Full disclosure. When the head honchos at Mondo called and asked me to write a case study for a project we delivered for one of Australia’s leading supermarkets I said no. I don’t think it’s a good idea, I told them. There’s no quantifiable success metrics and honestly we can’t conclusively prove our hypothesis. It just doesn’t fit the traditional case study format – here’s the problem, here’s what we did to deliver a solution etc. I told them it was not ‘case-studiable’. They told me that’s not even a real word. They just said something cryptic about the journey, not the destination something something. They politely asked me to get on with it and ended the call.
This online community is all about food, freebies and friends. Real people. With real stories. A community where members score products for free, redeem them via their loyalty card, contribute honest reviews and share them online. The Mondo crew worked with this customer a while back to build an automated campaign platform that replaced a legacy system that was taking 24 hours to run a campaign simulation and had a poor hit rate when allocating products to community members. So they already knew what Mondo was about – small teams moving fast to deliver business value from day one. Despite increased automation, the product allocation process was still proving to be highly manual, resource hungry and labour intensive.
With over 80k members and 1.4 million sample offers to allocate, this approach was resource intensive and simply not scalable.
In short, they needed a fresh approach, so they asked Mondo to come to the party.
In partnering with Mondo, this customer boldly embraced something different, and for them, something radical. Utilising serverless architecture, cloud computational heft and Google’s BigQuery data analytics platform, we agreed on trialling an experimental machine learning solution to the problem of product allocation.
The Mondo proposal was simple – build some ML models and train them to allocate the right products to the right members. This should reduce manual intervention and hopefully lead to higher customer engagement – higher rates of redemption, more positive reviews and increased re-purchasing. Mondo’s data scientist scoped out three key machine learning models – product redemption, positive reviews and post-purchase.
Our hypothesis was that over time, the ML models would allocate sample products that more closely aligned with the member’s previous engagement behaviour. Running in parallel, Mondo dug into the campaign platform and started co-designing new features and analytics that would provide the team with insight into whether the ML models were getting the results we were all expecting.
Machine learning is no magic bullet. We honestly didn’t know how accurate the models would be. Neither did our client. But Mondo’s non-prescriptive approach and rapid release cycles meant the cost of experimental failure was low. Build, test, iterate, learn and pivot. This is the essence of Mondo’s DNA. They gave us the go-ahead to give it a red hot crack.
So did it work? Well yes and no. Right now, sophisticated machine learning models are in place and are continually being trained to allocate product samples to customers with increased accuracy and way less manual intervention from the client. For member engagement, the results are still coming in and are being fine tuned. Hey I did say it was an experiment, right? But for us this is where the magic happened. Did we prove our original hypothesis?
Hand on heart, it’s still too early to say. What we do know for certain is that tackling a complex and potentially transformative problem doesn’t have to cost a gazillion bucks and take up months of internal resource effort. That done right, rapid prototyping, co-designed with a highly engaged partner, means the cost of experimental failure is low. And while our original hypothesis remains to be proven, the ‘journey’ ended up delivering a bunch of business value:
Non-prescriptive and experimental – no design/business requirements documents need apply. Together we agreed to embark on a radical and experimental approach without defining quantifiable outcomes. This freed Mondo up to focus on what the team needed from us to deliver real value.
Throughout the gig we heard just how stoked the team members were that a feature they requested yesterday was in their hands 24 hours later. We had a highly-engaged business partner with agency to make the right calls on the next priority via daily feedback. This was key to the success of this project, no doubt about it.
The Mondo DNA: Build fast, iterate, fail quickly and cheaply, learn, pivot. No big bang, ‘solution reveal’ party here. In short – small, fast-moving teams underpinned by clear and direct communication across the project team and business stakeholders.
Drop us a note and we’ll be in touch as soon as we can be.