E-commerce
June 28, 2026
"How to adjust the tension?" "The app won't connect." "What is the week 1 dosage?" Three problems, three different tutorials. A bot that spits out the entire help page is frustrating; a bot that sends step 4 of the wrong SKU makes it worse.
Heeya estimates that a well-chunked RAG on product documentation can deflect 55% to 72% of tier-1 contacts when routing is precise (Heeya, RAG support 2026). LaunchGPT points out that post-purchase chat benefits from combining order data and content RAG, rather than a generic widget everywhere (LaunchGPT, playbook chat 2026).
This guide #233 covers the product tutorial AI chatbot: matching customer problems with the right help. Distinct from the tutorial library (#232) (inventory) and the onboarding bot (#179) (proactive flows): here, it is about intent matching → tut_id → step in the dialogue.
Summary
Why does routing to the correct tutorial change the usage resolution?
The product tutorials chatbot does not answer in bulk: it identifies the problem, loads the correct tut_id, and proceeds step by step.
Failure of document dumping
Pasting a PDF link or an 800-word article into chat = the customer gives up and opens a ticket. GetAgent notes that 384-512 token chunks improve accuracy on step-by-step procedures (GetAgent, RAG KB 2026).
Three benefits
Accuracy: Order SKU + intent = correct tutorial
Progression: one step, confirmation, next
Adapted format: text, GIF, or video depending on the step
DTC Kitchen Example
Multifunction food processor, 76 tickets/month "does not work / soup mode". Bot routing tut_id + step: self-service 64%, FCR usage +22 points, repeat contact −31%.
How does it differ from the library, onboarding, and troubleshooting?
Five related pieces of content, one angle: bot that selects and guides the tutorial.
Library (#232)
Library (#232): taxonomy and TUT-meta. #233: how the bot consumes this catalog in conversation.
Onboarding (#179)
Bot onboarding: proactive Day-0. #233: reactive when the customer describes a blocking issue, even weeks later.
Troubleshooting (#229)
Troubleshooting (#229): TRB diagnostic tree. #233: educational guide; if TRB fails, handoff. Both follow each other.
Guided selling (#150)
Guided selling (#150): pre-purchase choice of SKU. #233: post-purchase usage of the SKU already ordered.
Promise #233
Tutorial intents, matching tut_id, step-by-step flow, multimedia, grounding, tests, KPIs, playbooks.
Which tutorial intents should be mapped for the bot?
Map the chatbot tutorial intents aligned with library #232.
Ten routing intents
tut_setup: first use, assembly
tut_routine: dose, mode, frequency
tut_care: maintenance, washing
tut_expectation: normal vs abnormal
tut_troubleshoot_light: reset, indicator light (before TRB)
tut_accessory: compatible part, consumable
tut_video_request: customer requests a visual demo
tut_step_blocked: stuck at step N in progress
tut_wrong_product: confusion with similar SKU
tut_unknown: no tut_id indexed
Verbatim mining
Exports of 90-day chat history: "how to", "user manual", "step", "video", "not working". Cross-reference with usage (#230) tags.
MVP Prioritization
Top 5 intents × top 10 TPU SKUs = 50 bot routes minimum before go-live.
How does the bot match the customer issue, tut_id, and step?
The tutorial matching bot follows a four-pass pipeline, not a single RAG prompt.
Matching pipeline
Order lookup: SKU, variant, delivery date
Classify intent: tut_* from verbatim
Resolve tut_id: metafield index or
sku→tutorials[]tablePick step: step 1 or resume session
last_step
Disambiguation rules
2 similar SKUs in history → ask which one
Vague intent setup vs troubleshoot → binary question
Customer mentions step N → jump to chunk step_N
Routing example
"The app is not connecting" + speaker SKU → intent tut_setup → tut_id PAIR-APP-X → step_1 (download app).
Fallback
No tut_id: TUT-UNKNOWN + help center link + handoff if urgent.
Routing JSON example (metafield)
{"sku":"SPK-100","match_rules":[{"keywords":["app","connect","pair"],"tut_id":"TUT-PAIR-SPK100","intent":"tut_setup"}]}. Priority: active order > keyword > fallback collection.
Which RAG sources and indexes for bot tutorials?
The RAG tutorials index powers grounded retrieval and citations.
Sources
Library chunks #232 (step + TUT-meta)
Shopify metafields
tutorial_library_refVideo URLs + transcript (#231)
Associated error codes tut_troubleshoot_light
Chunk schema
Each chunk: tut_id, sku, intent, step_n, action, media_url, version. Heeya: 400-600 tokens, 15% overlap, metadata for SKU filtering (Heeya).
Hybrid retrieval
Vector + SKU/error code keyword. Reranker if > 1000 chunks. Filter retrieval by sku=order.sku before similarity search: greatly reduces confusion between close variants of the same model and speeds up the response. See maintain knowledge base and train Shopify bot.
Which step-by-step tutorial conversational flow?
The chatbot tutorial flow progresses one step at a time with confirmation.
Standard sequence (6 turns max)
Greeting: "I see [SKU]. What is your issue: setup, usage, maintenance, or breakdown?"
Confirm tut_id + estimated duration
Step 1: action + Done / Blocked / Watch video buttons
If Done → step 2; if Blocked → sub-step or photo
Mid-flow tip or common mistake
Closing: "Resolved?" or TRB / handoff routing
Chat UX
Progress bar Step 2/5
Session resume via email magic link
"Restart tutorial" button
Additional onboarding
Same flow reusable proactively on Day 0 (#179) with pre-selected tut_id.
Full thread example (cosmetic device)
Customer: "How often should I use the serum?" → tut_routine → tut_id ROUTINE-SERUM-A → step_1 dose 2 drops → Done → step_2 frequency 2×/week → Done → tip SPF morning → inline CSAT closing.
How to combine text, video, and handoff in the feed?
The tutorial multimedia bot chooses the format by step, not by agent preference.
Format rules
Text: short policy, dose reminder
GIF: one click, UI toggle
Embedded video: ≥ 3 gestures or assembly (#231)
Bot templates
TUT-BOT-STEP: "Step [N]: [action]. [GIF/video if available]. Press Done when completed."
TUT-BOT-VIDEO: "Here is the 45s demo for this step: [player]. Transcript available below."
Handoff
After 2× Blocked on the same step or failed tut_troubleshoot_light intent → TRB (#229) or human with log tut_id + steps_done. Storebird insists: handoff without context is worse than no bot (Storebird, catalog RAG 2026).
What are the anti-hallucination rules for tutorial procedures?
The grounded tutorial bot never invents a step missing from the indexed chunk.
Guardrails
Whitelist actions by tut_id + step_n
Quote tutorial version in response
TUT-UNKNOWN if chunk confidence < threshold
No mixing of steps between SKU A and SKU B
System Prompt
"Answer only from the provided chunks. If a step is missing, say you are checking and offer a handoff." See anti-hallucination (#123).
Version Sync
Tutorial v1.2 published → reindexed within 24 hours. Bot refuses to quote v1.1 if deprecated flag is set.
How to test the routing tutorial before going to production?
The QA tutorials bot validates matching and trajectory, not an isolated response.
Test suite (30 scenarios)
10 correct tut_id routes by SKU+intent
5 steps N resumed in progress
5 similar SKU disambiguations
5 TUT-UNKNOWN → clean handoff
5 tut → TRB sequences
Shadow mode
Bot suggests tut_id + step, agent validates for 2 weeks. Measure discrepancies → correct index.
RAG Golden set
Heeya recommends 50-100 representative questions + faithfulness metrics. Target Precision@5 > 0.8 on tutorial chunks.
Which KPIs should be used to manage the tutorial bot?
Measure the tutorial bot by guided resolution, not by chat volume.
Routing KPIs
Tut_match accuracy: correct tut_id first try
Flow completion rate: final step reached
FCR tut_* without human
TUT-UNKNOWN rate: library gap
CS impact KPIs
TPU usage post-deployment
7-day repeat contact after tut flow
CSAT segment tut_guided
Monthly loop
Top TUT-UNKNOWN → create tutorial #232. Top Blocked step → enrich chunk or video. See Chatbot KPIs (#11).
Weekly dashboard
Columns: intent, tut_id, match OK/KO, max step reached, FCR, UNKNOWN. Sort UNKNOWN desc = content backlog for the week.
How does Qstomy route to the right product tutorial?
Qstomy exécute le pipeline matching SKU → intent → tut_id → step avec corpus bibliothèque #232.
Fonctionnalités tutoriel bot
Lookup commande Shopify + tut index
Intents tut_* + flow pas à pas
Embed vidéo/GIF par step metadata
Reprise session last_step persisté
Handoff dossier tut_id, steps, version
Export gaps vers roadmap #232
Scénario DTC chiffré
Marque skincare devices, 27 % chats post-achat usage, tut_match accuracy initiale 58 %. Index 42 tut_id chunkés + 10 intents + shadow 2 sem. Après 10 semaines : tut_match 87 %, FCR tut_* 71 %, TPU usage −29 %, complétion flow 68 %.
Explorez intégration Shopify, support client IA, demander une démo.
Which operational playbooks should be launched this week?
Playbook 1 : prérequis bibliothèque (2 j)
Minimum 20 tut_id indexés (#232) avec TUT-meta complet avant bot routing.
Playbook 2 : intents + pipeline (1 jour)
Mapper 10 intents section 3, règles matching section 4, test 10 verbatims.
Playbook 3 : flow + templates (4 h)
TUT-BOT-STEP, barre progression, reprise session, embed vidéo.
Playbook 4 : shadow + golden set (2 semaines)
30 scénarios section 9, ajuster index, bascule 25 % trafic usage.
Playbook 5 : revue KPI mensuelle (1 h)
TUT-UNKNOWN, étapes Bloqué, sync version tutoriels.
Maillage utile
Un bot tutoriel efficace ne remplace pas votre bibliothèque : il la rend conversationnelle. Quand chaque problème client trouve le bon tut_id, la bonne étape et le bon format en quelques tours, le support cesse d'être un moteur de recherche manuel et devient un guide qui sait où vous en êtes.
Enzo
June 28, 2026
