Applied AI & AI in business /

The long tail is the future of AI

February 2/3 min read

Promises of the massive value AI will bring have loomed large since the start of the latest AI boom. Touted as "the new electricity" by Andrew Ng, it's understandable how everyone from Fortune 500 companies to startups have raced towards becoming "AI first". Some succeeded, and some have been less successful. With everyone moving into AI, what can be learned from the ones who blazed the trail?

First to arrive on the AI scene were the Googles, Facebooks and Amazons of the world. They created state-of-the-art models that are used today and put AI into production at massive scale. In order to do so, Big Tech recruited the best-of-the-best and had 80% of all PhDs in AI working in their labs. (Source)

The early adopters followed in their footsteps. Large online-retailers, banking, telecom - they all started to hire top talent and build out their AI departments. They saw initial success, promising pilots that were ultimately put into production, generating some sweet ROI for these visionary companies. The next step was then to scale those initiatives, and that’s where companies often get stuck.

Two problems quickly emerged for companies looking to scale.

  1. Recruiting and retaining talent in the AI field proved to be extremely hard. As one of the most coveted positions around, machine learning engineers and data scientists can pick their landing spots and often want to work on the latest technology. That causes a retention problem draining companies of momentum to move onto the next use case.

  2. AI in production requires a lot of manpower. The first one or five use cases that the AI departments get into production require monitoring, retraining and development. Those tasks are not possible to just hand over to the IT department. Progress is made in the relatively new field of MLOps (machine learning operations), but this still requires allocating ML engineers to maintain models in production.

So where does this leave the early adopters? How can they battle this?

02/ The mature AI company

It's easier and more affordable to train business people in AI than it is to train AI people in business.

During the last two years we've seen several examples of the mentioned scenario play out. Stakeholders feel the time to value is slower than anticipated while data science teams need to focus on essential ground-work, the kind of work that isn’t very impactful on its own, but necessary to be able to use AI across organizations.

Recently, we've seen a trend at a few large enterprises that we refer to as "mature AI companies". The AI departments or AI labs of these companies have realized the difficulties they are facing and have accepted the fact that they won't be hiring or retaining 80% of the AI PhDs of Google and Facebook. So what do they do?

First, people have realized that not all AI is created equal. Depending on the company, there might be five or ten AI use cases that individually can move the numbers on the bottom line. These are usually tricky problems that require state-of-the-art models and constant attention of two or more team members. These are the high-impact work that a data science team should focus on.

But then there is the second class of AI use cases, the long tail. These are smaller, incremental tasks that AI can do which in itself won't affect the bottom line, but when added up will have a massive impact. These cases usually replace a manual repetitive task and don't necessarily need state-of-the-art models or the constant attention of data scientists. They are the self-watering plants of an AI enabled company.

So, if data scientists focus on the high impact work and there is a treasure trove of "easy" AI cases that are ripe for the picking, who will do those?

That is the second thing the mature AI companies have figured out. It's easier and more affordable to train business people in AI than it is to train AI people in business. If you put your data scientists or machine learning engineers on a business problem, they will turn out a good solution. But it will only be one solution for one business problem. If you instead put your business people on solving their problems with AI, you will turn out any number of solutions depending on how many problems they identify.

So, the problems the business should work with is the long tail of use cases, many of which are in their own domain. With the added help of a friendly AI platform, the steep learning curve into the land of AI is severely diminished and the business people and AI people can create massive value for the organization while focusing on what they know best. The hidden benefits from that will both come from making it easier to retain AI talent and in upskilling the business in AI.

We see a massive increase in possible applications when business people understand what you can and cannot use AI for. It's hard to put a number on that realization, but we believe it is crucial to create an AI enabled company.

03/ What's next?

This insight will spread from the mature AI companies to enterprises that follow. Some will accept and adapt while others will stick to old ways of doing things. Many times it is the smaller team that are happy to embrace this new reality and lift some weight from their shoulders. For companies that have yet to start their AI journey this will be a way for them to leapfrog into AI enablement without investing in huge AI labs and employing data scientists to build up such a function.

Granted, this is not an easy endeavor. The path of recruiting top-level experts and letting them do their thing might feel more fool-proof. But the trend that we are witnessing first hand is that companies that have already gone the traditional AI route are now headed down this trail.

Since they have already tested and learned on the way, it only makes sense to follow in their footsteps.

The long tail of AI use cases

    • Björn Treje

      Björn Treje

      Head of Technical Enablement

      Björn has a Master of Science in Electrical Engineering. He strives to put engineering into the business and business into the engineer. Secretly he hopes all projects involves helmets or reflex vests at some point.

    02/ More on Business & AI