Mike HeapMike is an experienced Product Manager who focuses on all the “non-development” areas of My AskAI, from finance and customer success to product design, copywriting, testing and more.
Table of Contents
- What does Barn Owl do?
- Which helpdesk does Barn Owl use?
- How did Barn Owl train their AI customer service agent?
- When did Barn Owl decide to turn on 'direct replies' to customers?
- What was the biggest thing Barn Owl did to improve their AI agent's resolution?
- How do Barn Owl customize their AI agent setup to work for their business?
- Guidance: voice, escalation triggers, handover rules
- Two Tasks: camera troubleshooting and manage subscription or account closure
- What impact is Barn Owl's AI customer service agent having now?
- Where do Barn Owl go from here?
Created time
May 26, 2026 09:56 AM
Title length (<60)
Author
Ecomm?
Tags
Image
Publish date
May 26, 2026
Slug
my-askai-barn-owl-case-study-2026
Featured
Type
Case study
Ready to Publish
Ready to Publish
Barn Owl resolves 47% of ~700 monthly Gorgias tickets with AI, holding 80% AI CSAT and saving ~27 hours every month.
Barn Owl sells off-grid cellular cameras to ranchers, farmers, contractors and remote-property owners. Their buyers are out tending animals or crops during business hours, and only get to a screen after dark. This leads to a variety of questions: technical questions about firmware, signal, solar, battery, SD cards, all landing on a small team.
Four to twelve thousand tickets a month, much of it after-hours, is a lot of support for a seed-stage operation. Barn Owl had already tried the AI agent baked into Gorgias and pulled it out before finding us. Today, across the chat and email tickets routed to the AI, ~700 a month land with the agent. 47% get resolved without a human, AI CSAT sits at 80%, and the team saves roughly 27 hours every month at five minutes a ticket.
Here's how it came together.
What does Barn Owl do?
Barn Owl makes off-grid cellular security cameras for places that don't have power or Wi-Fi (their own tagline: "built for rural life"). The hardware is the camera plus solar panels, batteries and cellular antennas. The recurring revenue is the cellular plan attached to it, sold contract-free at $7, $15 or $25 a month depending on data allowance.
The company runs out of Colorado, is Veteran Owned and Operated, and lists "more than 7,000 customers across Rural America" on its homepage. AgriBusiness Review named it Top Farm & Ranch Camera Company of 2026.
The buyer mix spans hunters and game managers, ranchers and farmers, oil and gas operators, and contractors on remote build sites. There's also a smaller-but-high-stakes segment: local governments and law enforcement (cities deploying cameras for public-safety reasons).
A VIP rancher with 15 cameras across one property, or a city running the network as a public-safety asset, is not a customer where a botched answer is recoverable.
Which helpdesk does Barn Owl use?
Barn Owl runs support inside Gorgias, with chat embedded on the storefront and email tickets flowing into the same inbox. Their Shopify backbone keeps order and customer data accessible from the same place (our Gorgias integration is the seam we sit in).
Before us, Barn Owl had Gorgias's own AI agent (Gorgias Automate) running on those same tickets. They pulled it out. The product was hallucinating; it was pulling answers from places no-one at Barn Owl had ever put on the platform; and the per-resolution pricing meant Barn Owl was being charged when the AI was wrong.
Two failure stories from that period made the case for ripping it out unambiguous (and we've seen similar shapes on calls with other Gorgias Automate customers). In the first, a VIP with 15 Barn Owl cameras asked the AI how to reset one of them, which is a standard, well-documented procedure the agent had answered before.
This time it returned advice on how to resolve a dog's loose-stool problem. That information had never lived anywhere in Barn Owl's documentation.
In the second, a customer was given a Google Maps link to a location in Sri Lanka, despite Barn Owl not selling cameras outside the United States. Barn Owl was charged $1 for each of those resolutions. You can take the wider Gorgias-Automate-is-underwhelming framing with a grain of salt if you like; the dog-stool ticket is harder to argue with.
Barn Owl then went looking. The closest match they found before landing with us was Redo, whose in-house AI guidance and training tooling they described as exceptional.
Redo, though, only runs inside Redo's own customer-service product. There was no path to keep Gorgias and add Redo's AI on top, and for a Shopify+Gorgias team that wasn't replacing its helpdesk, that was a dead end.
We offered the part Barn Owl had liked about Redo (the depth of training and guidance control), built to live inside Gorgias instead of around it. The integration is native; our per-resolution price came in 5-to-8x cheaper than Gorgias Automate at the volume Barn Owl was running; and our architecture isn't single-model, which had been a separate concern after Gorgias announced its own AI was moving to be 100% powered by a single underlying model.
How did Barn Owl train their AI customer service agent?
Barn Owl already had a lot of knowledge written down. It just lived in several different places. The training plan connected each of them in turn:
- Gorgias help centre: the first source, capturing the public-facing answers Barn Owl's team had been linking customers to for years.
- Shopify: wired in next, both for product information (cameras, plans, accessories) and the live order data behind every WISMO and warranty question.
- User Data API: the customer-specific data layer on top. Which cellular plan a customer is on, when they bought, what serial numbers are registered to them, whether they're a multi-camera VIP.
- Google Drive plus an internal site Barn Owl calls "Papers": the deeper product knowledge their team uses internally (the kind of context the public help centre doesn't carry).
- Custom Answers: a small set filling in the hardware-specific edge cases the help centre didn't yet cover. Firmware-version specifics, signal-strength troubleshooting, solar-and-battery checks for the more remote installs.
- Self-Learning: switched on across the lot. Every time a human agent closes a ticket the AI couldn't, we draft a new knowledge article from that exchange. The library grows on its own (useful for a team without a full-time CX-ops person sitting on top of it).
The historic-ticket-training step was where Barn Owl's situation differed most from the canonical playbook. Most ecommerce brands lean on years of past tickets as a training source. Barn Owl had 70,000 historic tickets sitting in Gorgias, but a large share of them covered camera hardware that's no longer being sold and firmware that's since been replaced.
We cap historic-ticket training at 5,000 by default, which Barn Owl treated as a feature. The cap forced selection toward the most recent, most representative slice of the backlog instead of teaching the AI on a generation of obsolete answers. Most other vendors import all-or-nothing on the historic backfill (fun fact: that was Christina's biggest single relief on the call).
Beneath all of that, our agent runs across a mix of models from OpenAI, Gemini and Anthropic rather than a single provider. Different steps in a single answer (rewriting the question, language detection, applying guidance, retrieving the right knowledge) route to whichever model handles that step best. The routing is A/B tested continuously against resolution rate.
When did Barn Owl decide to turn on 'direct replies' to customers?
Barn Owl went direct from day one across both chat and email (with us in the inbox from go-live). No internal-notes mode, no copilot-only stage. The reasoning came down to scope rather than bravery.
Forty percent of Barn Owl's total monthly support volume comes in through phone, which sits outside the AI's reach by design. Inside the 60% of volume that arrives via email, chat, Facebook comments and social messages, Barn Owl rolled the AI out on a sub-slice rather than the whole inbox. The combination (a defined channel scope plus the handover-and-escalation guidance below) meant a smaller blast radius than a notes-mode pilot would have produced anyway.
Across those routed channels today, ~700 monthly tickets land with the AI. Anything safety-critical, anything involving warranty claims or RMA, anything from a known government or law-enforcement account, escalates to a human before a customer sees a reply.
What was the biggest thing Barn Owl did to improve their AI agent's resolution?
Two things, working together: the Self-Learning loop on top of the live data layer.
The live data layer is the more concrete half. Once Shopify and the User Data API were connected, every "where is my order?" became a real lookup instead of a templated apology. The AI could see the order date (the field Barn Owl uses as the warranty start) and answer warranty questions against the camera's specific purchase timeline.
For multi-camera customers like our 15-camera VIP rancher, our AI could see which plan they were on and how many serial numbers were registered to them, then route the conversation accordingly. The class of ticket that used to default to "please contact support" turned into a class of ticket our AI could close.
The Self-Learning loop is the half that compounds over time (and the half we lean on hardest for seed-stage teams without a dedicated CX-ops reviewer). Every ticket where the AI didn't know the answer became a piece of training data the moment a human closed it.
Our AI compared its own attempt against the human reply, identified the gap, and drafted a knowledge article that fed back into the next answer.
Barn Owl's team didn't have to schedule a weekly review session to do this. It ran in the background.
Sitting underneath both of those is our answer-traceability tooling. When Barn Owl's team needed to double-check how the agent had arrived at a response, our inspect view showed the exact lines from the exact source documents the AI had quoted from.
In the early weeks, that's how Barn Owl caught documents left in the training set by mistake, surfaced pages with ambiguous wording, and tightened the guidance where our AI was over-reaching. Without it, the same step would have required guesswork (and after the Gorgias Automate experience, guesswork was the thing Barn Owl was actively trying to leave behind).
How do Barn Owl customize their AI agent setup to work for their business?
Two customizations carry the post: how Guidance shapes voice and escalation, and the two Tasks Barn Owl built for the recurring hardware workflows.
Guidance: voice, escalation triggers, handover rules
Barn Owl's communication guidance is built around the older, rural demographic the brand serves. Clarity over cleverness, no emoji, no marketing fluff. Plain instructions a customer can follow with one hand on a phone in a pickup truck.
The clarification layer handles the multi-camera context (which plan, which camera model, single-camera or multi-camera customer) before the AI commits to an answer. The wrong assumption is more expensive than a single follow-up question.
The handover-and-escalation layer is where the high-stakes scoping happens. Warranty claims, RMAs, faulty units, anything involving fraud, anything from a known government or law-enforcement account all escalate before a reply goes out. Our escalation logic reads intent rather than keywords (so an older rancher saying "this dang thing won't turn back on after the storm" gets treated as a faulty-unit signal and routed to a human, not as a generic troubleshooting question to be answered by the AI).
Two Tasks: camera troubleshooting and manage subscription or account closure
A Task is a structured workflow our AI runs when a specific kind of question comes in. Barn Owl built two:
- Camera troubleshooting covers the most-frequent technical thread. Firmware version, signal-strength check, battery and solar status, SD-card state. It walks a customer through the standard diagnostic flow Barn Owl's human agents would have walked them through, then either resolves the ticket or hands a fully-contextualized case to a human if the camera is faulty.
- Manage subscription or account closure covers the other recurring thread. Billing changes, plan upgrades and downgrades, account closure. Closure is high-stakes for Barn Owl. Closing a customer's account means the cameras attached to it go offline (and a rancher who came in expecting to pause for the winter doesn't want to find his cameras dark in November).
So our Task collects the closure context, surfaces the implications, and either processes the change or escalates with full context. Both Tasks sit on top of the same live-data layer our AI is already using for order lookups. They run autonomously where policy allows; they escalate cleanly where it doesn't.
What impact is Barn Owl's AI customer service agent having now?
- 47% AI resolution rate across the routed chat and email tickets
- ~700 tickets handled per month by the AI (~329 resolved without a human, ~371 escalated)
- ~27 hours saved per month at five minutes per ticket
- 80% AI CSAT across those resolved tickets
That 700-a-month sits inside a brand total of 4,000 to 12,000 monthly tickets, of which 40% comes via phone and is outside the AI's scope by design. Our footprint is deliberately a sub-slice of the AI-eligible channels rather than the whole inbox: chat and email tickets first, with the harder VIP and warranty paths routed to humans before the AI gets a chance to misread them.
For a seed-stage team running a high-touch hardware brand, the headline is what our AI is doing well: the right ~700 tickets a month, on Gorgias, with a 5-to-8x cheaper unit cost than the agent Barn Owl ripped out. (For a similar story on a different ecommerce helpdesk, see how Edel Optics moved from 25% to 79% AI resolution on Zendesk once the User Data API was wired in.)
Where do Barn Owl go from here?
Two threads on the roadmap.
The first is more autonomous Tasks. The order-update, refund, RMA and warranty workflows are next in line.
Barn Owl already asked us specifically about a warranty Task that would check the Shopify order date against the warranty policy, then either process the claim or escalate cleanly. That's exactly the kind of workflow where the live data layer pays off twice over (and the same pattern extends to address changes, exchange requests and replacement orders).
The second is a deeper Self-Learning corpus. The 5,000-ticket historic training default was the right starting point given the obsolete-hardware skew in Barn Owl's backlog.
But as the steady-state ticket flow accumulates inside our product, and Self-Learning keeps drafting new articles every week, our AI's understanding of the current camera generation only grows. For specific high-stakes segments like the government and law-enforcement accounts, layering in deeper historic training over time is on the table.
The reframe Barn Owl is settling into is that 100% automation isn't the goal. With 40% of volume on phone and a meaningful share of email and chat routed straight to humans by design, the realistic ceiling for our AI is probably 60-65% of the AI-eligible channels. Either way, that's plenty: a 47% resolution rate at sub-1k AI throughput is already saving a small team roughly a workweek every month, and the trajectory is up.
If you'd like to see more customer stories like Barn Owl's, browse all our case studies here. Or, if you want the same setup on your own Gorgias inbox, check our Gorgias integration page and pricing.
