In cybersecurity, an answer can't just be fast. It has to be right
Over the last 10 years, Huntress has earned exceptional trust from customers and partners that rely on Huntress to defend their organizations from modern cybercrime. As the company scaled its Agentic Security Platform globally, protecting that trust became imperative. As a source of truth for customers and partners alike, a bad answer from support doesn’t just create friction; it chips away at the trust of customers who rely on Huntress for something as high-stakes as security and partners who’ve put their own credibility on the line with Huntress.
Adam Wasmus leads Global Support across two very different teams. Product support handles the familiar work: break/fix issues and how-to questions. SOC support is something only a security company has: incident response for customers under active attack. As Adam puts it, it's "911 for when bad things happen." When a customer reaches out mid-incident, the answer they get can't be slow, and it can't be wrong.
As the company rapidly added more customers, expanded its partner ecosystem, and scaled its product portfolio, support demand rose with it at about 5,000 support tickets a month and rising. Growth meant a bigger queue and more pressure on a team known for a 99% CSAT for five plus years with a median first response time of 50 minutes on email and two minutes on chat. The goal was to hold that bar as the company scaled.
"It's always trust but verify. If you're the source of truth, you have to be sure the information you're putting out is correct."
Any tool that went in front of partners and customers had to clear that bar first.
Why their existing tooling wasn't the answer
Huntress tried Zendesk's Answer Bot. Classified as AI, but in practice "a glorified search". Huntress' answers are more complex, and building more workflows wasn't going to close the gap. They needed an intelligent, human-like solution that could understand their product as well as their best support engineers and could reason over their knowledge base.
The market is split into two camps. Most vendors were "support companies trying to solve AI," and several required migrating off Zendesk legacy chat to messaging before their tool would work. This was a non-starter for a team that needed to stay operational and didn't want to change two things at once.
Brainfish was the inverse: an AI company solving a support problem. It surfaced Huntress' own knowledge through a search-style experience rather than a chatbot pretending to be a person, generated answers without hand-built workflows, ran inside the product, and didn't demand a messaging migration as the cost of entry.
Build vs. Buy
Huntress seriously weighed building in-house. The test was disciplined: can a commercial solution solve 80% of our problem? Yes. Which made the deciding questions the ones that show up after launch.
"Anyone can build something that works for ten users. But who hosts it? Who maintains it? Who supports it at 3AM when it breaks? Who owns the SLAs and the liability? We're in the business of delivering exceptional service, not building an AI support solution."
How Huntress protected trust before going live
Before going live, Huntress ran Brainfish against their own team first, so they'd know exactly how it handled their hardest questions before customers ever saw it.
Using the Brainfish Slack integration against their own knowledge base, the support team (people who know the answers COLD) asked real questions and rated each response. A thumbs-down wasn't a failure; it was a map. Each one surfaced a hallucination, a knowledge gap, or an article that needed fixing. The team cleaned up their product knowledge before any customer saw the agent.
"AI is a multiplier. It multiplies the good, but it multiplies the bad too. We wanted to know our knowledge was good enough that Brainfish would pull the right information before we ever put it in front of a partner or customer."
Once their team signed off, they replaced Zendesk’s Answer Bot externally. The swap was smooth, ensuring that the most customers noticed was that the experience only got better. The team watches Brainfish analytics continuously, inspecting where each answer is sourced from.
"I sleep better at night because I can look at the answers and know we're sending the right things."
The results
A queue that grew every month is now flat and declining.
- April: ~50% prevention rate; tickets ~11% below forecast and down ~8% month over month
- May: a further ~4% month-over-month drop
- To date: ~7,700 interactions removed from the queue
- CSAT: held at 99% through the transition
What the numbers mean day to day: the repetitive, low-level questions are clearing out, so support agents can work the harder tickets and handle them with the care security issues require. That freed capacity is helping the team uplevel the support experience, funding a knowledge revamp and a live chat migration the team couldn't have started while buried in tickets.
Support teams are understandably wary of AI, and the headlines don't make it any easier. What this team found, though, was the opposite of what they'd braced for. As the volume of repetitive questions dropped, their work didn't vanish, it got more interesting. With the routine tasks handled, they had time to focus on the harder, more complex tickets that actually require judgment and a human touch, and the job itself leveled up alongside them. That was when "leveling up, not replacing" became the truth instead of a slogan.
Prevention, not deflection
The distinction is the philosophy.
Huntress’ support leader has been on the other side of a deflection story plenty of times. A customer already knows they need a person, and the bot keeps negotiating: what do you need help with, maybe I can help. They say they just want a human, and it asks again what they need a human for. Every attempt to reach support gets met with one more loop. That experience is exactly what Huntress refused to ship.
"Deflection is when you don't want people to come to support. Prevention is when you don't need them to but if they do, by all means, come to us."
Customers come to support because they want to get on with their day, and reaching out is already friction. So the goal was to take that friction away rather than add to it. Brainfish answers each question so that most people never need a ticket, and the ones who do can get to a person without fighting for it.
What's next
Brainfish is Huntress' AI support engine, in their product and on their website. The roadmap is widening: migrating from Zendesk legacy chat to messaging to open up more integrations; using the Brainfish MCP to auto-update knowledge base articles with Claude; migrating into the Brainfish knowledge base for usage-based article suggestions; and L2 ticket automation work with broader Brainfish functionality.
All of it serves one main purpose: building a support org that keeps pace with Huntress’ AI-centric platform, supporting five products today and a sixth before the end of 2026, while holding the trust it spent years earning.
"We're not trying to replace folks. We're trying to level them up and move the repetitive work out of the way so they can focus on the harder problems."
— Adam Wasmus, VP of Global Support, Huntress