Part 2: Creating a human-centered algorithm that is both flexible and rigorous
As explained in the previous article of this series, when designing a self-driving car, the purpose of the artificial intelligence is clear - move the passenger from point A to point B most efficiently without violating any traffic laws.
However, when designing an artificial intelligence to automate insight generation in digital consulting, defining a clear objective is significantly more difficult.
This is because the objective of a business is very much subjective to the owner of the businesses and could range from maximizing revenue to serving as a lifestyle business for owners to have a steady stream of income.
Even if the objectives are clearly defined, the strategies to achieve these objectives may vary greatly across industries.
For example, for the same purpose of maximizing revenue, an ecommerce store may run online ads to bring people to the website, whereas a government contractor may invest in an expensive face-to-face sales team.
For this reason, when tackling the challenge of automating the generation of insights and recommendations for digital marketing, we at Humanlytics immediately realized that our solution needed to have two important attributes:
On the one hand, the system needs to operate under a strict business and measurement framework that will enable the machine to solve specific business problems in structured ways.
On the other hand, there must be some degree of flexibility built into the business framework that allows the algorithm to interact with the users, and create a solution that is tailored specifically to their needs.
In other words, the algorithms we design must be human-centered.
This article will go more in-depth on how we at Humanlytics are working on a human-centric algorithm that creates harmony between flexibility and structure to automate insight generation in digital marketing.
Principle 1: Restricting analyses to only the metrics that are concrete and actionable
Human minds are designed for thinking, connecting, and creating. While those attributes are extremely important and valuable for businesses, the human brain is horrible at processing a large volume of information mechanically.
Unfortunately, this is what marketers have to do in the new digital age.
Let’s take your digital presence for example. Even if you are not running any advertising via Adwords or Facebook, you will receive data on a daily basis regarding the performance of your website (via Google Analytics), your social media platforms (via Facebook insights), and organic search rankings (via Google Search Console).
In each of these datasets, you will receive hundreds if not thousands of data points on the most nitty gritty details of your performance, including how many people visited your about page, how many people watched your videos, and how many people actually made a purchase (if you are an ecommerce store).
While all these data are overwhelming business owners across all company sizes and industries, there currently lack sa rigorous analytics and measurement framework to tell the business owners what they should measure in their business and why these data are important.
Therefore, in order to avoid data points that are not important to your business objective, the algorithm we designed at Humanlytics employs a customer-centric framework that tracks the effectiveness of your efforts in:
Acquisition: Bringing customers into your business website via channels such as Facebook and Organic Search
Engagement: Engaging with your customers so that they remain interested in your content
Re-engagement: Re-engaging with your potential customers via email and social media to further their interest
Conversion: Pushing your customers through the conversion funnel to persuade them to add value for your business.
In each step, our algorithm will use 4-5 key metrics (around 20 total), and as many dimensions (e.g. channels, demographics, etc) as possible, to identify the “missing pieces” that are preventing businesses from achieving their objectives. Our algorithm will also identify what the business should do to resolve these challenges.
Compared to the algorithms powering self-driving cars, the techniques employed in our solutions are much more primitive in the sense of feature engineering (self-driving cars use thousands if not millions of features).
However, while the algorithms for self-driving cars are designed for accuracy, our algorithms are designed to interpret data for even the most ordinary business owner. In other words, our algorithms are designed for humans.
Principle 2: Interacting with users to define objectives and metrics at each step
The centralized, restricted framework explained above is designed to resolve the chaotic nature of business data, but it doesn’t resolve the problem of varying business objectives depending on the preferences of business owners and marketing managers.
In order to accommodate the unique need of each business owner as much as possible, we built a series of experiences that are specifically designed to interact with business owners to understand their business needs and challenges.
This user experience uses AI to make recommendations on what objectives the owner should choose depending on their preferences, but leaves the ultimate decision power to the owners themselves.
After interpreting and understanding the needs of the owners, the system converts these requirements into concrete metrics that can be measured. The AI then fires up the data processing algorithm to produce the best ways forward for the owners to achieve their objectives.
However, the interactive experience between machine and user does not stop there.
After making recommendations for next steps based on the data processing algorithm, the decision-making process falls back on the business owners.
The front-end “wizard” is going to come back and work with the owners again to consolidate the final action steps the owner will take, depending on the preferences and resource constraints of the company.
Finally, the tool will create concrete implementation schedules for the owners to follow so they know exactly what they should do next and monitor their progress.
Due to the interactive and iterative nature of our tool, we like to say that we're building a “personal trainer” or “personal assistant” for digital analytics - it doesn't achieve your goals inside a black box, but rather asks for your feedback along the way. It's got the best of both worlds between "Do It Yourself" and "Do It For Me"
In this article, we examined the processes we used at Humanlytics to create a more human-centric AI that works with business owners to both (1) monitor the decisions made by the owners, and (2) make recommendations based on the results of their actions.
However, I understand most of our discussion remains surface-level, and I do not want to stop our introduction to the Humanlytics framework here.
As we develop our product, we will continue this series with concrete examples of how we are accomplishing each of the features and functions described in this article. We will also give in-depth explanations of the design decisions we make.
At that time, we will also offer you opportunities to try out our tools and offer feedback on how we can make the tools better for your needs as a business owner. This way, we can build something together that is truly valuable to improve your business.
Meanwhile, please comment below if you have any comments or questions about what we are doing!
We are also looking for companies to beta test the product for us and help us grow our consulting knowledge base. If you’re interested, please contact us @ email@example.com or sign up here: bit.ly/HMLbetatest
If you are interested and passionate about what we are doing, please contact us or sign up for our weekly newsletter below. Thanks!