B4Vela, your race office, backed by AI
Eight months of messages from participants of one organizer's seven events. B4Vela answers them in chat, by email and in tickets.
Book a demoRace packs, refunds, invoices, distance changes. It all lands on one desk, usually the week before the start.
From January to the end of August: 2,790 real emails, with spam and auto-replies removed. In May, the sign-up peak, 599 of them. Roughly four times the off-season level.
And works out which race each email is about, even when the participant doesn't say. That's 39% of emails: the agent reads the text and the quoted thread, and asks when it can't tell.
Shares from the real inbox. One email can cover several matters, so they add up to more than 100%.
Illustration built on the real proportions of one organizer's registration inbox: seven running events, about 20,000 participants a year, 1 Jan to 28 Aug 2026.
Email topics in the inbox: race pack, QR code and bib 25.4%; withdrawal, refund and transfer 23.7%; payment and invoice 18.8%; change of details or distance 12.1%; relay teams 11.3%; results and certificate 10.8%; discounts 6.3%.
Chat, email, tickets and the report share the same knowledge of your event and the same rules. You switch on only the ones you need.
A widget on your event website. It answers around the clock from the rules, the FAQ and the site content, in the participant's language, and remembers what they just asked. It adds a link only when that helps.
Live on: Białystok Biega, Bison MTB, Biegnę dla Niepodległej, Superbiegacz, Bison Ultra-Trail.
It reads the inbox, including one shared by several events. It works out the race and the matter, then replies or opens a ticket with every detail. When an email is unclear, it asks.
On a shared registration inbox since cutover: 127 emails delivered, zero errors, a peak of 22 emails in one hour.
It works inside the B4Sport ticket panel your office already uses. It drafts a reply for one-click approval, and where you allow it, answers on its own. You widen the scope one topic at a time.
Live on: Bison MTB, Bison Ultra-Trail.
Every morning by email: the previous day's emails, tickets and chats, what the agent prepared and what a person approved, topics and languages, response times and knowledge gaps.
There is a fifth part. It works before the agent answers anyone.
It checks that your sources agree.
Before the agent answers anyone, Vela Check compares the rules, the website, the FAQ and your corrections. For Izery Ultra Trail it checked 423 facts and found 56 discrepancies. Apparent ones are filtered out: "21 hours" and "23:59" are the same cutoff, written two ways.
In a conflict, the rules win by default. The organizer always makes the final call.
In race week we go through the whole website again. Before Białystok Biega that caught three changes at once, all in the agent the same day.
Until the decision, the agent leaves that time out rather than guess.
The agent is a member of your team, not an off-the-shelf bot. Colour, name, greeting, reply style and rules are set per event. Switch the event and see.
And more: name and avatar, email signature, office hours, event phase (before, during, after), a glossary of race names, separate rules per channel. The AI notice can't be removed: that's a transparency requirement.
This is the order in which the agent handles every message, on every channel. It doesn't guess.
A participant's message, Bison MTBWhen can I collect my pack? And can I take my friend's too?
A question, a request or a complaint; which language and which race.
question · PL · Bison MTBUp to three sub-questions, each checked separately.
2 sub-questions: when, collecting for someoneThis event and this race, not the whole internet.
rules, FAQAnd permanently pinned facts, such as "registration closed".
pinned fact: collection in person onlyBefore anything goes out.
match with sources: highPayments, data changes and complaints go to your office as a ready draft, details collected.
reply sentThe rollout agent does the heavy lifting: it reads the website, compares sources, writes the FAQ and runs the tests. You decide where the decision belongs to the organizer.
Youpoint to the website, rules and inbox
Agentpulls race data from B4Sport and reads the whole site
Agentcompares sources and lists conflicts in a sheet
Yousettle them in the "Who" column
Agentwrites 137 to 178 question and answer pairs; a second model checks them against the sources
Youedit in the sheet, no email chains
Agenttests on 227 to 285 questions, each asked three times
Yousee the results, every error analysed
Youset the thresholds and decide on launch, on test data
Agentanswers; every answer is scored against its sources
Youread the report each morning
Today we run this process together with you. A fully self-service mode, in Vela Studio, is in the works.
Example thresholds. We agree them together, and you decide on launch knowingly, on test data.
Zero made-up answers is a target we approach measurably, not a day-one promise. So after launch every answer is scored against its own sources. A low score means a draft for a person, and you see it in the report.
AIDesigned to meet the AI Act and GDPR
20 minutes, on questions from your own event, not slideware examples.
One channel on one edition. See the result before you decide on the rest.
We don't publish a price list. We'll prepare a quote after a short call, matched to your participant numbers, channels and registration platform.
