70-80% of Support Queries Can Be Automated, But Only If the AI Doesn't Hallucinate
Most ecommerce merchants who have looked at generative AI for customer support hit the same wall: the bot confidently makes things up. Ashu Dubey, CEO and co-founder of Gleen AI, built his entire company around solving that single problem first. According to Ashu, 80% of Gleen's technical work is focused on preventing hallucination, and they were the first company to publicly claim and demonstrate a zero-hallucination AI product. For serious retailers, he argues this is not a nice-to-have. An AI that gives customers wrong answers 50% of the time is worse than no AI at all.
When hallucination is properly controlled, the upside is concrete. Gleen targets automating 70% to 80% of inbound support queries, covering common questions about order status, returns, and store policies. That reduction in support volume directly lowers operating costs while freeing agents to handle the tickets that actually require human judgment.
Why You Shouldn't Build This Yourself With ChatGPT
A DIY prototype using ChatGPT is achievable over a weekend. Shipping something production-ready is a different problem entirely. Beyond hallucination, a self-built solution requires integrations with your Shopify store, your FAQ content, and your help desk platform, whether that's Zendesk, Gorgias, or another tool. Each of those integration points needs to be built, maintained, and updated whenever your store data changes. Without a dedicated development team, that work can stretch into months. The maintenance burden alone, keeping the AI current as products, policies, and pricing evolve, makes the DIY path impractical for most merchants.
How to Evaluate Any AI Solution Before You Deploy It to Customers
Ashu outlines three areas merchants consistently overlook when assessing AI tools. First, maintenance. A generative AI solution is not a set-it-and-forget-it install. It needs ongoing feedback loops so the model can correct mistakes over time. When evaluating a vendor, ask specifically how you provide corrections and how quickly they propagate. Second, security and compliance. Your customer data needs to stay protected, and if you operate in regulated industries like healthcare, HIPAA compliance becomes a hard requirement. Third, and most important, test the actual business impact. Vendors, including Gleen, will describe best-case outcomes. Your job is to verify those outcomes in your specific store environment before committing to a full rollout.
The Crawl-Walk-Run Framework for Merchants Who Are Skeptical of AI
For merchants who are not ready to let AI interact directly with customers, Ashu recommends starting with internal use only. In this model, support agents query the AI themselves, review the suggested answer, edit if needed, and then send it. Customers never interact with the AI directly. This builds team familiarity and lets you evaluate answer quality at zero risk to the customer experience.
The next step is embedding the AI inside your existing help desk interface so agents can pull AI-generated answers with a single click without switching tools. Once answer quality has been validated internally, you can begin exposing the AI to customers, but incrementally. Ashu draws a direct parallel to A/B testing: start at 1% exposure, move to 5%, then 50%, then 100%. Each threshold gives you data and confidence before the next step. The worst outcome, he notes, is doing nothing at all while competitors move forward.
What Gleen Actually Does Beyond Customer Support
Gleen is positioned as a platform rather than a single-use tool. Beyond automating support tickets, it functions as a virtual store concierge that helps shoppers find products and make purchase decisions, targeting conversion rate improvement alongside cost reduction. The platform also supports omnichannel deployment, meaning the same AI configuration can operate across your website, email, WhatsApp, and Instagram without rebuilding logic for each channel — a concern that Rep AI covers from a different angle in their episode on replicating in-store sales experiences on Shopify. For teams with multiple use cases, separate AI profiles can be created within a single account, covering support, product discovery, and even replacing on-site quiz tools.
On the compliance side, Gleen holds SOC 2 Type 2 certification, which Ashu describes as the highest compliance grade available in the US market. For merchants working with enterprise retailers or handling sensitive customer data, that certification matters in vendor evaluations.
For merchants already on Shopify, setup is designed to take under five minutes through the Shopify App Store. Searching for Gleen Accurate Answers pulls up the app, and the core integrations with platforms like Shopify, BigCommerce, Zendesk, and Gorgias are pre-built. Custom integrations outside that list are handled by the Gleen team and targeted for completion within one to two weeks.
Key Lessons From This Episode
- Hallucination is the primary barrier to production-ready AI in ecommerce. Any solution you evaluate should have a clear, verifiable answer for how it prevents the AI from fabricating information.
- A ChatGPT prototype can be built in a weekend, but production-grade integrations with Shopify, your help desk, and your product catalog can take months without a dedicated engineering team.
- Always test metrics impact in your own store. Vendor claims about conversion lift and support automation rates need to be validated against your actual traffic and ticket volume.
- AI adoption does not have to be all-or-nothing. Using AI to assist internal agents before exposing it to customers is a low-risk starting point that builds confidence and catches errors early.
- When rolling out AI responses to customers, treat it like an A/B test. Start at 1% exposure and increase incrementally based on measured outcomes.
- Omnichannel consistency matters. Managing separate AI configurations for your website, email, and social channels adds unnecessary complexity. Look for solutions that handle all channels from a single setup.
Hear the full conversation with Ashu Dubey, including his breakdown of Gleen's origin in Discord communities and how the product roadmap is shaped by merchant feedback, in the complete episode and transcript below.
In This Conversation We Discuss:
- [00:40] Intro
- [01:47] Bridging Discord and Ecommerce gaps
- [03:08] The importance of reliable AI in business
- [04:25] Balancing AI risks and benefits
- [04:49] Limitations of ChatGPT integration
- [06:09] Aligning AI abilities with customer expectations
- [07:05] Improving your AI tool through feedbacks
- [07:35] Safeguarding customer information
- [08:35] Editing and customizing AI responses
- [09:43] Incorporating AI into help desks
- [10:26] Testing AI exposure levels on customers
- [11:03] Seamless Setup with Gleen AI
- [12:10] Gleen’s quick turnaround for integration requests
- [12:27] Unified suite of AI services with Gleen AI
- [14:17] Gleen’s special offer for Honest Ecommerce listeners
- [15:01] The omnichannel capability of Gleen
- [15:50] Find Ashu on LinkedIn and Twitter/X
Resources:
- Subscribe to Honest Ecommerce on Youtube
- Generative AI platform for customer success https://gleen.ai/
- Follow Ashu Dubey https://www.linkedin.com/in/ashudubey
If you’re enjoying the show, we’d love it if you left Honest Ecommerce a review on Apple Podcasts. It makes a huge impact on the success of the podcast, and we love reading every one of your reviews!
Transcript
Read more

Mussayab Ehtesham built DMoose from a single fitness product in 2017 to a near 8-figure brand by applying project management discipline to ecommerce. He shares how an Amazon suspension pushed him t...

Shaina Rainford built Bask & Lather Co. from unlabeled bottles sold on Instagram to a million dollars in revenue within eight months, all before running a single paid ad. She shares how she val...
