The making of onsite marketing
Wisepops’ tech roadmap to build the first onsite marketing platform
I spent 8 years at Amazon and Facebook growing data-driven ecommerce platforms. As I implemented and managed large-scale data products, I saw first-hand their staggering effectiveness at impacting billions of online shoppers.
At the same time, I realized how ill-equipped brands are without the well-optimized, production data science of centralized ecosystems. Convinced that customers expect more authentic and personalized connections, I embarked on a journey at Wisepops to empower the brands they love with automated marketing tools that were so far reserved to the biggest organizations.
In this post, I would like to take you through our tech roadmap to build the first marketing platform to directly engage customers onsite .
The potential for innovation and growth is immense. In 2021, online businesses increased their digital ad spend by 35% (up to $189B¹) to build their brand and acquire traffic. Meanwhile, their conversion rates dropped by 12% (down to 2.5%²). To revive their growth in an ever more competitive ecommerce environment, we believe brands should refocus their efforts on the 97.5% of traffic that leaves without converting. We estimate that less than 5% of digital marketing budgets today is allocated to this exciting, untapped opportunity.
Our plan to tackle this new market category is to add actionable levers to the typical conversion funnel. This will create more opportunities to influence visitor behavior when they are onsite and to convert them to loyal customers:
In a nutshell, our roadmap aims to:
- understand our clients’ onsite audience
- find the best moment & channel to reach them
- engage them in a personalized way
- drive conversions
- build loyalty
Improving each of these stages by 15% could yield 2x returns overall.
Here is our action plan:
(1) Onsite audience
➧ We plan to build a privacy-centric behavioral data store.
To understand our clients’ onsite audience, we need to segment their behavior and intent by profiling their activity as they browse the store. Automated and anonymous onsite signals will be the foundations of our data store design. This means we will rely on first-party data pertaining to the brand, and will not access personal information from unauthorized third parties.
On the one hand, this is a technical challenge as we’d have less data to predict the potential of incoming traffic. On the other hand, this is an opportunity to provide privacy-centric solutions since we’d use data compliant with the strictest regulations.
(2) Onsite reach
➧ We plan to build a real-time campaign decision engine.
To reach out to potential customers, we need to match their onsite activity with our clients’ most relevant campaigns. The key challenge lies in capturing the customer’s intent in real time (e.g. right before they abandon their cart), and answer their needs via the channel they will be most responsive to (notification, pop-up, embed etc.), while not interrupting their shopping experience.
With tens of thousands of concurrent sessions across our clients, this requires a system capable of handling up to 100K TPS (transactions per second) at 100ms P90 latency (the threshold for a responsive UI) anywhere in the world.
(3) Onsite engagement
➧ We plan to implement accurate and serendipitous recommender systems.
Personalization of communication has been shown to make engagements less intrusive and more relevant. The content served should thus be aesthetically delightful as well as spot-on to fit the customer’s needs.
To implement personalized recommendations, we will need to overcome several challenges: i/ the cold start problem, where we may lack previous knowledge about visitors as most are not logged in; ii/ the data sparsity of ecommerce websites, where users usually interact only with a small portion of the catalog; and iii/ the tight accuracy SLA and controls for serendipity to ensure our recommendations are not only relevant, but also novel and unexpected.
(4) Onsite conversion
➧ We plan to build self-optimizing, online-learning algorithms.
To generate better ROI for our clients, we need to optimize conversions with dynamic pricing and promotions. This entails selecting the kind and the amount of offer that will be most meaningful for a particular purchase (i.e. smart A/B testing).
The main challenge is to improve upon past learnings — not merely reproduce them — and adapt to ever-changing customer habits. It is a tremendous opportunity to leverage contextual bandits and other online learning algorithms that continuously explore and exploit campaign variations.
(5) Loyalty
➧ We plan to build integrations with top digital channels.
To build loyalty, we need to integrate closely with our clients’ existing channels. Customer trust is built over multiple cohesive purchase journeys through onsite and offsite campaigns.
During the first visit, we should incorporate elements reflecting where the traffic originated, while enabling future engagements with audience-building functionalities. For subsequent visits, we should recall prior touch points as part of the experience, while driving repeated sales in the future.
As a bootstrapped startup, we have less resources than VC-funded players. We also have less friction to innovate. This makes us more nimble to iterate quickly between ideation and implementation with short feedback loops. Being profitable and consistently delivering high-quality projects over the past 6+ years grants us plenty of room for experimentation and failure, as well as the foundations to succeed in the long run.
I am tremendously excited about the growth potential and the innovation opportunities of our business here at Wisepops. I hope this peek into the breadth and depth we aim to reach has sparked your interest as well. I look forward to engaging with you in the comments or through DM.
In future posts, I will share our progress and deep dive into specific challenges. Interested to know how we set up our data lake to ingest 2–10B monthly events as a stream for less than $0.1 per 1M events? Follow us for more insights and findings!
