Built with Chatbase
AI Sales & Support Agent for a Premium Skincare Brand
Khadijah (Dija) 💥
Building a 24/7 Support & Sales Agent for DTC Skincare
About this project: This is a technical showcase designed to demonstrate what a premium Chatbase setup looks like. While "Glowell" is a conceptual brand, the exact workflows, safety rules, and integrations detailed below are how I solve real operational bottlenecks for my e-commerce clients.
The Business Challenge
Direct-to-consumer skincare brands face a unique support problem. Because skincare is so personal, customers constantly need advice before they buy.
When a brand grows rapidly, this creates massive bottlenecks:
Drowning in Routine Questions: Support teams spend hours answering the exact same questions about order tracking and product ingredients.
Lost Sales: If a customer asks a question at 8 PM, they won't get an answer until the next morning. In e-commerce, that usually means an abandoned cart.
The Danger of Bad Advice: You can't just plug a basic AI into a skincare website. If the AI gives bad medical advice or recommends the wrong product, it's a massive liability.
To solve this, I built a system that acts less like a "chatbot" and more like a highly trained digital employee. Here is how I built it:
Phase 1: Giving the AI a "Single Source of Truth"
An AI is only as smart as the information you give it. If you just copy and paste a whole website into a bot, it gets confused and starts making things up (hallucinating).
Instead, I built a highly organized knowledge library:
Clean Product Data: I formatted the brand's catalog so the AI could perfectly understand it. For every product, the AI knows the exact price, what it's used for, what ingredients it contains, and what other product pairs perfectly with it.
Clear Policies: I mapped out the most common customer questions (like "Where is my order?" or "How do I return this?") to single, clear, approved answers. This ensures the AI never guesses on company policy.
Structuring the catalog data so the AI never guesses on policy or price.
Phase 2: Setting Strict Rules and Boundaries
To make sure the agent sounded like a premium brand consultant and never gave dangerous advice, I set up strict, unbreakable rules.
The Persona (How it acts):
I instructed the agent to act as an elite Digital Esthetician. The primary rule: when recommending a product, it must explain why it works for the customer's specific skin type, and suggest one complementary product to help increase the average order value.
The Guardrails (What it CANNOT do):
Safety is critical. I implemented a strict liability firewall:
By locking these in as hard constraints, the AI is structurally blocked from breaking safety rules just to be "helpful."
Testing the medical liability firewall in real-time.
Phase 3: Building Step-by-Step Customer Journeys
This is what separates a basic toy from a business tool. I built "Procedures"—step-by-step workflows the AI must follow when a customer asks certain questions.
I built three core journeys:
Journey 1: "The Routine Builder" (Driving Sales)
When a customer asks for product recommendations, the AI doesn't just throw links at them. It follows a script:
Ask the customer about their skin type.
Recommend the best product and explain the ingredients.
Suggest a complementary product that pairs well with it.
Journey 2: "Order Support" (Saving Time)
When a customer asks where their package is:
Ask for their order number.
Silently connect to the brand's actual database to pull the live tracking info.
Present the exact delivery date to the customer in a friendly way.
Journey 3: "The Escalation" (The Safety Net)
If a customer reports an allergic reaction:
Immediately express empathy and tell them to stop using the product.
Ask exactly what happened and record their symptoms.
Automatically escalate the conversation to a human manager and promise an email within 30 minutes.
The AI acting as a consultant to naturally drive cross-sales.
Phase 4: Connecting to Real Business Data
To prove the agent can actually do things instead of just chatting, I gave it custom tools.
Live Database Connection: I built a custom connection that allows the AI to securely look up real order data. This means the AI can tell a customer exactly where their package is, just like a human agent checking Shopify.
Smart Escalations: The AI is locked out of using the "escalate to human" tool unless it is following a specific, approved customer journey. This ensures your human support team doesn't get flooded with unnecessary alerts.
The AI pulling live order tracking data automatically.
Phase 5: Seamless Brand Integration
A premium DTC brand cannot have a generic, out-of-the-box chatbot sitting on its website. The AI interface must feel like a native extension of the brand's aesthetic.
I completely customized the Chatbase UI to match Glowell's design system:
Custom Typography & Colors: I set the widget to a sleek Dark Theme with Glowell's exact coral pink (
#FF6B6B) as the primary accent color. I overrode the default fonts, selecting a clean, modern sans-serif (Inter) to match their website typography.Bespoke Branding: I replaced the default chat bubble with a custom-designed launcher icon (a minimalist water droplet and leaf) and uploaded a bespoke agent avatar.
Initial Engagement: The widget doesn't just sit there waiting. It greets users with a custom message: "Hi! I'm Glowell's skincare expert. Tell me about your skin and I'll build you a routine," alongside three pre-set suggested buttons to drive immediate interaction.
Customizing the UI, typography, and colors so the AI feels like a native extension of the brand.
Phase 6: The AI-Powered Helpdesk
The final piece of the puzzle is the human team. I configured a Helpdesk to catch anything the AI couldn't handle automatically.
The Shared Inbox: If a web chat gets too complex, it routes straight to the human team's dashboard.
AI Email Drafting: When a customer sends a complex email, the AI automatically reads it, checks company policy, and types out a draft reply. The human agent just has to review it, tweak a few words, and click send. This turns the AI into a powerful co-pilot, helping your real team work significantly faster.The AI auto-drafting email replies for the human support team.
The AI auto-drafting email replies for the human support team.
The Build Philosophy
This showcase highlights my core philosophy: AI shouldn't be a black box that you hope works well. By combining strict rules, step-by-step logic, and deep integrations into your real business data, I build reliable, safe systems that founders can actually trust with their customers.
Ready to automate your support without sacrificing quality?
If your e-commerce brand is drowning in routine tickets, or you want to turn your support widget into an active revenue driver, let's talk.
Get in touch to discuss a custom AI architecture for your business. Book here
Like this project
Posted Aug 22, 2026
Engineered a 24/7 AI agent that automates support tickets, tracks live orders, and drives cross-sales for a premium brand.
Likes
1
Views
0
Timeline
Aug 22, 2026 - Aug 22, 2026
