WhatsApp
Tecifo Solutions logo: orange hexagon cube with blue hexagon dotsTecifo.Book a call
← All work
AI/MLgotobeat.com

Gotobeat

Technology and data OS for the live music industry. We built the prediction and agent layer behind artist and venue demand forecasting.

Sector
Live music data
Engagement
Extend — AI layer on a live platform
Timeline
Sep 2024 → Jun 2025
Team
1 ML engineer, 1 backend engineer
01 — The problem

Gotobeat could see every ticket sale in its network and still could not tell a promoter whether a show would sell.

Four years of gig data — artists, venues, capacities, ticket curves, cities, support acts — sat in Postgres and got used for reporting. Booking decisions were still made from instinct and a spreadsheet, and a mistake meant a half-empty 400-capacity room or an artist turned away from a venue that would have sold out.

Internally the same knowledge problem repeated. Answering "how did this artist perform in Manchester last spring?" meant asking the two people who remembered, and promoters were sending those questions to a shared WhatsApp number at all hours. Nobody was answering them at 11pm on a Friday, which is exactly when promoters ask.

02 — How we solved it

What we built, and why in that order

01
Prediction Agent
An XGBoost model over historical sell-through, seasonality, venue capacity and artist trajectory produces the number; an LLM turns the feature contributions into a paragraph a booker can act on. The forecast is never shown without its reasoning.
02
GBrain knowledge layer
A retrieval layer over internal data, documents and past events gave the team one place to ask questions in plain language, with answers grounded in records rather than recall.
03
Hermes promoter agent
A WhatsApp agent on Telnyx handles promoter enquiries end to end — availability, past performance, submission status — and escalates to a human with full context when it should not answer.
04
MCP-style tool layer
Internal services were exposed as typed tools, so each agent calls the same audited functions instead of holding its own copy of business logic.
StackPythonXGBoostLLM agentsGeminiMCPTelnyx
Outcome
±11%
Median forecast error on sell-through
74%
Promoter enquiries resolved without a human
4 yrs
Historical gig data put to work
<3s
Median agent response time
We stopped arguing about whether a show would sell and started arguing about the number. That's a much better argument to be having.
Founder, Gotobeat
03 — Where they are now

Forecasting is now part of how Gotobeat books, not a report it reads afterwards.

Our role: AI/ML and agent engineering

Prediction Agent output appears next to every show under consideration, and the team has started using low-confidence forecasts as a signal in itself — a wide band means the artist-city pairing is genuinely untested and gets treated as a risk, not a guess. Hermes handles the overnight and weekend enquiry volume that previously went unanswered until Monday.

The platform has since expanded into more cities on the same model architecture, retrained quarterly as new ticket curves land. Gotobeat runs and retrains the models in-house now; we hand over after each expansion review.

Next case studyCleverseedAI/ProductSimilar problem?Start a projectinfo@tecifo.com

Have something that needs to ship?

Scope agreed in writing, architecture first, weekly increments you can review. Start with a 30-minute call or send the brief.