Dynojet Research · Belgrade, Montana · Global powersports
Same storefront, same traffic, same products. The only difference is whether the visitor asked a question.
Dynojet's tech team covered phones and support tickets from 8 AM to 5 PM Pacific, Monday through Friday. Any inquiry submitted outside that window sat until morning.
Most riders have a day job. The bike gets worked on late at night, after dinner, or over the weekend. That means critical questions like "Will this part fit?" or "Why is my engine running lean?" usually pop up at the exact moment no one on the Dynojet tech team is at their desk. A rider stuck mid-project at 9 PM had two options. Struggle through the website alone, or send an email and wait for the next business day.
The website didn't offer much help either. The fitment finder was buried, forcing riders to wade through long product lists without finding a definitive answer.
Even if a rider found the correct product page, determining fitment on a modified bike isn't a simple lookup. A stock 2019 Harley-Davidson Road Glide takes one specific part. Bolt on an aftermarket exhaust, and the required part changes. Flash the ECU, and it changes again. The rider knows every modification on their bike, but a static website only sees a basic year, make, and model.
Dynojet possessed all the right answers. They were buried in install guides, wiring diagrams, service bulletins, and the heads of the tech team. But none of that knowledge was reachable at 9 PM. And the question that stops a build is the exact question that kills a sale when no one is awake to answer it.
Standard e-commerce chat tools rely on simple, static lists of frequent questions. Dynojet's questions are nothing like that.
In powersports, an incorrect answer carries severe, real-world consequences. Recommending the wrong tuner doesn't only waste time. It can destroy an engine during setup. To make matters worse, once a part like a Power Vision or Power Commander is installed on a bike, store policy means it can no longer be returned. If a rider buys and installs the wrong unit, they are stuck with an expensive piece of hardware that doesn't work for them.
On top of physical fitment and strict return policies, the system also has to navigate state laws, like knowing which Power Vision models can legally ship to California.
Building an automated assistant to handle these engineering rules and shipping constraints with zero room for error was a massive challenge. Both Brock and Klaus were built to catch those mistakes before a rider hits checkout.
An Ask AI button on dynojet.com. Riders ask about their vehicle, their modifications, and what actually fits, and get an answer in the moment.
Digs straight into Dynojet's own install guides, wiring diagrams, service bulletins, and specs, indexed deep enough that a question about wire routing pulls up the actual diagram.
Decodes the VIN to pin down the exact vehicle, then pulls live pricing, stock, and order status instead of guessing.
Responds in the customer's own language.
The same help by phone, for the people who would rather talk than type. Launched to cover the hours the team was closed, then moved to 24/7.
Handles fitment, product, and troubleshooting calls end to end, in any language.
When a call needs a person, writes the ticket into the right department's pipeline so the correct tech picks it up in the morning.
Confirms the caller's name, number, and vehicle before the handoff, so nothing gets chased down twice.
Nobody at Dynojet had time to listen back through a year of phone calls. Klaus did. He read back through customer conversations and produced a ranked list of 49 issues, grouped by category and severity, with 92 callers behind them and a recommended fix on each.
One of the three critical issues was a documentation error on the website that could have damaged a customer's ignition module. Because Klaus surfaced it, the team could see a gap in the system and acted on it.
Visitors who talked to Brock bought at 6.4%, against 1% for those who did not.
452 visitors chatted with Brock. 29 bought, spending $9,528.57 between them.
554,792 messages, roughly 94 a day, with zero support headcount added.
634 positive ratings out of 679 total rated conversations.
The 452 figure counts tracked storefront browser sessions only. The 34,215 is the assistant's own total and includes Help Center and Dynojet University traffic that carries no analytics.
Ask. On the site or on the phone, at any hour.
Runs on Chatbase, pulling from a vector knowledge base built out of Dynojet's own engineering and support files.
Runs on ElevenLabs, handling calls end to end in any language.
Connects both agents straight into the live business, so a price, a stock count, or an order status comes from Dynojet's system itself, never from memory.
Klaus writes directly into HubSpot, opening contacts, logging notes, and queuing callback tasks in the right department's pipeline.
Every link Brock sends tags source, campaign, SKU, and vehicle straight through to GA4.
A wrong answer costs a customer real money, so Klaus runs a fixed chain: knowledge base first, then the website, then Brock, then a ticket for the tech team if all three come up empty. No skipping, no guessing. That chain is where 252 of Brock's sessions came from, one system asking the other for an answer it wasn't allowed to make up.
Both agents rank their sources in a fixed order: the engineering documents first, dynojet.com second, the other AI third, and their own training data last for anything touching fitment, part numbers, or pricing. When a document and the other system disagree, the document wins.
If a call ends while Klaus is mid-troubleshoot and he has contact information, the callback task is created anyway, without the caller asking for it. Support failures are rarely dramatic. They are usually someone whose call dropped and who never heard back. That path is closed by default.
Every link Brock sends has carried source, campaign, SKU, and vehicle through to GA4 since the day it launched. That is why the numbers above exist. Nobody had to go back and add tracking to find them.
Most of what I build starts the same way. Somebody describes the task they do manually every week and already knows there has to be a better way. Tell me what yours is and I'll be straight with you about what's possible.