By Kelvene Requiroso
In a highly connected digital environment, artificial intelligence (AI) influences the daily decisions of consumers. Its ubiquity makes it an intriguing target for investment. AI promises enhanced customer experience and increased profitability among its many benefits. For the desire to improve operations, products, and services, companies rush the adoption of AI — partly due to the pressures of the competition. But only 13% of data science projects reach production, a small fraction of the number of companies venturing into the new frontier of business and technology. It might also be because 80% of AI projects through 2020 “remain alchemy, run by wizards.” Only a smaller number of people have the expertise, talent, and skills.
There’s no magic wand to make AI work in an instant. Its implementation is a complex process that requires planning, infrastructure, resources, and a lot of patience. Missing any one of these components renders an AI project a failure.
Reasons why AI projects fail
Here’s a familiar story. A company has everything ready to integrate AI solutions into its IT infrastructure—money, hardware, software, and personnel. AI is quickly integrated into the system and has undergone all tests and dry runs necessary. It’s time to scale up, and then the new system suddenly becomes unusable.
The AI implementation failure stems from a confluence of factors: understanding how AI works, data, production environment, and people running the project. It is, after all, human work.
Undefined business problem
Gartner points to a lack of understanding of the use of AI as a primary reason for failure. It’s alright to be driven by competitive pressures, but it is critical to have a clear understanding of the need for AI and how it is used for the business. Healthcare, retail, or media enterprise, for instance, deal with large amounts of data where AI is in the best position to provide solutions, as it can predict events, recognize patterns, identify trends, and process natural language.
The best way to address lack of understanding is to identify the tasks, challenges, and opportunities for AI to play a role in. And to define the problem, break it down into smaller units and solve them one at a time, but with the awareness that not every problem can be solved by AI. Once the problem is defined, you can also quantify the outcome, whether AI can improve customer experience, enhance the company’s operational agility, increase revenue, help save on costs, and promote better employee engagement. You can also leverage use cases, including AI’s impact, risks, and advantages.
Further reading: Retooling Supply Chains After COVID-19
Problem with data
Most problems that AI adoption encounters have to do with the veracity and volume of data. Bad data sets, even if you have a good database and IT system, cause your AI model to malfunction. Latency and throughput issues in the flow and processing of data can also crash the production environment.
As wrong data results in a bigger problem, ensuring data quality becomes imperative. You need to identify the right data. And that is contingent on data architecture, which must be flexible and expandable. There has to be a set data strategy in place. To make your data usable in the implementation of AI requires a delicate process of data cleansing and data preparation.
Production environment
An AI model starts with a small unit or a subset of data in a larger ecosystem of different parts that make up the whole. AI requires storage or hardware relative to its size. But as it expands, a larger, well-designed production environment becomes necessary. Not considering the exponential impact of AI is a sure formula for failure.
Designing a dependable, flexible, and scalable production environment takes time, and it involves many moving pieces. You need to assess AI’s viability, whether it can be integrated into the existing IT infrastructure, or if the algorithm can be deployed in the operating system.
The human players
More than software or hardware, adopting AI into your IT system needs people. Humans play a critical role in AI integration, requiring continuous training and support for the implementing team.
The key to benefiting the most from machine learning is AI Operations (AIOps), which includes the players and tools to make AI operational. An AIOps team builds the solution, integrates AI, performs tests, releases and deploys the model, and manages the system. It is about organizing the right team to run the project, composed of domain experts, technical personnel like developers and engineers, and data scientists. And it calls for a collaborative environment.
Your AIOps team can be in-house, which could incur higher overhead costs, but you will have more control. You can also contract out to external service providers to lower overhead costs, but you risk giving up some control. Or a hybrid, a combination of in-house and outsourced. But in the longer term, you need to invest in the training of your personnel. And that requires both time and money.
Security risks and unplanned situations
Cybersecurity is a risk in implementing and scaling AI projects. As you expand your AI solution, it exposes vulnerabilities in the existing security system. And there are unplanned situations that arise as you scale up AI implementation.
Part of the role of the AIOps team is to identify weaknesses and reinforce the identified vulnerabilities. The team must also devise contingencies — both technical and operational — to accommodate unforeseen issues and anticipate problems before they occur.
A successful AI adoption in a nutshell
Having the problem identified, the role the AI model will have in the business, is a prerequisite in its adoption. There’s a rigorous assessment of existing data, whether they are the right data, and supporting the set objective. Executing the adoption of AI into an existing IT infrastructure takes time. And it also requires a well-designed and well-planned production environment that can accommodate unforeseen situations in its expansion.
Key to AI integration is the AIOps team, the personnel implementing and maintaining the project. The team is also responsible for keeping data quality by providing an audit trail and other measures to ensure that the AI project succeeds.
