by George Nott

CIO50 2017 #5: Paul Keen, Airtasker

Nov 21, 2017
Technology Industry

The rise and rise of the so called gig-economy is in no small part down to platforms like Airtasker.

The company has emerged as the leading platform for individuals to access tasks – from assembling IKEA furniture to apartment cleaning to lawnmowing – worth a total of $11.5 million a month.

As of this month, the platform has 1.7 million registered users (growing at a rate of 20,000 a month), a number likely to increase rapidly as the company expands to overseas markets. Since it was founded in Sydney in 2012, the business has transformed from one with 10 per cent unprompted brand awareness to a household name.

In October the company announced it had raised $33 million in a funding round to grow its business in the UK.

It is a company in ‘hypergrowth’. Its priority for the last year has been to scale the company’s platform, people and process to support its expansion. Leading that charge is chief technology officer Paul Keen.

Keen arrived at Airtasker at the beginning of 2016, from retailer Dick Smith, where he was CIO.

Tasked with building a scalable and solid infrastructure that could handle the massive spikes in traffic the website would be experiencing, Keen first took Airtasker to the cloud.

His team migrated the platform from running on just four managed servers (“less powerful than your MacBook Pro,” Keen says), to an elastic cloud solution that allows the company to deal with huge traffic spikes resulting from TV content integration (like during its award-winning adsaired during the AFL grand final half-time break last year).

“Integral to the migration, we implemented a strong DevOps capability with immutable infrastructure that allows automated scripts to rebuild our environment from one-click,” says Keen. “Deployments can be made by anyone in the engineering team simply by typing in Slackbot commands.”

The massive migration work complete, focus moved to ensuring every Airtasker user’s experience is slick and secure, no matter how high membership numbers get.

Innovating for growth

A recent development is Carl – named after Carl Linnaeus, the Swedish, 18th Century botanist known as the ‘Father of Taxonomy’ – a machine learning tool that detects different tasks, groups them and creates category names.

“Airtasker has never categorised its tasks, not wanting to limit what users can do on the platform,” explains Keen. “Tasks could be traditional jobs like cleaning and gardening, but unexpected trends appear, such as hiring someone to queue for the latest iPhone. By letting the marketplace naturally decide, we can immediately leverage these opportunities.”

To train the model, Keen’s team manually classified 200,000 tasks then further enhanced the model to have 92 per cent confidence level on categorising tier one tasks (for example ‘cleaning’) and 83 per cent confidence categorising tier two tasks (for example ‘end of lease cleaning for renters’).

“Using this data, we can now ensure each task on the platform, remarket based on activity and help provide guidance to users to better describe and price tasks. This ensures both parties are happy with the outcome,” Keen adds.

Another fix for Airtasker’s massive growth has been Airgun, which pushes new task post notifications to other services.

“As a two-sided marketplace the activity on the Airtasker platform is exponential to user traffic. A single task posted on average would receive three bids and eight comments, as well as hundreds of notifications letting Airtasker users wanting to perform a task know about the tasks,” Keen says.

In 2016, notifications to users – sent via email or push notifications – rocketed from 10 per second to 300 per second, totalling 150 million notifications a month.

“To deal with these asynchronous loads, we built a service called Airgun,” Keen says.

Airgun is an event-based service that moves data downstream to third party applications in an “eventual consistency” manner.

“When a new task is posted on the platform, a number of services would like to be aware of the content. We first push the event to a message queue that triggers a Lambda serverless function.

The Lambda service understands what downstream services need to be aware of the service – for example Elasticsearch or Salesforce – and sends the data to the appropriate message queue. Finally, another Lambda serverless function is triggered performing the integration at the throttling limit for that service,” Keen explains.

“Airgun uses the AWS technology Kinesis and Lambda to have highly redundant, highly fault tolerant systems that provide virtually limitless scalability at a very low cost. As part of giving back to the community, Airgun is open sourced for others to use.”

Mentor, not dictator

Keen considers his role being “to mentor the team, not dictate”.

Feature teams – which are focused around a particular customer group – are given autonomy and are self-organising, which gives members their own motivation.

“Our team is intrinsically motivated in their roles as they get to choose what they work on,” Keen adds.

As Airtasker grew from 30 to 65 employees, the company moved to a Spotify feature-team model to “keep the nimble start-up roots of building, testing and iterating quickly, allowing our team to take ownership of their ideas”.

The result is a diverse team, with a number of nationalities represented. Keen is now the “token Australian”, he says.

“As our marketplace is diverse, we also recruit with diversity in mind with engineers coming from South America, Philippines, Canada, France, as well as the token Australian.”

Team members can also take ownership of their own development.

“We ensure day-to-day learning is gained through challenging code reviews and sharing articles and concepts on Slack. Every fortnight, we have a Tech Council providing a forum for engineers to share new concepts that can be adopted across the business. Each engineer has a formal learning path with budgets assigned to ensure they meet with their learning goals,” Keen says.

There are the ”standard” ping pong table, Friday drinks and card games, Keen says, but again, teams can choose their own extra-curricular activities, given budgets “to do things out of the ordinary from cooking classes to axe throwing”.

“At Airtasker we focus on three main areas: culture – to like and be inspired by the people you work with; to work on things you think will make a difference; and growth in terms of your day-to-day on the job learnings and formal growth,” Keen says. “Our belief is that without focusing on all three pillars simultaneously, our team won’t be fulfilled in their role.”

As their work over the last year shows, Keen’s team is driven, fulfilled and up to the task.