E-commerce
June 28, 2026
The customer opens the best-selling dress. Twelve colors, six sizes, seventy-two combinations. They type in the chat: "Between midnight blue and navy, which one leans more toward black?" "I usually wear a size 38 with you, how does this cut fit?" A static selector doesn't answer. A generic bot invents.
Heeya shows that a pre-purchase assistant can reduce variant-related returns by 20 to 35% when it guides the decision in 3 to 5 contextualized questions about the open SKU (Heeya, pre-purchase returns 2026). Text.com confirms that size/color selection directly within the chat widget, with add-to-cart, completes the purchase journey without leaving the conversation (Text.com, variant in chat 2026).
This guide #242 covers the AI chatbot centered on product variants: color, size, format, compatibility. A new use case distinct from guided selling (#150) (product choice) and variant error reduction (#241) (PDP/CS ops): here, the AI recommends and pre-selects the right variant on the PDP.
Summary
Why a chatbot dedicated to variants on the PDP?
The PDP variant chatbot comes into play when the SKU is already selected but the option is not. This is the moment when wrong-size and wrong-color returns are decided.
Three jobs of the variant bot
Clarify: color rendering, format dimensions, model compatibility
Recommend: size, SKU variant, with a traceable reason
Execute: selector pre-selection or add-to-cart of the correct variant
Market signal
AeroChat estimates that incorrect Shopify orders mainly stem from unclear sizing, confusing color variants, and rushed mobile decisions (AeroChat, wrong orders 2026). Zipchat observes 30 to 45% fewer sizing returns when the AI asks qualifying questions and then maps the size guide (Zipchat, size advisor 2026).
DTC Cosmetics Example
Serum with 4 formats (30/50/100 ml), 38 tickets/month asking "which format to choose?". Format variant bot: 2 usage + duration questions → recommends 50 ml + add-to-cart. Format tickets −61%, pilot PDP conversion +9% in 6 weeks.
How does it differ from neighboring content?
Four closely related contents, four levels of granularity.
Guided selling (#150)
Guided selling (#150): guide towards the right product in a catalog. #242: a variant on an already opened product.
AI size recommendation (#199)
AI Sizing (#199): fit engine architecture, measurements mapping. #242 extends to color, size, compatibility with unified variant flows.
Reduction of variant errors (#241)
Ops variant (#241): PDP audit, customer service tags, feedback loop. #242: conversational AI layer that complements the PDP, rather than replacing it.
Human size support (#128)
Size support (#128): agent macros. #242 automatizes the variant advice; #128 manages the escalation.
Promise #242
Variant data by axis, intent flows, PDP triggers, Shopify stock sync, pre-selection, handoff, KPIs, playbooks.
What catalog data should the bot read?
A reliable variant bot never guesses: it reads structured fields by SKU and by option.
Native Shopify Layer
variants[]: id, title, option1/2/3, sku, price, inventory_quantity
options[]: axis names (Color, Size, Format)
images linked by variant_id for color rendering
Recommended Metafields per Axis
Size: fit_note, size_chart, model_size, fabric_stretch
Color: color_description, undertone, material_finish, swatch_hex
Format: capacity_ml, dimensions_cm, usage_duration_days, price_per_unit
Compatibility: compat_models[], voltage, connector_type
Structure via Shopify product and variant metafields (Shopify, metafields). See product support data flow (#198).
Feedback Layer
Reviews tagged by variant, Loop wrong-size/color returns, var_* tickets from #241. Feeds offsets ("73% reviews: runs small on this SKU").
How to structure flows by intent variant?
Four distinct bot variant flows, one common SKU resolution engine.
Size flow (4 questions max)
Bot reads fit_note + product_type (skipped if obvious)
"What size do you wear in [category] with us?"
Adjusted vs comfortable preference (skipped if fit_note is clear)
Between two sizes? → tie-break rule + exchange plan B
Output: size variant + reason + pre-selected selector. Engine detail: #199.
Color flow (2-3 questions)
"Are you looking for neutral, bright or dark?"
"Do you often wear [similar shade]?" → swatch map
Displays variant image + color_description + "true to color" review
Format flow (2 questions)
Usage: daily, travel, family, pro?
Desired duration or budget constraint → compare capacity + price/unit
Compatibility flow (3 questions)
"Which model / device do you own?"
Lookup compat_models[] → match or no match
If no: alternative compatible variants or handoff
Unification rule
Intent detected in NLU → flow branch. Multi-axis product: resolve color first, then size, never the other way around if color changes the decision-making visual.
How do you trigger the bot at the right time on the PDP?
The variant trigger bot must capture hesitation, not interrupt reading.
Five behavioral signals
Selector dwell: 25-45 s without variant click (Heeya: 45-90 s depending on category)
Repeated toggle: 3+ color/size changes in 60 s
Size guide click without add-to-cart within 90 s
Scroll back up after viewing selector (mobile)
Language intent: "between two sizes", "which color", "compatible with"
Contextualized opening messages
Size: "This cut [fit_note]. I can recommend a size in 30 seconds."
Color: "[N] colors available. Hesitating between two shades?"
Compat: "Enter your model, I will check compatibility before ordering."
What not to do
Immediate popup upon PDP arrival, generic message "How can I help?", trigger on all product pages including single-variant SKU.
Widget placement
Sticky bottom right mobile, inline under variant selector on desktop if the theme allows. Dori calls this pattern "Product Pal" on PDP (Dori, Product Pal 2026).
How to pre-select a variant in Shopify?
The bot variant pre-selection closes the loop: recommendation → action, no recommendation → confusion.
Three levels of execution
Level 1: "Choose size M" message + selector anchor link
Level 2: JS theme event: programmatic option selection + gallery update
Level 3: direct add-to-cart variant_id in the chat (Text.com)
Shopify API required
Before add-to-cart: retrieve exact variant.id via selected options. MSG91 insists: check live stock and capture the correct variant_id before cart (MSG91, variant ID 2026).
Sticky post-recommendation recap
"You chose: [Color X] / [Size M] based on assistant advice. [Modify] [Add to cart]." Reduces add-to-cart on forgotten default variants.
Mandatory QA testing
15 color×size combinations on 3 pilot SKUs, desktop + mobile. Verify: gallery sync, updated price, OOS stock blocked, correct cart line item.
Which stock and availability rules should be hardcoded?
The stock variant guardrail prevents the bot from recommending unavailable items.
Non-negotiable rules
inventory_quantity = 0: never recommend; suggest adjacent size/color or back-in-stock alert
Impossible combination: do not suggest color X + size Y if SKU does not exist
API Refresh: re-query stock before add-to-cart, not only at the beginning of the flow
Pre-order: if pre-order variant, communicate explicit lead time before validation
Out-of-stock phrasing
"Size M in Midnight Blue is out of stock. M in Sky Blue is available (2-3 days delivery). Size L Midnight Blue: estimated back in stock [date]." Three concrete options, not "come back later."
Smart bracketing
If the customer hesitates between S/M and M is out of stock: recommend S + free exchange reminder rather than encouraging double-size ordering.
How to handle edge cases and handoff?
Bot variant edge cases require confidence thresholds and contextual handoff.
Handoff Threshold (< 70% confidence)
No size_chart or fit_note on SKU
Request for atypical morphology, pregnancy, pro sports
External brand reference without a cross-brand table
Model compatibility missing from compat_models[]
3 turns without variant resolution
Agent Transfer
Transcript + SKU + options already explored + measurements collected. See bot → human context transfer. Agent does not ask the same questions again.
Disclaimers
"Recommendation based on size guide and customer reviews. Free exchange within 30 days if the variant does not fit." Reduces disputes and aligns expectations on color/format.
Monthly Audit
50 variant conversations: recommended variant traceable, zero invented SKUs. Chatbot response audit.
What anti-patterns cause variant bots to fail?
Five variant bot errors generate returns and loss of trust.
LLM alone without live catalog
The model recommends "M" without reading the fit_note "runs small" or stock. 2026 Pattern: LLM interface + variant rules engine.
Identical flow for all axes
Asking for measurements to choose between red and burgundy. Intent detection first.
Recommendation without execution
"Take L" but selector remains on S by default. Pre-selection or add-to-cart mandatory.
Ignoring mobile
8-question flows unreadable on small screens. Max 4 questions, quick-reply buttons.
Zero feedback loop
Bot deployed, wrong-variant returns continue on the same SKU, no fit_note adjustment. Connect Loop + tickets #241.
Which KPIs should be used to measure the impact of the bot variations?
Measure the bot variant ROI by intent and pilot SKU, using A/B testing if possible.
Primary KPIs (30-90 days)
Add-to-cart conversion: bot variant vs. control PDP sessions
Wrong-variant return rate: engaged bot sessions vs. non-engaged
Var_* tickets / 100 SKU orders: delta vs. baseline #241
Bracketing rate: multi-size orders of the same SKU
Handoff rate: % of flows escalated to a human agent
Quality KPIs
Variant accuracy: "bot advised me poorly" feedback / bot sessions
Time-to-recommendation: target < 90 s on mobile
Flow completion rate: % of sessions reaching recommendation
A/B Methodology
50% of pilot PDP traffic with bot variant triggers, 50% without. 4 weeks, min. 1,500 sessions/arm. Note: sizing gains visible from W+4 on fit-sensitive categories.
How does Qstomy handle variant selection on Shopify?
Qstomy combines contextual variant flows, Shopify catalog sync, and cart fulfillment.
Variant selection features
Intent var_size / var_color / var_format / var_compat on PDP
Real-time reading of variants[] + metafields
Traceable recommendation: variant_id + reason + image
Selector pre-selection or in-chat add-to-cart
Inventory guardrail before any validation
Transcript handoff to Gorgias if confidence is low
Quantified DTC scenario
Fashion + tech brand, 4,200 variant-heavy PDP sessions/month, 26% variant pre-purchase tickets, selectors alone being insufficient. Deployment of Qstomy variant flows on 35 SKUs + section 5 triggers + inventory sync. After 8 weeks (50/50 A/B): pilot PDP conversion +14%, bot sessions wrong-size/color returns −27%, var_* tickets −31%, handoff 9%, flow completion 78%.
Explore Shopify integration, AI customer support, request a demo.
Which operational playbooks should be deployed in 30 days?
Playbook 1: audit data variant (1 day)
20 top revenue SKU variant-heavy. Verify section 3 metafields by axis. Complete missing fit_note, color_description, compat_models.
Playbook 2: flows 4 intents (3 days)
Draft section 4 decision trees, recommendation templates, disclaimers. Test 20 scenarios: between two sizes, close color, budget format, unknown compatibility.
Playbook 3: PDP triggers (1 day)
Configure dwell, toggle, messages in section 5. Exclude mono-variant. Shadow mode for 1 week without auto add-to-cart.
Playbook 4: stock sync + pre-selection (2 days)
Connect inventory API, QA 15 combinations in section 6. Activate sticky recap.
Playbook 5: A/B pilot + feedback loop (4 weeks)
Measure section 10 KPIs. Merge Loop wrong-variant feedback. Adjust fit_note and color_description for problematic SKUs. Share support wins.
Useful links
Choosing a variant is not about ticking a box: it is about resolving an uncertainty. When the chatbot reads your catalog, asks the right questions, and executes the correct option, the PDP becomes a sales advisor available at midnight, with no support ticket the next day.
Enzo
June 28, 2026
