Wasla turned Monday morning firefights into a pull request to approve

Wasla runs same-day delivery in nine cities across the UAE and Saudi Arabia, from Dubai and Abu Dhabi to Riyadh and Jeddah. Its five-person data team owns 420 dbt models on Snowflake, orchestrated by Airflow, feeding the dashboards that dispatch drivers every morning at 6:00.

Courier partner feeds change without warning. A renamed field or a new status code would break a staging model overnight, the failure would cascade through the marts, and Monday started with an hour of reading Airflow logs before anyone could even say what had gone wrong. Divya Krishnan, Head of Data, counted 4 hours a week of Monday firefighting across the team, plus the cost of dispatchers working from stale numbers.

Wasla connected Airflow, the dbt repository and Snowflake in one afternoon. Tessel grouped the first weekend’s 38 alerts into 3 incidents, and opened fix pull requests for two of them: a renamed column in one courier partner’s feed and a new delivery status in another.

Six months in, 91% of failures arrive as a pull request with a test attached, and Monday incident time is down to about 20 minutes, most of it spent reviewing. The team uses the hours it got back on the routing model it had postponed for a year.

Stack: dbt Core, Airflow, Snowflake, GitHub, Slack.