AI for Managers Part 2: Fear and Loathing in AI

 
 

The focus of the second part of this series is to address some of the key fears and concerns with integrating AI into your organization. We have focused on four key areas for this post:

  1. Data - What is it? Where is it?

  2. The art of the possible (or impossible)

  3. Risk - How can we manage it?

  4. Scale - How can scale help manage risk?

The intent of this post is to help navigate some of the most common fears around AI (don’t worry it’s not coming for your job) and to help you make better decisions about what is right for you and your organization.

Data - What is it? Where is it?

Data - I’m sure you hear about it a lot. Someone has mumbled about it being the new oil. Other people are shouting about data leaks. Data isn’t oil. There’s lots of junk, and putting stuff in lakes without knowing what it is should be called pollution. And no one wants to swim (or drink from) a polluted lake.

Data is everywhere in your organization. It comes in many different shapes, sizes and types. For the purposes of this post we are going to focus on structured vs. unstructured data and ways to increase the value you are capturing around that data.

Art of possible

Right now AI learns extremely narrowly and only from the data it was shown. This means that you cannot take an AI model trained on pictures of cats and expect it to recognize dogs. For that matter if the model has only seen a cats face it will not be able to recognize a cat’s tail. To see what we mean, try playing with this.

What this means is that AI doesn’t generalize well. It learns task-specific patterns and makes predictions on the basis that these patterns hold true across the data set. Being cognizant that your model and predictions are only ever as good as the data that you show it is critical to managing your organization's expectations about the art of the possible.

Risk - How can we manage it?

As with all innovation, there is risk involved. Risk takes many forms. In discussing risk we are going to talk about three different types of risk: compounding, linear and finite.

Compound risk is best expressed in this scenario as the risk of doing nothing. The risk is compounding because by doing nothing, you are compounding the impact of the outcome. Failing to effectively capture, curate and consume your data not only are you missing out on business value today but you are also increasing the cost of catching up. Eventually, the cost of catching up will be so high that you will never be financially capable of doing so.

Linear risk is a risk that increases in a linear predictable fashion. In the context of AI this is the risk associated with poorly planned or executed data capture and curation strategies. While you may be capturing and curating some data your approach is incomplete or ill-conceived. The chasm between you and the competition stays consistent but you are perpetually playing catch up. Implemented poorly enough the risk starts to look more and more like compound risk.

Finally, the finite risk is the leap you take when investing in a large scale AI project. The cost of failure is finite (corresponding to the cost of all inputs). In AI adoption it is rare to find finite risk in isolation.

The best way to mitigate each of these types of risk is to design and implement a capture, curate and consume pipeline for your data. To minimize the risk short term we are huge proponents of starting off small and building on success(‘brick at a time’). This approach should contemplate the end goal of the organization and operationalize incremental change to mitigate the short term finite risk balanced with the long term compound risk.

The bottom line is that it isn’t appropriate for organizations to become ‘AI first’ tomorrow. This is a huge finite risk ignoring that the early stages of compound risk appear fairly linear. Instead, we advocate a measured approach that doesn’t require an organization to boil the ocean by changing overnight. Incremental change in the early stages of AI adoption mitigates the large finite risk while taking advantage of the minimal impact of compounding in the early stages of the transition.

How can scale manage risk?

One of the ways that we at IMRSV handle risk is through defining the scale of our problem. Our secret? Start small. You don’t have to boil the ocean your first time out the gate! With that in mind, we try to define the problem we’re trying to solve by targeting something small that can have a large impact.

In our process we make a plan with our client, come up with recommendations, design the solution and help integrate it into their system.We work with you to define your unique business problem and help you unlock the potential of your data to find a solution that meets your needs. All this while balancing the value of information with the cost of achieving your end goal.

One of the benefits of this approach is that we can help find a small scale project that answers a specific question, save people time, and won’t break the bank to implement. One great example of this is how we helped Innovapost leverage data gathered from call centre interactions. By using ML to screen calls for quality and accuracy, based on metrics they already knew, we were able to increase call agent performance and decrease operational costs in one fell swoop. To read more about how he helped Innovapost, visit our case study.

Summary

In summary, are there things to be worried about when trying to introduce AI in your company? Sure. But are there real ways to limit those risks? Absolutely!

You likely have all the data you need just sitting there as a wasted opportunity. What’s important now is to recognize what is possible to do with that data and to do so in a way that you’re not taking on larger risks than you can afford. Just be mindful of the scale of the project you’re trying to undertake. Is there a way to break your problem down into manageable parts? By keeping these things in mind, you can make room for innovation while limiting the risk.

Don’t believe that there is a small project with a large impact using data you already have? We can help!