. Size Finder reduces the risk of ordering the wrong size. Virtual Try-On reduces the risk of ordering something the shopper cannot picture. Together, they create a stronger pre-purchase confidence loop.

That is also why VTS can talk about reducing returns by 35% through fit methodology and an additional 12% from virtual try-on on top of size recommendation alone. The systems are not redundant. They solve different doubts in the same buying moment.

Data security is part of the product

Body scanning only works if shoppers trust the process. VTS treats customer body data as sensitive. Images are processed through a secure AI pipeline and deleted within 24 hours. Measurements are stored encrypted with AES-256 and associated with anonymous IDs. Data is protected in transit with TLS 1.3. VTS does not sell or share biometric data for advertising.

This is not a legal footnote. It is part of the user experience. A shopper who is asked for body photos deserves direct language about what happens next. A merchant using VTS should be able to explain the privacy posture clearly.

The best fit algorithm in the world will fail if shoppers do not trust the scan. Security, retention, deletion, and transparency all affect completion rate. That is why the system has to be built with privacy as a core feature rather than an afterthought.

What merchants should measure

After launch, merchants should not only watch total return rate. They should watch the full fit funnel. How many shoppers start the size finder? How many complete the scan? Which products get the most scans? Which sizes are recommended most often? How do scanned sessions convert compared with unscanned sessions? Which return reasons remain after the fit tools are used?

VTS gives merchants the analytics needed to ask those questions: revenue impact, active customers using size finder and try-on, conversion rate lift, performance trends, scans completed, try-ons generated, high-demand articles, return reason analysis, and size distribution.

Those signals help the merchant improve the store. If many shoppers scan on one product but conversion stays low, the issue may be price, photography, styling, or product trust. If one size is recommended far more often than expected, inventory planning may need to change. If returns persist for a product even after scan usage, the product data may need cleanup.

The algorithm is not only a recommendation engine. It is a measurement system for shopper uncertainty.

The future of size recommendation is explainable fit

The future is not another quiz. It is not a longer size chart. It is not another "tell us your favorite brand" flow. The future is explainable fit: measure the shopper, compare the garment, show the fit, and let the merchant learn from the data.

That is what VTS is building for Shopify stores. AI Size Finder extracts more than 20 body measurements from two photos. Fit heatmaps make the recommendation understandable. Virtual Try-On shows the product on the shopper's own body. Merchant analytics turn those interactions into business insight.

The technical deep dive ends with a simple point. Sizing returns are expensive because shoppers are forced to guess. A good algorithm does not make the guess prettier. It removes the guess.

When the shopper knows their true size and can see how the product will look, the cart gets cleaner. When the cart gets cleaner, returns drop. When returns drop, margin comes back. That is the job.