E-commerce
July 22, 2026
"I'm stuck at step 4." "Which screw for crossbar B?" "I'm afraid of breaking it if I force it." Three mid-assembly messages where a PDF link is not enough: the customer needs a conversational guide, one step at a time, with confirmation before continuing.
SharkNinja has deployed an unboxing agent that guides step-by-step and displays videos when visuals help more than a paragraph (PYMNTS, unboxing agent 2026). Salesforce documents a targeted 30% self-resolution rate on restructed manuals designed for intelligent chunking (Salesforce, SharkNinja 2026).
This guide #354 formalizes the product assembly AI chatbot: guiding with caution, without over-promising. It completes assembly errors (#353) (policy liability ops) and tutorial bot (#233) with the focus on step-by-step guided assembly AI use cases and safety guardrails.
Summary
Why automate bot-guided assembly?
An assembly bot does not replace the manual: it unrolls it in a dialogue, verifies understanding, and escalates before a customer error becomes an after-sales dispute.
Three limitations of the PDF alone
Cognitive overload: 20 steps at once, the customer loses track
No feedback: original customer doesn't know if they oriented the part correctly
No adaptation: SKU revision or mid-step blockage ignored
Fusion CX points out that post-delivery home appliance contacts are often due to an assembly discrepancy, not a product defect (Fusion CX, installation 2026).
Angle #354
#353 documents policy ASM-LIABILITY and photo evidence. #233 routes general usage tutorials. #354 defines the dedicated step-by-step assembly bot: one step, confirmation, caution, handoff if damaged.
Prudence Principle
The bot never says "force it" without quoting the manual. It asks for visual confirmation before critical steps. It hands off if there is a broken part or liability dispute.
DTC Example
Flat-pack furniture brand, 112 assembly tickets/month, PDF sent repeatedly. After ASM-BOT step-by-step bot: 68% self-guided assembly, assembly_bot_resolution 71%, assembly_dispute_rate 6%, assembly bot CSAT 4.6/5.
Journey Stage
Day 0 to Day +3 post-delivery: peak assembly_bot sessions. Trigger: packaging QR, Day +0 delivery message, or if customer opens chat with "I am assembling the product."
Cost without guided bot
VSight cites about 68% of No Trouble Found returns after setup failure (VSight, visual support 2026). Step-by-step guidance reduces NTF before an agent ticket is created.
Peak assembly bot sessions
Bot volume rises 72 hours post-delivery and over the weekend. Plan parallel #353 handoff capacity if launching a new flat-pack SKU. A bot without a complete tut_assembly_id on a new product worsens disputes.
Impact on post-assembly conversion
A customer who finishes assembly with the bot = product used, positive review, fewer "too complicated" returns. Successful guidance transforms unboxing friction into a brand moment.
How does it differ from support #353 and neighboring bots?
Seven assembly contents, seven distinct bot roles.
Ops assembly errors (#353)
Guide #353: policy ASM-LIABILITY, macros ASM-*, human escalations. #354 implements the step-by-step AI guidance layer with handoff to #353 in case of litigation.
Tutorials bot (#233)
Tutorials bot (#233): matching tut_id wide usage. #354 focuses on tut_assembly only, sequential flow with confirmation gates.
Installation support (#228)
Installation (#228): human workflow INS-*. #354 automates step guidance from the same indexed assets.
Onboarding bot (#179)
Onboarding (#179): proactive D+0 to D+30 global. #354 = active assembly session mid-assembly, not welcome series.
Manuals bot (#331)
Manuals bot (#331): open Q&A on long PDF. #354 enforces the order of steps, not random section search.
Video support
Support video: when to send clip. #354 embeds video per step in the ASM-BOT flow.
Promise #354
Intents assembly_*, flow ASM-BOT-GATE, caution safeguards, handoff liability, anti-skip, KPI assembly_bot.
Which assembly intents should the bot classify?
Map the assembly_bot intents before flows.
Ten assembly bot intents
assembly_start: customer starts assembly
assembly_step_next: next step confirmed
assembly_step_blocked: blocked at step N
assembly_step_repeat: repeat current step
assembly_tools: which tool for step N
assembly_orientation: which part, A or B
assembly_safety: safety instruction for step
assembly_broken_part: broken part reported
assembly_complete: assembly finished, final check
assembly_abort: customer gives up, handoff
Mandatory session fields
sku, tut_assembly_id, current_step, total_steps, step_confirmed[], tools_required[], safety_ack (bool), photos_pending, liability_flag. See taxonomy (#135).
Mining 90-day tickets
Export "assembly", "step", "assemble", "screw", "blocked". Cross-reference assembly_stuck (#353) tags. Prioritize flows for the top 20 kit SKUs.
Distinction: assembly_start vs tut_setup
tut_setup (#233) includes pairing, first use. assembly_* (#354) = physical kit assembly only. Classify separately for KPIs.
Verbatims to train
"How do I assemble step 3?", "Which screw for the crossmember?", "I don't understand the diagram", "Can I skip this step?", "I broke it while assembling". Tag in the assembly intent classifier.
Relevant Verticals
Flat-pack furniture, bicycles, outdoor, pergola, fitness, toys, smart home sensors. Assembly bot flow is high priority if assembly_steps_count > 5 and TPU assembly is high.
How to build the ASM-BOT-GATE flow step-by-step?
The ASM-BOT-GATE flow enforces one step at a time with confirmation before progression.
Six sequential gates
SKU Gate: order line → correct tut_assembly_id
Inventory Gate: package parts verified at step 0
Step Gate: current_step N, content indexed
Caution Gate: safety_ack if critical step
Confirmation Gate: client confirms step OK
Exit Gate: step_next, repeat, handoff, complete
Step 0 package inventory
Before step 1: bot lists hardware bag parts, client confirms everything is present. Missing → handoff assembly_missing_part (#353 branch B), do not proceed with assembly.
Step N loop
Display step N instruction (text + GIF/video) → list tools → safety if applicable → "Confirm when step completed" → step_confirmed push → current_step N+1.
Anti-skip rule
Client asks to "skip to step 8": refusal unless step_confirmed[1..7] are true. Explain risk of instability or breakage. Exception: expert handoff if partial assembly is documented with photos.
Max turns per step
3 clarifications max on the same step → offer step video or expert handoff. No LLM infinite loop.
Which data sources does the assembly bot read?
The assembly bot only guides from verifiable sources.
Five indexed sources
tut_assembly_id: JSON steps array library #232
Metafield product.assembly_steps_count: total steps
Metafield product.assembly_tools: global required tools
Metafield product.assembly_revision: v1, v2 SKU
Chunks MANUAL-RAG: manual assembly sections #331
Heeya estimates 55 to 72% tier-1 deflection when RAG is chunked and routed by SKU (Heeya, RAG support 2026).
JSON step structure
Each step: step_number, title, instruction, tools[], safety_warning (bool), video_url, confirmation_prompt, critical (bool). Bot cites step_number, never invents instruction.
Sync INS-ASSEMBLY #228
Same assets as INS-ASSEMBLY human macros. One source, bot + agents. See Shopify bot data.
SKU revision
Order variant → assembly_revision metafield → correct tut_assembly_id. Mismatch → handoff, no guidance on wrong revision.
Multimedia per step
Each step JSON includes optional video_url or gif_url. Bot prioritizes visual elements for orientation and tightening steps. Align support video matrix.
Estimated assembly duration
Metafield product.assembly_duration_minutes displayed step 0. Calibrates customer expectations and reduces mid-flow abandonment.
What safety precautions should be imposed on the assembly bot?
The ASM-BOT-GUARD safeguards protect the customer and brand during active guidance.
Five strict rules
Grounded steps: bot only cites steps indexed under tut_assembly_id
No force: never say "tighten hard" without manual citation
ASM-PAUSE: broken part → stop flow, handoff #353
Safety ack: critical step requires safety confirmation
Anti-skip: no step N+2 without confirmation of N
Assembly bot system prompt
"You guide one step at a time. You never skip a step. You ask for confirmation before progressing. If a part is broken or there is a dispute, you handoff without promising a spare." See anti-hallucination.
Critical steps
Mark critical=true: structural tightening, irreversible orientation, electrical connections. Bot displays safety_warning verbatim + requests photo before proceeding.
Handoff liability
Intents assembly_broken_part, assembly_abort, assembly_liability_dispute: payload sku, current_step, photos[], session log → agent #353. Bot does not decide on free spares.
Regression tests
25 scenarios: full start, blocked step 4, skip request, broken part, wrong revision, missing hardware step 0, complete check. Target assembly_hallucination is 0%.
Which bot flows for pre-assembly and mid-assembly?
The assembly bot operates on multiple touchpoints.
D+0 packaging QR Flow
QR Scan → pre-filled chat sku + assembly_start → step 0 inventory → step 1. Proactive post-delivery if onboarding_active tag (#179).
Reactive mid-assembly Flow
Customer "blocked at step 6" → assembly_step_blocked → bot resumes current_step 6 if session exists → otherwise asks for sku + step number + photo of the area.
Part orientation Flow
assembly_orientation: bot displays step diagram + "part A logo side facing up" → customer confirmation before step_next.
Complete check Flow
assembly_complete: final checklist (screws tightening, stability, remaining parts) → link to troubleshooting (#229) if post-assembly anomaly occurs.
Example bot reply step 4
« Step 4/12: Attach crossbar B to the left upright. Use the provided 4 mm Allen key. Do not tighten fully, allow some play for step 5. [15 s GIF]. Confirm when done to proceed to step 5. »
Assembly handoff payload
sku, tut_assembly_id, current_step, step_confirmed[], session_duration, photos[], liability_flag, customer verbatim.
Session resumption Flow
Customer closes chat and returns: bot reads current_step + step_confirmed from session store. Resumes with "Let's pick up at step [N]" without asking for SKU again if the order is identified.
Assembly bot tone
Encouraging, never pushy. "Take your time" on critical steps. No technical jargon that is not present in the step JSON.
How do I connect the Assembly Bot to Shopify?
The Shopify bot assembly setup aligns tut_assembly, metafields, and post-delivery triggers.
Technical Checklist
Create tut_assembly_id for top 30 kit SKUs (#232)
Populate assembly_steps_count, assembly_revision metafields
Index RAG JSON step chunks by step_number
Configure assembly_* intents and ASM-BOT-GATE
Draft ASM-BOT-GUARD prompt section 6
Connect fulfillment delivered trigger + packaging QR code
Test 25 regression scenarios
Weekly assembly_bot_resolution dashboard
Product Guide Page Widget
Page /pages/montage-[sku]: bot widget takes over order if logged in. Same tut_assembly_id as package QR code.
Shopify Flow Trigger
fulfillment delivered + product tag requires_assembly → tag assembly_bot_eligible + proactive chat message on Day 0.
Progressive Rollout
Phase 1: top 5 SKUs, assembly_start up to step 5. Phase 2: complete flow + handoff #353. Phase 3: proactive Day 0 for all kit SKUs.
Which KPIs should be measured on the bot assembly?
Without assembly_bot KPIs, it is impossible to prove AI guidance ROI.
Seven key metrics
assembly_bot_resolution: assembly completed without agent / bot sessions
assembly_step_completion_rate: confirmed steps / proposed steps
assembly_bot_handoff_rate: handoff #353 / assembly sessions
assembly_blocked_step_top: most blocking step N
assembly_session_duration: median duration of full flow
NTF_return_post_bot: NTF returns after bot guidance
CSAT intent assembly_bot: post-session satisfaction
DTC Benchmark
Target assembly_bot_resolution > 65%, assembly_step_completion_rate > 85%, assembly_bot_handoff_rate < 20%, CSAT assembly bot > 4.5/5.
Correlation #353
The bot must reduce assembly_stuck tickets without increasing assembly_dispute_rate. Stable handoff liability = guards OK.
Monthly review
Top blocked_step by SKU → redesign step tutorial or video clip. Fill content gaps #173.
Blocking step alert
Threshold > 15% blocked sessions on same step / SKU / month: audit step instruction JSON + priority clip production.
Segment by trigger
Packaging QR vs reactive chat vs day-0 proactive: distinct assembly_bot_resolution KPIs. Day-0 proactive often converts better if the tutorial is indexed before batch shipment.
What edge cases and escalations should be anticipated?
Five edge cases assembly bot require handoff.
Missing part step 0
Stop ASM-BOT flow. Handoff assembly_missing_part (#353). Do not continue incomplete assembly.
Broken part mid-flow
ASM-PAUSE activated. Bot collects ASM-PHOTO (#353) photos. Liability handoff, do not promise spare.
Customer forces skip
Bot refusal + explanation of risk. 3 skip requests → expert handoff or ASM-LIABILITY (#353).
SKU mismatch revision
tut_assembly_id missing for variant order → handoff, no generic LLM guidance.
Partial assembly resumed Day+7
Customer returns a week later: bot requests photo of current state + last step_confirmed → resume step N or restart if in doubt.
Refund request mid-flow
Empathy + offer 15 min expert. Handoff to supervisor if insistent. Bot does not promise return.
How does Qstomy guide product assembly?
Qstomy processes assembly_* intents from tut_assembly_id, order history, and ASM-BOT-GUARD guardrails.
Capabilities
assembly_step_guide: complete ASM-BOT-GATE flow
assembly_video_embed: step-by-step clip
assembly_safety_gate: critical step acknowledgment
Handoff #353: session payload + liability photos
Proactive Day+0: post-delivery SKU kit trigger
Encrypted DTC Scenario
Flat-pack furniture brand, 112 assembly tickets/month, PDFs on repeat.
After Qstomy assembly bot: 70% complete bot sessions, assembly_bot_resolution 73%, assembly_dispute_rate 5%, CSAT assembly bot 4.7/5.
Explore AI support, Shopify, request a demo.
What is the checklist for launching the bot assembly?
Checklist bot assembly (10 steps)
Audit assembly tickets 90 days (#353)
Create tut_assembly_id top 30 SKU kit (#232)
Document ASM-BOT-GATE flow section 4
Write ASM-BOT-GUARD prompt section 6
Configure assembly_* intents
Align ASM-LIABILITY handoff policy #353
Test 25 regression scenarios
Activate D+0 trigger post-delivery
Weekly assembly_bot_resolution dashboard
Monthly blocked_step review and tutorial overhaul
In brief
#354 = step-by-step assembly bot, not liability ops (#353)
ASM-BOT-GATE: one step, confirmation, anti-skip
10 assembly_* intents: start, next, blocked, broken
Caution: safety ack, ASM-PAUSE, handoff liability
assembly_bot_resolution KPI: target > 65%
FAQ
Difference with support #353?
#353 = liability policy and human macros. #354 = step-by-step AI guidance.
Difference with tutorial bot #233?
#233 = general use tut_id routing. #354 = assembly sequential flow only.
Can the bot skip steps?
No, except for expert handoff if partial assembly is documented.
Broken part during bot flow?
ASM-PAUSE, photos, handoff #353. The bot does not promise a spare.
Without indexed tut_assembly_id?
Agent handoff. No LLM guidance alone.
Going further
Test staging flow: assembly_start test SKU, verify that the bot refuses to skip a step without confirmation.
Share this guide #354 with product content: an assembly bot is only reliable if tut_assembly_id is structured step by step.
Enzo
July 22, 2026
