Martin Alcock
All work

Tubr: from chat to feed

Should a forecasting app be something you talk to, or something you glance at?

Tubr uses machine learning to predict trade for hospitality businesses. I moved the app from chat-first to a feed of insights, then changed how the team ships.

Organisation
Tubr
My role
Fractional Chief Product Officer
Period
2026–present
Type of work
0 to 1
Capabilities
Product direction, Product design, Front-end development, Technical leadership, AI-assisted delivery
Sectors
Hospitality

The starting point

When I joined, the app was chat-first: you asked it questions about your forecast. In my first six weeks I rebuilt and replatformed the customer experience, in time for a launch in the UK and US.

Chat or feed?

  • Considered

    Chat-first

    The existing approach. Flexible, but the user has to know what to ask.

  • Chosen

    Information-first feed

    Push insights to the user as a feed, with a dashboard view of what's coming.

Decision

I chose the feed. It puts the information in front of people instead of making them ask for it. Later I built it into a system of record: people can comment on forecasts and look back at what actually happened.

Making the feed work

I designed the data model for a multi-tenant feed across companies and their outlets, so the right people see the right insights by user, role, outlet and company. Per-user state lives apart from the generated content, so regenerating an insight can't wipe out someone's comments.

How the team ships

When I arrived, deploying the infrastructure meant running Terraform from someone's laptop. I've been moving deployment into the merge process. The newer parts of the infrastructure now build a preview environment for each pull request, and tear down any that haven't merged after seven days.

I lead a lead data scientist and an ML engineer, and still ship front-end and back-end code myself. I've moved the team onto AI-native ways of working, including agentic workflows for parallel development and bug fixing, and added PostHog instrumentation so we can see how the app is used.

Where it is

Still early. We're working towards launch, partners are coming on board, and the first customer found us on their own.