Skip to main content
RapidDev - Software Development Agency
AI ImplementationsE-commerce & Retail20 min read

White-Label AI Chatbot for E-commerce Support for DTC & 3PL Agencies

Three paths: buy Stammer.ai Full White-Label at $197/mo and add a Shopify connector, hire RapidDev to build a custom chatbot for $13K–$25K, or scaffold it with Lovable in a weekend for ~$50. Research recommends buy-saas — Stammer at $197/mo resells at $349–$499/mo per merchant with 43%+ margin and zero engineering risk, making it the fastest path to profitability for agencies under 50 clients.

4.9Clutch rating
600+Happy partners
17+Countries served
190+Team members

Decision matrix

Should you buy, hire, or build it yourself?

Three paths to launch a AI Chatbot for E-commerce Support, side-by-side. Pick the one that matches your budget, timeline, and how much control you actually need.

Recommended

Buy Stammer.ai Full White-Label

Buy SaaS
Time to launch
1–2 weeks (setup + connector)
Upfront cost
$0 build
Monthly cost
$197/mo Stammer + Shopify connector dev ($0 or 1-time $2K)
Ownership
Stammer owns the AI layer; you own the client relationship
Customization
Custom domain, branded chat widget, client dashboard — full visual rebrand

Best for

Agencies with 5–50 merchant clients who want to launch a managed support-chatbot service in two weeks without hiring a developer

Risks

  • Stammer runs on OpenAI under the hood — you cannot swap to Claude or Gemini if a client requires a specific model
  • Deep Shopify features (refund API automation, Shop Pay order modification) are not available in off-the-shelf Stammer; they require a custom connector add-on
  • At 100+ merchants the $197/mo flat fee is cheap, but Stammer's roadmap controls your feature velocity — you cannot accelerate it
  • If Stammer raises prices or changes terms, migrating 50+ client installations is a 2–3 week operational project

Hire RapidDev

Hire agency
Time to launch
5–8 weeks
Upfront cost
$13,000–$25,000
Monthly cost
$200–$500 infra
Ownership
You own the code
Customization
Unlimited — your roadmap

Best for

Agencies at 50+ merchants or those needing deep Shopify refund automation and custom escalation routing that Stammer cannot provide

Risks

  • Build cost requires 5–8 months of $349/mo per-merchant revenue to break even at 10 clients
  • Shopify Admin API rate limits (40 req/min on standard plans) require careful queueing logic in the build
  • Maintaining vector store freshness when help docs update is an operational overhead that needs a sync pipeline
  • PCI-DSS scope risk if the build ever touches payment-related conversation data — requires explicit design guardrails

Build with Lovable

Build yourself
Time to launch
1 weekend
Upfront cost
$25 Lovable Pro + ~$25 API credits
Monthly cost
$50–$150 + API
Ownership
You own the code
Customization
Limited by your Lovable prompt skill

Best for

Individual Shopify consultants wanting a proof-of-concept for a single client before committing to Stammer or a full build

Risks

  • Lovable cannot scaffold Shopify Admin API OAuth securely without explicit edge function prompts — raw API keys in frontend code is a critical security failure
  • The pgvector help-doc embedding requires a manual Supabase SQL migration step that Lovable will not execute automatically
  • Multi-tenant widget embedding (one chat widget per merchant domain) requires CORS configuration Lovable will not wire up correctly by default
  • A Lovable MVP is not production-ready for handling real customer refund requests without a full security and load review

What a AI Chatbot for E-commerce Support actually does

Handles post-purchase customer support — order status, returns, refunds, and product troubleshooting — via a branded AI chat interface embedded on any Shopify or WooCommerce storefront.

The post-purchase support chatbot works through a RAG pipeline: a tenant's help docs, return policies, and FAQ content are ingested and embedded into a vector store. When a customer opens the chat widget and asks 'where is my order?', the bot queries the Shopify Admin API for real-time order status, enriches the response with shipping carrier data, and surfaces the answer without human intervention. Escalation to Zendesk or Gorgias fires when the bot detects frustration signals or a return/refund request requiring policy exceptions. The key cost advantage is $0.002 per 3-turn support conversation on Claude Haiku 4.5 versus $0.85+ per conversation on Zendesk AI Agents.

This category is unusual in the ecommerce cluster: real white-label SaaS options actually exist. Stammer.ai's $197/mo Full White-Label plan lets agencies deploy under their own brand with custom domains, branded chat widgets, and client-level dashboards — a genuine rebrand, not just a logo swap. For agencies under 50 merchants, the buy-path has a structural cost advantage: Stammer's flat $197/mo eliminates per-message overage risk that makes build economics variable. The inflection point where a custom build wins is approximately 80–100 merchants, where the SaaS fee plus Shopify connector development time starts to exceed the amortized build cost.

AI capabilities involved

Order-status and WISMO resolution via live Shopify API lookup

Claude Haiku 4.5GPT-5.4 nanoGemini 3.1 Flash-Lite

Return and refund initiation with policy-aware guardrails

Claude Haiku 4.5Claude Sonnet 4.6GPT-5.4 mini

RAG-based help-doc retrieval for product troubleshooting

Claude Haiku 4.5GPT-5.4 miniMistral Small 3.2

Multi-language support across 100+ languages

Gemini 3 FlashClaude Haiku 4.5GPT-5.4 mini

Who uses this

  • DTC Shopify agencies serving 10–50 merchant clients who want to add AI support automation as a managed service line
  • 3PL operators offering support-as-a-service as a value-add to fulfillment clients
  • Shopify Plus partners looking to bundle an AI chatbot tier into their agency retainer
  • DTC brands with 5,000+ monthly orders who have outgrown Gorgias automation templates and need deeper Shopify API integration

SaaS alternatives on the market

Real products you can sign up for today — with current 2026 pricing, honest pros and cons.

Stammer.ai

DTC agencies serving 5–50 Shopify merchants who want to launch a managed AI support service in two weeks without writing a line of code

7-day trial

$197/mo (Full White-Label)

Custom quote

Pros

  • +Genuine full rebrand: custom domain, branded widget, white-labeled client dashboard — no Stammer branding visible
  • +Flat $197/mo with no per-message overage — economics stay predictable even at high conversation volume
  • +Clients can train their own bots from help docs and product catalogs without agency involvement
  • +Fastest path to a rebrandable ecommerce chatbot: one Shopify connector and you're billing clients in week two

Cons

  • OpenAI under the hood — no option to route to Claude or Gemini if a client requires a specific model or data residency
  • Shopify Admin API integration for live order status and refund automation is not native — requires a custom connector
  • Conversation history analytics are limited compared to Gorgias or Zendesk; no native CSAT scoring
  • At $197/mo flat, margins compress if you're serving high-volume merchants running 10K+ conversations/mo
Shopify order-status integration is not native — expect 1–2 days of connector dev (or $2K one-time if outsourced) before clients get real WISMO automation.

ChatLab

Agencies whose clients have strong model-preference opinions (Claude vs. GPT) and are willing to pay a premium for multi-LLM routing under one dashboard

14-day trial

$360/mo (White-Label)

Custom quote

Pros

  • +Multi-LLM routing: GPT-5.4, Gemini 3 Flash, and Claude Haiku 4.5 under one interface — clients can choose their preferred model
  • +Full visual rebrand including domain, colors, fonts, and chat bubble
  • +Native analytics dashboard with conversation-volume and deflection-rate metrics
  • +API access on all white-label tiers for custom workflow integrations

Cons

  • At $360/mo (vs. Stammer's $197/mo), the economics require reselling at $499+/mo to maintain 28%+ margin
  • No native Shopify Admin API integration for order-status lookups — same connector requirement as Stammer
  • Less documentation and fewer G2 reviews than Stammer — harder to evaluate reliability before signing
  • White-label tier pricing may change — verify current floor before committing to client contracts
At $360/mo base, you need at least 2 merchant clients at $349/mo just to break even — the margin math is tighter than Stammer from day one.

Gorgias AI Agent

Shopify merchants buying support tooling directly — not the right fit for agencies who need to resell under their own brand

7-day trial

$360/mo Pro (with AI add-on)

Custom quote

Pros

  • +Deepest native Shopify integration in the category: refund processing, order cancellation, address edits via Gorgias rules — no connector build required
  • +Trusted by 15,000+ ecommerce brands — strong social proof for client adoption
  • +Automation library covers 80% of standard WISMO and return flows out of the box
  • +Multi-channel: chat, email, SMS, Instagram, Facebook — one platform for all touchpoints

Cons

  • No white-label tier at any price — your client sees Gorgias branding on every touchpoint
  • Gorgias positions itself as a direct merchant tool, not an agency-resold product
  • Price is per-merchant, not agency-flat — at $360/mo per merchant, reseller economics require significant markup
  • AI features (Automate add-on) are metered per resolution — unpredictable cost at scale
Gorgias has no white-label tier at any price point. If your agency pitch includes 'our AI support platform,' Gorgias will always undermine that with its own brand on the widget.

The AI stack

The production pipeline needs four layers: a chat LLM for conversation handling, an embedding model for help-doc RAG, a Shopify API connector for live order data, and an escalation router. The critical cost decision is routing routine WISMO queries through Claude Haiku 4.5 ($0.002/conversation) and reserving Claude Sonnet 4.6 only for policy-exception disputes.

01

Conversation LLM (primary chat layer)

Handles the full customer support conversation — understanding intent, querying context, generating responses

Claude Haiku 4.5

$1/$5 per M tokens

All standard WISMO, FAQ, and return-status queries where volume is high and individual conversation complexity is low

+ Lowest cost for high-volume support conversations; excellent instruction-following for policy-constrained responses 200K context cap — adequate for single conversations but limits long multi-session context

Claude Sonnet 4.6

$3/$15 per M tokens

Escalated disputes, refund policy exceptions, and complex multi-item order troubleshooting

+ Better nuanced reasoning for policy-exception cases ('my order arrived damaged but the 30-day window passed') 3× the cost of Haiku — only justified for escalated conversations where reasoning depth matters

GPT-5.4 nano

$0.20/$1.25 per M tokens

Merchants processing 50,000+ support conversations/mo who need maximum cost efficiency on routine queries

+ Cheapest available option for highest-volume merchants; adequate for pure WISMO with templated responses Weaker instruction-following than Haiku — policy guardrails require more prompt engineering to enforce

Our pick: Use Claude Haiku 4.5 as the default for all tenants. Add a routing layer that escalates to Sonnet 4.6 when the conversation contains refund/dispute keywords. Offer GPT-5.4 nano as a cost-optimized tier for highest-volume merchants at a lower price point.

02

Help-doc embedding (RAG layer)

Converts merchant help docs, return policies, and product FAQs into searchable vector embeddings for grounding bot responses

text-embedding-3-small

$0.02/M tokens

All standard help-doc embedding across all merchant tiers; the cost difference from large model is negligible at this scale

+ Extremely cheap for the ingestion volume typical of an ecommerce help center (100–500 documents per merchant) Slightly lower semantic accuracy than large model — mitigated by using cosine similarity thresholds with a fallback

Our pick: text-embedding-3-small on pgvector in Supabase for all tenants. Re-embed documents on a nightly sync schedule when the merchant updates their help center. Cosine similarity threshold of 0.75 before including a context chunk.

Reference architecture

The pipeline is a real-time chat flow: customer message → intent classification → Shopify API lookup (if order-related) → help-doc RAG retrieval → LLM response generation → escalation check. The hardest engineering challenge is Shopify Admin API rate limiting (40 req/min on standard plans) combined with the need for sub-2-second response times — requiring a Redis queue and rate-limit retry logic in the backend.

01

Customer opens chat widget on merchant storefront

Next.js chat widget (embedded via script tag) + Supabase Realtime

The widget loads from a CDN with tenant-specific configuration (colors, greeting, logo). A Supabase Realtime channel opens for the session. Tenant and store domain are sent in the session init — no cookie or localStorage access.

02

Customer message is classified by intent

Supabase Edge Function → Claude Haiku 4.5

A lightweight intent classifier (Haiku 4.5 with a short system prompt + few-shot examples) routes the message to one of three flows: order-lookup (WISMO), help-doc (FAQ/troubleshooting), or escalation (refund/dispute). Classification cost is ~$0.0001 per message.

03

Shopify Admin API lookup (order-related intents only)

Supabase Edge Function → Shopify Admin GraphQL API

Order number or email is extracted from the conversation context. The edge function queries Shopify Admin GraphQL for order status, fulfillment, carrier, and tracking URL. Response is cached in Supabase for 5 minutes to avoid rate limit hits.

04

Help-doc RAG retrieval (FAQ/troubleshooting intents)

Supabase pgvector → cosine similarity search

The customer message is embedded with text-embedding-3-small. A pgvector similarity search returns the top-3 help-doc chunks above 0.75 cosine threshold for the tenant. These chunks are included in the Claude Haiku 4.5 context window.

05

LLM generates response with policy guardrails

Claude Haiku 4.5 (or Sonnet 4.6 for escalated flows)

The system prompt includes the merchant's return policy, refund limits, and escalation triggers. Claude generates a response grounded in the order data or help-doc context. A JSON output mode enforces structured responses with a 'should_escalate' flag.

06

Escalation routing on trigger conditions

Edge Function → Zendesk/Gorgias REST API

If 'should_escalate' is true, the full conversation history is posted to the merchant's Gorgias or Zendesk queue via API. The customer sees 'A support agent will follow up within 2 hours' — the bot does not attempt to resolve the escalated issue.

Estimated cost per request

~$0.002 per 3-turn WISMO conversation (Claude Haiku 4.5); ~$0.015 per escalated refund dispute (Sonnet 4.6, ~3K in + 500 out)

Cost calculator

Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.

Calculator models a 20-merchant white-label deployment with each merchant handling 500 support conversations per month. Adjust merchant count and conversation volume to see real infra cost versus Stammer SaaS fee.

20 merchants
1200
500 conversations
5010,000
15 %
550

Estimated monthly cost

$63.51

$762 per year

Supabase Pro (DB + Auth + Realtime + pgvector)$25.00
Vercel Pro (edge functions + CDN widget delivery)$20.00
Redis (Upstash Shopify rate-limit queue)$10.00
Claude Haiku 4.5 (standard conversations)$1.00
Claude Sonnet 4.6 (escalated conversations)$7.50
text-embedding-3-small (help-doc RAG per conversation)$0.01
Fixed: $55.00/moVariable: $8.51/mo

Calculator notes

  • Sonnet cost applies to ~15% of conversations (escalation rate) — adjust the escalation_rate_pct slider to match your merchant mix
  • Help-doc embedding ingestion cost is a one-time $0.50–$2 per merchant (500–5,000 docs at $0.02/M tokens) — not included in per-conversation cost
  • Shopify Admin API calls are free but rate-limited; Redis queue cost is $10/mo flat for Upstash on most agency scales
  • At the default 20 merchants × 500 conversations, total AI cost is ~$25/mo — compare against Stammer's $197/mo flat fee
  • The custom build breaks even against Stammer at approximately 85 merchants paying $349/mo, assuming a $20,000 build cost amortized over 24 months

Build it yourself with vibe-coding tools

If you're validating before buying Stammer, a Lovable weekend build gives you a working chatbot with Shopify order lookup and help-doc RAG — enough to demo to one client and confirm they'd pay $349/mo for it.

Time to MVP

12–16 hours (1 weekend)

Total cost to MVP

$25 Lovable Pro + ~$25 API credits (Anthropic + Shopify dev store)

You'll need

Supabase project with pgvector extension enabled and a help_docs table (id, tenant_id, content, embedding vector(1536))Anthropic API key for Claude Haiku 4.5Shopify Partner account with a development store and Admin API key (read_orders scope)text-embedding-3-small OpenAI API key for help-doc ingestionLovable Pro subscription ($25/mo)

Starter prompt

Lovable Prompt

Build a white-label AI ecommerce support chatbot with a multi-tenant architecture using Next.js and Supabase. Tenant model: each merchant has a name, Shopify store URL, Shopify Admin API key (stored encrypted in Supabase vault), and a collection of help docs. All tenant data is isolated via Supabase RLS. Core features: 1. Chat widget: an embeddable chat bubble script tag that merchants install on their Shopify store. The widget loads from /widget/{tenant_slug}.js and initializes a full-screen chat panel. No Lovable branding anywhere. 2. Message handling Edge Function: on each customer message, call Claude Haiku 4.5 with a system prompt that includes the merchant's return policy (stored as tenant config). Before calling the LLM, run a pgvector similarity search on the tenant's help_docs table using text-embedding-3-small embedding of the customer message (cosine threshold 0.75). If order-number is detected in the message, call the Shopify Admin GraphQL API to fetch order status and include it in the LLM context. 3. Escalation: if the Claude response contains 'ESCALATE' (enforce via structured JSON output with a should_escalate boolean), show the customer a 'Connecting you to a support agent' message and create a Zendesk ticket via Zendesk REST API with the full conversation history. 4. Agency dashboard: a Next.js admin panel at /admin where agency staff can: add new merchant tenants, upload help docs (PDF/TXT → chunk → embed → store in pgvector), view conversation logs per tenant, and see a simple deflection-rate metric (conversations resolved / total conversations). 5. Auth: Supabase email/password auth for agency dashboard. Widget sessions are anonymous (no customer login required). Env vars: NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_ANON_KEY, SUPABASE_SERVICE_ROLE_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY (for embeddings), ZENDESK_API_KEY, ZENDESK_SUBDOMAIN.

Paste this into Lovable

Follow-up prompts (run in order)

  1. 1

    Add Shopify webhook processing: create a Supabase Edge Function that receives Shopify orders/fulfilled webhooks and updates the order-cache table in real time. This eliminates the Admin API call on every WISMO message for merchants with high order volumes.

  2. 2

    Add multi-language routing: detect customer message language using a 1-token Claude Haiku classification call. If non-English, set the LLM system prompt to respond in the detected language. Add a language-distribution chart to the agency dashboard metrics page.

  3. 3

    Add help-doc sync: a Crawl Help Center button on the merchant settings page takes a Shopify help center URL and uses Firecrawl to scrape all articles, chunk them at 500 tokens, embed with text-embedding-3-small, and upsert into pgvector. Run nightly re-sync via a Supabase cron function.

  4. 4

    Add conversation analytics page: a per-tenant dashboard showing conversation volume by day, intent distribution (WISMO vs. FAQ vs. escalation), average resolution time, and CSAT score (collected via a 1-question post-chat survey: 'Did we solve your issue? Yes/No').

Expected output

A working chat widget that handles WISMO queries via live Shopify order lookup and FAQ questions via help-doc RAG, with agency-level multi-tenant management — runnable for ~$55/mo infrastructure.

Known gotchas

  • !Shopify Admin API OAuth for production stores requires Shopify Partner app approval — your dev store key works for testing but merchants cannot install an unapproved app; plan for a 2–4 week Shopify app review if you go custom instead of Stammer
  • !pgvector cosine similarity must use a threshold; without one, the bot will always include irrelevant chunks and hallucinate policy details that don't exist in the merchant's help center
  • !Lovable will scaffold the chat widget as a React component, not an embeddable script tag — you need a separate build step (Vite library mode) to turn it into an installable widget JS file
  • !Claude Haiku 4.5 will sometimes ignore the 'respond only in X language' instruction for short messages — add a language-enforcement check at the edge function level
  • !Never store Shopify Admin API keys in Supabase as plaintext — use Supabase Vault (encrypted secrets) or your custom encryption before storing any merchant credentials
  • !Supabase Realtime has a 200 concurrent channel limit on the free tier — upgrade to Pro before your first production merchant or chat sessions will silently fail at peak traffic

Compliance & risk reality check

An ecommerce support chatbot processes customer PII (name, email, order history), payment-adjacent data (order totals, shipping addresses), and sometimes refund-triggering conversations — making PCI scope, GDPR consent, and data retention the three non-negotiable compliance requirements.

Critical

PCI-DSS scope avoidance

Any chatbot that discusses payment amounts, card details, or order-total refunds is adjacent to PCI scope. A single instance of the bot echoing a partial card number from order data brings the entire system into PCI audit scope. The Shopify Admin API does not return card data, but conversation logs may contain user-typed card information.

Mitigation: Add a real-time PII scrubber (regex on card-number patterns before any LLM call and before conversation storage). Never store raw Shopify Admin API responses — extract only order_id, status, tracking_url, and carrier. Run conversation logs through a post-processing scrubber before writing to Supabase.

Important

GDPR/CCPA conversation logging and retention

Conversation logs containing customer names, email addresses, and order details are personal data under GDPR Article 4 and CCPA Section 1798.140. Retaining these logs indefinitely without a disclosed retention policy exposes merchants to GDPR Article 5 (storage limitation) violations.

Mitigation: Set a 90-day default retention period on conversation logs with a configurable per-tenant setting. Surface a data-deletion endpoint per customer (GDPR right to erasure). Display the retention policy in the chat widget's privacy disclosure. Provide DPA templates to merchants for their EU end-customers.

Good to know

FTC pricing-honesty rules on chat-initiated discounts

If the chatbot autonomously offers discount codes as retention tools (e.g., 'Here's 15% off to keep your order'), the discount terms must be clearly disclosed and non-deceptive per FTC Section 5. AI-generated discount offers that misstate terms or expiry create merchant liability.

Mitigation: Only allow discount-code offers that are pre-configured by the merchant in the tenant settings panel — never let the LLM invent discount amounts. Always include the discount code terms (minimum purchase, expiry date) in the offer message, pulled from tenant config, not generated by the model.

Build vs buy: the real math

5–8 weeks

Custom build time

$13,000–$25,000

One-time investment

4–7 months (vs. custom build cost); 0 months (vs. Stammer at $197/mo)

Breakeven vs buying

The buy-path has no break-even period — Stammer at $197/mo costs less than two months of one developer's time. A custom build at $13K–$25K breaks even against Stammer at the 67–127 month mark (5–10 years) if Stammer stays at $197/mo. The build becomes rational only at 80+ merchants where Stammer's flat fee is negligible as a percentage of revenue AND the agency needs Shopify Admin API features (refund automation, order modification) that Stammer cannot provide. For agencies under 50 merchants, buy Stammer, add a Shopify connector wrapper, and invest the saved $20K in client acquisition.

Skip the DIY — RapidDev builds the production version

A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship.

1

Discovery call (free)

30 min

We map your exact AI Chatbot for E-commerce Support use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.

2

AI-accelerated build

5–8 weeks

Our engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.

3

Launch + handoff

1 week

We deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.

What you get

Full source code (GitHub repo)
Deployed on your infrastructure
Audited prompts & model configs
Cost monitoring + budget alerts
3 months of bug-fix support
Direct Slack channel with engineers

Timeline

5–8 weeks

Investment

$13,000–$25,000

vs SaaS

ROI in 4–7 months (vs. custom build cost); 0 months (vs. Stammer at $197/mo)

Get your free estimate

30-min call. Fixed-price quote within 48 hours. No commitment.

Frequently asked questions

How much does it cost to build a white-label AI E-commerce Support Chatbot?

If you buy Stammer.ai's Full White-Label plan, upfront cost is $0 and ongoing is $197/mo — plus 1–2 days of Shopify connector setup (or a one-time $2K fee to a developer). A custom build with RapidDev costs $13,000–$25,000 with a 5–8 week timeline. For most agencies under 50 merchants, the buy-path is the correct choice — save the build budget for client acquisition.

How long does it take to ship this?

The Stammer.ai buy-path launches in 1–2 weeks including Shopify connector setup and client onboarding. A custom build takes 5–8 weeks. The Lovable weekend build gives you a proof-of-concept in 12–16 hours — useful for client demos before committing to Stammer or a full build.

Can RapidDev build this for my company?

Yes. RapidDev has shipped 600+ applications and 200+ AI implementations in production. We scope the Shopify Admin API integration, multi-tenant architecture, pgvector RAG pipeline, and escalation routing — and deliver in 5–8 weeks. Book a free 30-minute consultation at rapidevelopers.com to get a fixed-price quote.

Can the chatbot actually process refunds, or does it just provide information?

It depends on how you build the Shopify Admin API integration. Stammer.ai out-of-the-box provides information only — no refund API calls. A custom build can trigger Shopify refund-creation API calls programmatically, but this requires careful policy guardrails (max refund amount, time window, reason validation) to prevent abuse. Most agencies start with information-only and add refund automation as a premium tier after validating the information-flow economics.

What happens to conversations when a customer interacts with the bot — is that data private?

Conversation logs contain customer PII (name, email, order details) and must be handled under GDPR/CCPA. Set a 90-day retention period, scrub card-number patterns before storage, and provide a customer data-deletion endpoint. In the Stammer buy-path, review Stammer's data processing agreement to understand how they store and process conversation data before signing client contracts.

How does the chatbot know what's in my merchant's help center?

Help-center articles are chunked at 500 tokens, embedded with text-embedding-3-small, and stored in a pgvector index per tenant. When a customer asks a question, the customer message is embedded and a cosine similarity search returns the 3 most relevant help-doc chunks. These are included in the Claude Haiku 4.5 context window as grounding — the bot answers based only on these chunks, not from its training data. This means you must re-embed whenever the merchant updates their help center.

RapidDev

Want the production version?

  • Delivered in 5–8 weeks
  • You own 100% of the code
  • AI cost monitoring built in
Get a free estimate

30-min call. No commitment.

Matt Graham

Written by

Matt Graham · CEO & Founder, RapidDev

1,000+ client projects delivered. Columbia University & Harvard Business School alumnus, U.S. Navy veteran. About the author →

Want this built for you?

We ship production apps at a fixed price — $13K–$25K, 6–10 weeks, source code yours. You've seen what it takes; we do it every week.

Get a fixed-price quote

We put the rapid in RapidDev

Need a dedicated strategic tech and growth partner? Discover what RapidDev can do for your business! Book a call with our team to schedule a free, no-obligation consultation. We'll discuss your project and provide a custom quote at no cost.