Project: Automate Consulting, The Humanlytics Approach (Part 1)
Part 1: Why is automating digital consulting so difficult?
Consulting is way too expensive.
As someone who graduated from business school and led my family business through multiple costly consulting engagements, it pains me to see how much money companies spend on consultants for expertise such as human resources management, strategy, and digital marketing.
For this reason, slightly under a year ago, I started Humanlytics with one specific goal in mind — automating consulting with AI technologies so small- and medium-sized businesses (SMBs) could employ the business and analytical expertise of a consultant at a much lower cost.
After interviewing over 100 SMBs (you can read about it here), we chose digital marketing as our focus not only because it is a high-growth field that many SMBs benefit from, but also because this is the exact area within which most of the business owners we interviewed struggle.
It deeply troubles me that, despite an abundance of data from free tools like Google Analytics, SMBs still fail to use this information on a daily or even weekly basis to improve their business. Especially because, for those who did, their online revenue increased drastically — some of them even doubled revenue in just a year’s time. We are determined to help all businesses see those results!
After almost half a year of trying, and two failed alpha prototypes that did not even make it pass the internal “sanity test”, we finally landed on an AI framework that we are confident can automate digital marketing consulting.
The purpose of this post is to share that framework with you, and encourage feedback and discussion. Because it is really important for us at Humanlytics to build our product and service while gathering input and suggestions from industrial experts, business owners, and entrepreneurs who are already established in the field.
We want to hear from people like you who are also passionate about the challenge we are tackling, and want to work toward a solution together with us.
To accomplish this, we are going to present a two-part series about our view of automating consulting, and our current approach in accomplishing it.
In this first part, I will set the stage, talking about why automating consulting is so much more difficult than other automation tasks.
Then, in the second part of this series, I will talk about how we at Humanlytics are going to use a combination of psychology, AI, business, and design to tackle all the problems explained in today’s article.
Challenges in Automating Consulting
Market Tech Today published an article earlier this month arguing that we will have self-driving cars before we have an automated marketing engine, and I agree with the sentiment.
Automating digital marketing consulting — or any consulting in general — is a task that is exponentially more complex than self-driving cars, and its complexity can be summarized primarily in three points.
First of all, the business world is complex, with multiple criterias of success and paths to get there.
The goal of a self-driving car is simply to move a passenger from point A to point B in the fastest way possible — without violating any traffic laws or crashing into pedestrians or other cars, of course.
However, for a business, the concept of point A and point B (and the path between them) are inherently unclear and multifaceted.
First of all, a business’s success might be judged in multiple criterias, such as customer loyalty, revenue generated, long-term growth, or simply the owner’s happiness.
For example, in the case of the famous outdoor clothing company, Patagonia, the owner forewent the path to maximize revenue in favor of providing its customers with the highest quality product that they would enjoy. This sentiment would throw off most AI systems (and human consultants) unless it was explicitly communicated by the owner before analyses were conducted.
Furthermore, even if the definition of success is clear for a business (let’s just say it’s “increase revenue”), there might be multiple, uncertain paths towards that particular objective.
Driving from point A to point B is what AI technologists call “deterministic game” because all of the roads are mapped. It is absolutely certain to the AI algorithm which options are available for it to optimize.
On the other hand, the business challenge is un-deterministic or stochastic — filled with uncertainties and probabilities.
What this means is that a consultant, AI or not, can only tell you that their recommendations have a high likelihood of working for your business. They cannot provide you with a “silver bullet” that can resolve your business problems once and for all.
For this reason, the best consultant — or consulting AI — is going to be the one that helps you establish rigorously tracked systems of your actions; then tell you whether the recommendations they made to your business actually produced the desired results, and iterate upon those results to further improve your business.
In order for consulting automation to occur, the system must develop some sort of mechanism to communicate with the business owners it serves, understand the concrete goals and objectives of that specific business, and then work WITH the business owners to chart the best course of action to achieve success in their mind — and none of this will be required for self-driving cars.
Secondly, there lacks a comprehensive repository of past consulting data to train the artificial intelligence algorithms.
The true power of AI lies in its ability to process and analyze an enormous amount of data in a relatively short period of time, and detect patterns in those data better than humans are able to.
Therefore, the foundation of AI is the ready availability of vast amounts of data to train the algorithms.
Unfortunately, this abundance of data is not readily available in the consulting industry.
Currently, there is no systematic standard for consultants all over the world; there is no centralized place to store their past client cases and enable training of an Artificial Intelligent system.
Even in large companies such as Bain and McKinsey, their past experiences are at best stored in the format of case studies — which is not a format that can help consultants understand, statistically, what solutions works better in which specific situations.
At this point, even if those companies decide to go back to their cases and make them more statistically sound, only part of the information may be captured since a lot of necessary context data was simply not collected during the creation of those cases.
Essentially, automating consulting is a chicken and an egg problem — you need to have data for recommendations to work, but only by providing recommendations can you get enough data.
This means that the AI consultant of the future needs to be serving people, while also collecting data to constantly improve itself. This concept of dynamic AI is significantly more difficult to design for than a regular AI system.
Even if we have established a systematic way of collecting past case data in consulting, variations of data structure among companies may be another challenge that prevents consulting from being automated.
This challenge originates from the fact that different companies have dramatically different data storage practices and data management needs. This results in a simple piece of information — such as revenue — stored drastically differently from one company to another.
AI technologies, as they stand right now, can only be trained using data of a fixed set of features, and do not have the intrinsic ability to develop or recognize new features automatically without the manual addition of a human programmer.
The most feasible way for AI technology to tackle this challenge is to create a layer of machine learning algorithm on top of the clients’ database with the sole purpose of mapping fields in the database into features that the machine learning algorithm can understand. This is another layer of complexity that an automated consulting AI has to deal with.
Lastly, there also needs to be a mechanism in place that holds businesses accountable for executing actions recommended by consultants or the AI system.
A typical consulting engagement goes like this: consultant comes in, works on the problem presented by the client for a couple of days, presents a solution and an implementation plan that they think is the best fit for the client at the current stage, and then leaves. The implementation is entirely up to the business.
In many cases, the plan does not get implemented in the end, and consultants are not entirely responsible for the execution and tracking of it — unless they are paid another sum to help clients’ monitor the long-term effectiveness of their recommendations.
This implementation gap is what frustrates me most about the consulting industry, and will inevitably be one of the biggest challenges for AI consultants: to show concrete results and return on investment.
Unfortunately, if we are talking about consulting for SMBs, this problem is even more severe.
As a small team, the employees and owners of SMBs usually have to wear multiple hats at the same time, ranging from personnel management to even delivery driving.
For this reason, they only have, at most, a couple of hours every week to focus on improving their digital presence, and will most likely forget about spending these few hours in the first place if their week gets too busy.
Therefore, the last challenge preventing consulting from being automated is that the consulting AI need not only serve as a super consulting intelligence that offers good recommendations based on clients situations, but also as a good personal trainer that holds its clients accountable for executing those recommendations. And then come up with new, personalized recommendations based on the results of those actions.
This means that in order to design for an AI that can truly automate consulting and provide concrete results to its users, we need to chart beyond the field of Artificial Intelligence and Machine Learning, and also incorporate techniques and findings from behavioral psychology and human-centered design. That is, the AI needs to be created with human behaviors in mind.
Automating consulting using Artificial Intelligence is significantly more complex than a task such as self-driving cars.
For a consulting AI to work, it not only requires an extremely expansive and flexible intelligence system, but also requires incorporations of fields outside of AI to solve the problems of human behavior.
Upon reading this article, you may think that automating consulting is not possible or is perhaps a project for a later generation. I do not believe so!
While automating consulting is definitely not an easy task, I believe we have the technology and theoretical support to make it happen (at least to some degree) now. This discussion is going to be the focus of the second part of this series.
In the next post, we are going to describe the “three-pillar” approach we use at Humanlytics to automate digital marketing analytics, and show you how this unique combination of human-centered design, AI, and behavior science will resolve or bypass most of the challenges presented in this article to make automated digital analytics a possibility.
Meanwhile, please comment below if any of the content here resonates with you, or if you want to beta test our AI marketing analytics tool for free. Until next time!