How to put AI to use in your business?

From Science Fiction, to a real game changer

Artificial intelligence (AI) is one of the greatest opportunities of our time. Fueled by vast amounts of data and important advances in deep and machine learning, it has the potential to add almost 16 trillion dollars to the global economy by 2030.

Though, as in the first days of electricity discovery, AI adoption has been slower than anticipated and has taken on an air of mysticism with promises of grandeur and out of the reach of "mere mortals". Companies of all sizes and across all industries are struggling to adopt AI while they fully understand its potential (cf. here or here).

A game changer serving important business purposes…

What is AI ? There is many definitions of AI, and it is perhaps more accurate to define many AIs instead of one unique form of AI. Anyway, this discussion is beyond the scope of this short article. Let us say that AI is an umbrella term for a set of techniques that allow machines to learn from data and to act on what they have learned rather than following rote instructions created by a programmer. This relative autonomy of machines is referred to as "intelligence" - or "intuition" for more cautious minds.

In all cases, what is commonly called AI is behind several spectacular advances in many areas, e.g. speech processing, image/video recognition, autonomous vehicles, translation engines, robotics, paint creators and more.

For businesses, AI is a way of radically improving three areas : predictions, automation and optimization.

Improving predictions allows to take advantage of what’s going to happen in your business in the few coming days, months or years, e.g. customers behavior, environment shift, peak periods, special customer' events, evolution of demand, evolutions in the financial markets, etc.

On the other hand, there’s tremendous value in automating critical yet time consuming, and repetitive business processes , which are often done by humans, e.g. document classification, contract generation, face/object recognition, reporting, counting & checking objects, treating customer queries, etc.

Finally, AI is also about optimization, whether that means optimizing customer experience, routing and logistics, marketing spend, or asset portfolio allocation.

AI is then coupled with strong digital/IT capabilities to meet users new requirements :

• Immediacy - easy and immediate access
• Customization - tailor made products and services
• Pro-activity - responsive and active service/product providers
• Meaning - engaging and with a credible signification of actions
• Collaboration - embarking customers in value/brand building, creation process, design, etc.

However one should overcome and tackle serious challenges before winning the grail

While the potential of AI is generally well known in the business world, there's several challenges which are facing businesses concerning the adoption of AI. Those challenges can be summarized in six important issues.

Managing data

This is by far the most important challenge, since data is the foundation and the fuel for AI. Generally, organizations are facing three types of issues concerning data management :

• Data collection problem - not enough data
• Data engineering problem - data scattered all over the place
• Data quality problem - garbage data - 60% of business leaders are challenged by managing data quality

Lack of understanding

AI can't do - should not be used for - anything. In fact, the first step for any organization is to ask the right questions, specify the business problems they are trying to solve, and identify whether AI is the right tool to achieve business goals - which is not necessarily the case. This is a crucial step to avoid wasting time and valuable resources to "kill a fly" or to "teach a fish to run".

Lack of relevant skills

The extent of the global AI skills shortage is laid bare in the Global AI Talent Report 2019. While this study is focusing on cutting edge data scientist (PhDs), many businesses are also struggling to find intermediate data scientists with the relevant skills. On the other hand, AI requires even the most experienced software engineers to relearn a lot of what they take for granted, as the regular software development workflows of Continuous Integration (CI) and Continuous Deployment (CD) don’t easily apply.

The issue of trust/transparency

It is critical to ensure that AI recommendations or decisions are fully explainable - enabling adoption by business experts - and traceable - making possible to audit the lineage of the models and the associated training data, along with the inputs and outputs for each AI recommendation. 

The issue of deployment

The deployment of AI is the process for making models available in production environments, where they can provide predictions to other software systems. It is only once models are deployed to production that they start adding value, making deployment a crucial step. There is many specific challenges in the deployment of machine learning models : managing data dependencies; maintaining config clarity with constant model iterations; features entanglement; detecting model errors; management of tasks between data scientists, data engineers, software dev and IT; plus the challenges of traditional code…

Culture and business model

Lacking a shared vision and/or roadmap, strong governance and/or fast innovation cycles - fast deployment of the solutions. One should have the ability to rethink the process altogether and do things that were previously impossible.

Is there a shortcut or a special recipe ? If not, how to set one foot firmly planted to go beyond ?

There is no one and simple way to make AI adoption pop up in your organization like magic. This process is complex and have a component which should be tailor made taking into account organization's culture and specifics. Fully leveraging AI is hard - especially in some industries - and this is why most companies are struggling. Yet, one can avoid repeating other's errors by adopting some practices and/or adapting them to your organization.

There is no AI without achieving success first with how you collect, process and organize data

Data collection :

• Business data - internal processes, supply chain partners, etc.
• Users' touching points - loyalty programs, technological facilities, etc.
• Use open/alternative data - public data sets, satellite imagery data, etc.

Data organization :

• Make data clean, available, simple and accessible

Data analysis :

• Create a business-ready analytics foundation
• Set up a strong IDE and collaborative tools - such as git

There is no AI without a strong information architecture on which data is organized and structured across a company. The IT infrastructure should be (re)designed for AI

• No data silos
• Support all data types 
• Setting a unified approach to govern data and AI life cycles
• Be open to hybrid multi-cloud platforms and/or IaC (Infrastructure As Code) solutions - it's perhaps the future of data architecture

There is no AI without HI - talents :

• Create an engaging/fun environment
• Propose 'special' packages/incentives and claim proportional results
• Build or reinforce an internal "data community"
• Define a clear and transparent distribution of tasks and stick to it !
• Give meaning, a why, a narrative - people embrace what is meaningful, not what is objectively true or expedient