How Arbiter Sports scaled to 24/7 support with AI handling a third of contacts

Support that couldn't keep pace with growth

Arbiter has spent more than 40 years as a leader in K-12 athletics technology, helping schools manage registration, eligibility, scheduling, and payments. Its customers include state associations, schools, families, assignors, and officials, essentially anyone with a role in a K-12 athletic event. Over the past seven years, the company has acquired FamilyID, rSchoolToday, BigTeams, and Vertical Raise, each one adding customers, products, and support volume onto a team that hadn't scaled to match. Two of those acquisitions landed close together, one in December 2024 and another the following September, which is what made the support gap impossible to ignore.

Live chat existed, but it wasn't reliable. Of the 92,500 total contacts in the first half of 2025, only 2,700 came through Zendesk chat or messaging, because the team could only staff it one to two hours a day. KayLee Roemmich, Senior Director of CX at Arbiter, said it "wasn't a great experience for our customers who knew that we had chat, but it wasn't available to use all the time."

The acquisitions made the support gap impossible to ignore. As KayLee put it, "the real need for AI did come through all of our acquisitions." The team knew it needed a way to serve legacy customers from newly acquired companies without simply hiring its way out of the problem.

KayLee had been thinking about this since 2024, after a conference presentation convinced her that AI support only works if the knowledge base underneath it is solid first. She acted on that immediately, hiring someone internally that year to audit and tag every help article in Zendesk before evaluating a single AI vendor.

Choosing a vendor that didn't require building a workflow for every answer

Arbiter ran small demos with a few vendors in early 2025, including Zendesk's own native chatbot. How much manual setup each option required before it could answer a question was the deciding factor.

Zendesk's chatbot needed a workflow built for every individual response, and KayLee didn't see the logic in that, calling it a case of building "every single workflow" just to get the bot to "give an answer," at which point, in her words, "why are we using AI."

Assembled worked differently. Pointing the platform to a knowledge base that Arbiter’s customers were already using meant T0 and T1 answers were already answered out of the box. Workflows were reserved for the edge cases that the knowledge base couldn’t cover.

Rolling out in phases, starting with knowledge and adding workflows later

Phase 1: Connect the knowledge base and go live on help articles alone. Arbiter went live in January 2026. Brendan Larabee, who moved into a new role as Zendesk and AI Support Operations Specialist to run the rollout, called the setup "plug and play," with all of Arbiter's help articles flowing into Assembled within an hour of connecting the integration. The sync runs hourly, so new or updated articles show up in Assembled without a separate step. The bot was also instructed on which products each customer uses, and it asks users their role at the start of a chat so it can surface the right articles for that segment.

For the first five months, that was the entire setup: a clean knowledge base and chat instructions, no workflows. That alone was enough to start containing volume.

Phase 2: Add workflows for the cases the knowledge base can't handle. Around five months in, Arbiter started building workflows for situations that don't belong in a public help article, including features that were rolled back and need an explanation or requests that have to route to a customer's own association rather than to Arbiter. Each workflow defines trigger conditions, a scripted response, and links to relevant help content, and can be tested inline before going live. The user segment collected at the start of a chat also flows back into Zendesk automatically, updating case fields and driving ticket routing without manual tagging.

The customer who used to call every time

One assigner had historically preferred phone support over anything else. As KayLee described it, ever since Arbiter switched to Assembled, he "gets 95% of his questions answered through the chat." He still calls in when he needs to make an account change, but the everyday questions that used to mean picking up the phone now get handled without one.

An unexpected benefit: No dedicated AI admin needed

Arbiter didn't hire a full-time person to run Assembled. Brendan's role covers both the AI support operation and Zendesk administration generally. When another AKKR portfolio company asked what to prepare for in advance, KayLee's advice was to start light, getting going "without a vast amount of workflows" and identifying gaps as they showed up, building a workflow only once something "is not really answered in a help article." On staffing, she didn't think a dedicated hire was necessary, just "someone who can keep an eye on things, make updates, and make sure it's running properly.” During her own leave, the system ran without her. A customer support manager and a team lead covered it alongside Brendan, with nobody dedicated to it full time.

What changed, and what it added up to

Day to day, the support team went from staffing chat for a couple of hours to offering it around the clock, and from manually reading every chat's context to having user segment and routing information land in Zendesk automatically. In KayLee's own accounting, Assembled now "handled 30% of those [contacts]," and within that group, "only 10% of those were then handed off to a human agent." In other words, roughly 27% of all contacts now get resolved by AI without a human ever stepping in.

KayLee called the shift "really awesome," pointing to Assembled AI's ability to "manage our customers and help us in taking care of those customers quickly and efficiently."

What's next

Arbiter launched chat on its SR platform in July and is targeting an August launch for Assembled AI for email, and will keep watching containment on the SR platform as chat volume there builds. Once the current busy season wraps, Arbiter will look at how support capacity should evolve alongside the AI rollout.

‍