Chatbots
Why this matters for your business
Two human-bottleneck problems exist in every store:
- Customers want answers immediately. They're shopping at 2 AM, your support team isn't online, and the question ("does this come in size M?", "what's your return policy?", "where's my order?") is the difference between a sale and abandonment. By morning, the customer has bought from a competitor.
- The merchant wants answers immediately. "How do I set up cart recovery?" "What's the difference between campaigns and journeys?" "Why didn't this rule fire?" Hunting through docs interrupts the flow of work.
Chatbots solve both. The shopper chatbot answers customer questions on your storefront 24/7 — using your product catalog, order data, FAQ, and policies as the knowledge base. The admin chatbot answers your operational questions inside the platform — using the docs, your shop's data, and current state.
The unlock isn't replacing humans entirely; it's handling the 70-80% of questions that have well-known answers, so humans focus on the 20-30% that need judgment.
What this typically unlocks
| Outcome | Result |
|---|---|
| Storefront questions answered without human | 62-78% typical |
| Avg shopper-question response time | 8 seconds vs. hours |
| Reduction in support ticket volume | −45% |
| Cart abandonment from "I had a question" | −22% |
| Merchant time spent in docs | −60% |
What you actually get
Shopper chatbot
| Capability | Description |
|---|---|
| Storefront widget | Embeds on your site (one snippet) |
| Knowledge base | Auto-trained on product catalog + order data + your FAQ |
| Order tracking | Customer asks "where's my order?" → real-time tracking |
| Sizing / availability | "Do you have this in M?" → real-time inventory |
| Returns / policy | Trained on your policies |
| Cart assistance | "Can you help me check out?" → walks through |
| Human handoff | If confidence < 0.7 or customer asks → routes to support |
| Multi-language | Detects + responds in customer's language |
| WhatsApp + email handoff | Continue chat via WA/email if customer prefers |
Admin chatbot
| Capability | Description |
|---|---|
| In-app overlay | Bottom-right of every admin page |
| Docs trained | Knows every page of these docs |
| Shop-state aware | Knows your current setup, recent activity |
| Action handoff | "Show me my cart-recovery rule" → opens it |
| Walkthrough mode | "How do I set up X?" → guided step-by-step |
| Diagnostic Q | "Why didn't this fire?" → checks logs |
How the shopper chatbot works
Knowledge sources
The shopper chatbot pulls from:
- Product catalog: title, description, variants, inventory, price
- Order data: customer's own orders (after identification)
- Your uploaded FAQ: PDF / text uploads (return policy, shipping info, etc.)
- Recent campaigns: so it can answer "what's on sale?"
- Sizing charts (if uploaded)
- Reviews (high-confidence only)
It does not invent answers. If the question is outside knowledge, it says "I don't know that — let me get a human."
Handoff to human
Three triggers:
- Customer explicitly asks ("can I talk to a person?")
- AI confidence below threshold (default 0.7)
- Topic is in the always-escalate list (e.g. legal, refunds over $X)
Handoff includes the full conversation transcript, so the human picks up without making the customer repeat themselves.
Real merchant scenarios
Scenario A — Apparel brand cuts support volume 45%
Setup. $5M apparel brand, 3 part-time support agents. Inbound questions: ~180/day across email + WA + storefront contact form.
Pre-chatbot: 180/day handled by 3 agents = 60 each = 4 hours of work each = team always behind.
Post-shopper-chatbot (60 days):
| Metric | Before | After |
|---|---|---|
| Total inbound questions | 180/day | 220/day (more shopper engagement) |
| Bot-handled | 0 | 142/day (65%) |
| Human-handled | 180 | 78/day |
| Avg response time | 4 hours | bot: 8s, human: 30 min |
| CSAT | 4.3/5 | 4.6/5 |
The CSAT improvement is the surprise. Faster response (even from a bot) often beats waiting hours for a human.
Scenario B — Sizing helper for footwear brand
Setup. Footwear brand, returns 28% on sizing alone.
Chatbot training: Uploaded sizing charts for every product; trained on customer review patterns ("runs small", "fits true to size").
Customer flow:
- "I'm a women's 8 in Nike — what size in your brand?"
- Bot: "For [product name], reviews suggest sizing up half a size. Try women's 8.5. If you typically wear narrow, you might try our regular fit. Want me to add an 8 and 8.5 side-by-side to compare?"
60-day result:
- Sizing-related returns: 28% → 17%
- Pre-purchase sizing chat engagement: 38% of product-page visitors
- Return-cost savings: ~$24K
Scenario C — Admin chatbot reduces founder's docs time
Setup. Founder running solo, building first journey.
Pre-admin-chatbot: Searched docs for "how to add a condition to step 3 of a journey" — found a relevant page in ~3 minutes, scanned for the exact answer in another 5 min.
With admin chatbot:
- Asked: "How do I add a condition to step 3 of my welcome journey?"
- Response: "Go to Sales engine → Journeys → Welcome Series → click step 3 → 'Add condition' button. Want me to take you there?"
- Click "Yes" → opened the right page
Time: 8 seconds vs. 8 minutes.
Scenario D — Multi-language support without hiring
Setup. Brand selling in US + EU + Brazil. No multi-lingual support staff.
Chatbot setup: Auto-detects language from customer's browser + initial message. Responds in same language. Languages supported: English, Spanish, Portuguese, French, German, Italian, Dutch.
Result: Brazilian customers (Portuguese) saw 4× the engagement on storefront chat vs. when chat was English-only. Avg order from Brazil up 22% in 60 days.
Best practices
✅ Train on your actual support backlog. The 100 most common questions become 80% of your bot value.
✅ Set escalation thresholds carefully. Too high = customers stuck with bot; too low = humans drowning.
✅ Update knowledge base monthly. Product catalog changes, new FAQ entries, policy updates.
✅ Use admin chatbot as docs replacement. It's faster than search.
❌ Don't make the bot pretend to be human. Honesty about "this is an AI" builds trust; pretending erodes it.
❌ Don't disable handoff. Customers must be able to reach a human; bots-only is a complaint generator.
❌ Don't use the bot for high-stakes decisions (refunds over policy, cancellations, complaints). Always escalate.
Plan tiers
| Capability | Free | Starter | Pro | Agency | Enterprise |
|---|---|---|---|---|---|
| Shopper chatbot | — | — | ✓ | ✓ | ✓ |
| Admin chatbot | — | ✓ | ✓ | ✓ | ✓ |
| Multi-language | — | — | ✓ | ✓ | ✓ |
| Custom knowledge upload | — | — | ✓ | ✓ | ✓ |
| WhatsApp / email handoff | — | — | ✓ | ✓ | ✓ |
| Conversation analytics | — | — | ✓ | ✓ | ✓ |
| Multi-brand chatbot config | — | — | — | ✓ | ✓ |
| Custom AI provider | — | — | — | — | ✓ |
See also
- Communications overview
- WhatsApp engine — P3 AI Concierge similar concept, on WA
- Storefront widgets — chatbot embeds via widget framework
- Customer 360 — chat history attaches here