Glossary

Adaptive Learning

In one line: Adaptive learning is AI that personalizes a learner’s path — adjusting difficulty, topics, and pacing based on how they’re actually doing.

Why education retailers should care

If you sell digital learning products — courses, workbooks with companion apps, exam prep tools — adaptive learning is the feature that increasingly separates premium products from commodity ones. Learners (and parents) pay more for content that adjusts to them than for content that doesn’t.

How it works

The system tracks each learner’s responses — right, wrong, time taken, confidence. It builds a model of what they know and what they don’t. Then it serves the next question, lesson, or exercise calibrated to where they are: hard enough to challenge, not so hard they give up. The path through the content is unique to each learner.

Where AI adds beyond traditional adaptive learning

• Generative explanations — when a learner gets stuck, AI generates a fresh explanation tuned to their level
• Plain-language tutoring — AI assistants now hold conversations with learners about the material
• Parent and teacher dashboards — surface what’s actually being learned, not just time spent

Related terms

Personalization, Recommendation Engine, Generative AI

Glossary

Inventory Optimization

In one line: Inventory optimization uses AI to decide how much of each item to hold, where, and when to reorder — balancing stockouts and overstock continuously.

More than forecasting

Demand forecasting predicts what you’ll sell. Inventory optimization decides what to do about it — how much to order, when, from which supplier, allocated to which location. The two are related but distinct: a perfect forecast doesn’t help if your reorder logic isn’t sound.

What it accounts for

• Supplier lead times (some take 2 days, some 6 weeks)
• Minimum order quantities and case-pack sizes
• Storage costs and shelf life
• Service level targets (95% in-stock vs 99% costs very different)
• Promotion calendars that will spike demand
• Allocation between multiple locations or channels

What it achieves

Typical wins: 15–30% reduction in stockholding cost, fewer stockouts, less write-off of expired or obsolete stock. For retailers, the working capital freed up by smarter inventory often funds the next investment in the business.

Related terms

Demand Forecasting, SKU-Level Forecasting, Dynamic Pricing

Glossary

Virtual Try-On

In one line: Virtual try-on uses AI and your phone camera (or an uploaded photo) to show how clothing, eyewear, or accessories look on you before you buy.

Where it works well

Accessories — eyewear, watches, jewelry, hats — are mature use cases. The AI maps the item onto your face or hand with high accuracy, lift in conversion of 20–40% is typical. Outerwear and shoes work reasonably well. Body-fit garments (dresses, tailored pieces) are still maturing — the visual fidelity isn’t yet good enough to fully replace trying on, but it helps with style judgment.

Why it matters

The two biggest reasons shoppers don’t buy online are “I’m not sure how it’ll look on me” and “I’m not sure of the size.” Virtual try-on directly addresses the first. The result is higher conversion, fewer abandoned carts, and lower return rates — especially for accessories where sizing isn’t a factor.

Related terms

Computer Vision, Personalization, Recommendation Engine

Glossary

Computer Vision

In one line: Computer vision is AI that interprets images and video — recognizing products, counting customers, reading shelves, and detecting patterns in visual data.

What it does in retail

Modern computer vision turns CCTV and product images into actionable data. It can count how many shoppers entered today, recognize when a shelf needs restocking, identify which products customers picked up but didn’t buy, flag damaged packaging on receiving docks, or match a customer photo to similar items in your catalog.

Common use cases

• Footfall analytics — count visitors, time spent, dwell zones
• Shelf monitoring — detect out-of-stock items in near real time
• Visual product search — “find something like this” for shoppers
• Loss prevention — flag suspicious patterns at self-checkout
• Quality inspection — spot damaged stock before it reaches the floor

What you need

A camera setup — existing CCTV often works — and a service that processes the feed. Most modern offerings are cloud-based and don’t require new hardware. Privacy and consent rules apply, especially in physical stores.

Related terms

Virtual Try-On, Personalization, Inventory Optimization

Glossary

Customer Lifetime Value (CLV)

In one line: Customer Lifetime Value is the total revenue you expect from a customer over their entire relationship with you. AI makes it predictive instead of historical.

The traditional way

Old-school CLV is a backward look: average order value × purchase frequency × expected lifespan. It tells you what past customers were worth, on average. Useful for benchmarking, useless for decisions about an individual customer in front of you right now.

The AI way

Predictive CLV uses each customer’s actual behavior — what they bought, how often, what they returned, when they last engaged — to forecast their individual future value. The same model also estimates churn risk: how likely each customer is to never come back.

What you can do with it

• Spend marketing dollars smarter. Pay more to acquire customers who’ll be worth more.
• Identify VIPs early. Treat high-predicted-value customers like the gold they are — not just the ones who’ve already spent the most.
• Win back at-risk customers. Intervene with churn-risk customers before they’re gone.
• Tier loyalty programs intelligently. Base perks on expected future value, not just past spending.

What you need

Transaction history per customer (at minimum 12 months), enough customers to establish patterns (typically thousands), and a way to identify the same customer across visits — usually a loyalty program, account login, or unified email/phone matching.

Related terms

Personalization, Recommendation Engine, Demand Forecasting

Glossary

Cold-Start Problem

In one line: The cold-start problem is what happens when an AI system has no history to learn from — new products, new customers, new stores. Solvable, but worth understanding.

Where it shows up

You launch a new product — the recommendation engine doesn’t know who’ll like it. A new customer signs up — the personalization engine has nothing to personalize on. You open a new location — the forecasting model has no sales history. In all three cases, the AI is starting from zero data.

How good systems handle it

• Borrow from similar items. A new “navy medium” t-shirt? Predict based on existing similar SKUs.
• Use content features. For new products, the model uses category, brand, price, and description — not just purchase history.
• Bootstrap from a peer store. A new location forecasts based on the demographically closest existing store.
• Default to popular. Show new shoppers the bestsellers until you have enough behavior to personalize.

Why it matters in evaluations

When evaluating AI tools, ask explicitly: “how does this handle new products / new customers / new locations?” If the answer is “wait for data,” the tool will quietly fail you every time you launch something new. A serious system has a strategy.

Related terms

Recommendation Engine, Demand Forecasting, SKU-Level Forecasting

Glossary

SKU-Level Forecasting

In one line: SKU-level forecasting predicts demand for each individual product variant — not just category totals. It’s the difference between useful and useless inventory data.

Why the granularity matters

“We’ll sell 200 units of t-shirts next week” is interesting but unhelpful. Which colors? Which sizes? Without that, you can’t order intelligently. SKU-level forecasting answers: 47 black mediums, 32 white larges, 18 navy smalls, and so on for every variant.

Where it’s critical

• Clothing — size curves vary dramatically; getting them wrong means lost sales and excess markdown
• Electronics — model variants, storage capacities, colors all sell differently
• F&B — specific ingredients and dishes, not just “total covers”
• Education — specific titles by syllabus level, not generic “math books”

The technical challenge

SKU-level forecasting is computationally harder because there’s much less history per item (a new product variant might only have a few weeks of sales). Good models handle this by sharing patterns across similar SKUs — learning that a new “navy medium” will probably behave like the average of similar items, even without its own history.

What it enables

• Precise reorder quantities and timing
• Smarter allocation between stores
• Earlier markdown decisions on underperforming variants
• Better buying decisions for next season

Related terms

Demand Forecasting, Inventory Optimization, Cold-Start Problem

Glossary

Dynamic Pricing

In one line: Dynamic pricing adjusts product prices in near real-time based on demand, competitor moves, inventory levels, and other signals.

Not just “sales when stock is low”

The simplest version is a markdown rule (“drop price 15% if inventory still holds 30 days from end of season”). The AI version is far richer: it watches competitor pricing, demand elasticity for each product, your margin position, and inventory pressure — and recommends repricing windows that protect margin and maximize sell-through.

Where it works well

• Electronics — commoditized products with visible competitor pricing
• Fashion — seasonal markdown cycles, end-of-season clearance
• F&B delivery — daypart-based pricing, surge offers
• Perishables — price-to-clear before waste

Where to be careful

Customers notice when prices change between visits and they don’t love it. The retailers who use dynamic pricing well are transparent about it, change gradually (not minute by minute), and avoid using personal data to charge different shoppers different prices for the same product — which is legally risky and trust-destroying.

Related terms

Demand Forecasting, Inventory Optimization, Personalization

Glossary

Personalization

In one line: Personalization in retail means tailoring what each customer sees — products, offers, messaging, even pricing — based on their individual behavior and preferences.

Beyond “hi {{first_name}}”

Real personalization isn’t just putting someone’s name in an email. It’s sending each shopper the offer they’re most likely to act on, surfacing products they’re likely to want, and adjusting the experience as their behavior changes. AI makes this possible at scale — doing for thousands of customers what a great shopkeeper does for the regulars they know personally.

Where it shows up

• Product recommendations on the website, in cart, after checkout
• Email and SMS campaigns targeted by behavior and life-cycle stage
• Search results ranked differently per shopper
• Dynamic pricing and offer timing
• Loyalty rewards tuned to individual habits

The tradeoffs

Personalization done well drives conversion and loyalty. Done poorly — or in the wrong contexts — it feels invasive. The line moves with category, geography, and trust. The retailers winning here treat customer data as a privilege, are transparent about how it’s used, and let customers opt out without penalty.

Related terms

Recommendation Engine, Customer Lifetime Value, Dynamic Pricing

Glossary

Recommendation Engine

In one line: A recommendation engine suggests products to a customer based on what they (and similar customers) have looked at, bought, or shown interest in.

How it works

Most recommendation engines fall into three families. Collaborative filtering recommends what similar customers bought (“people who bought X also bought Y”). Content-based recommends items similar to ones the customer already engaged with (“you liked this jacket, here are others with similar attributes”). Hybrid systems blend both, plus business rules like inventory level, margin, or compatibility.

Why generic engines underperform in specialist retail

Off-the-shelf widgets trained on general e-commerce data often miss what makes your category unique. In electronics, that’s compatibility. In clothing, that’s sizing and style. In education, that’s curriculum fit. The lift you get from a properly tuned engine is dramatically larger than from a generic one.

What you need to use it

• A clean product catalog with structured attributes (specs, tags, categories)
• Customer behavior data — views, cart adds, purchases, and ideally returns
• A way to display recommendations on product pages, cart, and email

Related terms

Personalization, Cold-Start Problem, Customer Lifetime Value