Why you’re deploying AI incorrectly

AI (artificial intelligence) and ML (machine learning) can be invaluable assets to business success, enabling automation of labour-intensive tasks to drive faster and smarter business decisions

However, successfully deploying AI requires proper attention and care. In order for AI to help and not hurt your business, it is crucial to follow best practices. Here are five key errors to avoid when leveraging AI to meet company goals. 

1) Incorrect use cases  

By now, many businesses realise the benefits of AI and feel obliged to keep up with competitors in this area. McKinsey found that by the end of 2021, 56% of all respondents reported AI adoption in at least one business function. 

Despite the pressure to deploy AI as soon as possible, arbitrary and rushed attempts to implement it are ill-advised. It’s important to apply AI to the right use cases only. Rather than leading with, “Can I apply AI to this situation?”, the better question is, “Am I applying the right AI to the right situation?”  

If AI is running incongruously with business goals, time and company resources will be wasted. Unfocused AI implementation isn’t ultimately worth the investment.  

2) Misguided hiring 

The UK tech landscape has been struggling with a hiring problem for the past two years, with leaders coming up against a skills shortage. In the third quarter of 2021, there were more than 64,000 vacancies for UK tech jobs according to the latest BCS State of the Nation report 

Given this challenge, the need for diligence in the AI hiring process is extremely important in order to build a team with a variety and balance of skills. Rather than hiring only generalist data scientists, pay attention to each candidate’s specialised experience and match this to your business needs. For example, deep expertise in modeling is critical for thorough research and solution development, while data engineering skills are essential to execute an AI solution. 

3) Data hygiene  

Data is the fuel that drives the AI engine. One of the biggest mistakes companies can make is not taking care of their data. This oversight often starts with the misconception that data is solely the responsibility of the IT department. In fact, before data is even captured, business subject matter experts and data scientists should be involved to ensure the right data is being identified and maintained appropriately. All teams must be educated to have a shared sense of responsibility for curating, vetting, and maintaining data.  

Later in the game, data management procedures are also a key component. Processes for data management and governance need to evolve to handle the increased volume, velocity, and variety of data while ensuring compliance with government and corporate regulations.  

4) Lack of appropriate intervention  

It cannot be assumed that once deployed, AI can be left to its own devices. AI requires consistent intervention to serve as an effective solution over time.  

When it comes to AI-based pricing systems, effectiveness will be limited if the system is not designed to accommodate market shifts. One way to measure the value of AI is through sales team performance, specifically analysing how well the AI system is being adopted. Common pricing related KPIs include profit margin and revenue. If the AI-driven recommendations are not improving KPIs, it may be time for intervention. 

Intervention, which must be scalable through automation, should include two components: (1) reviewing inputs to the AI system and (2) ensuring the system output is as expected. Each of these practices should be set up as planned checkpoints based on clearly defined metrics. Waiting for AI to malfunction before intervening will limit the positive impact of AI and potentially harm your business. 

5) Potential bias  

AI and its derived outputs can be biased when exposed to a limited or nonrepresentative dataset. Rather than this being the fault of the AI itself, the problem often lies with the input data and system design decisions.  

A good practice is to avoid data that can be inadvertently biased against race, gender, class, etc. For example, modeling based directly on consumers’ geography and income may produce biased output.  

Explainable AI can also be a good solution here. By identifying the key factors that are driving the AI model’s predictions or recommendations, explainable AI can make it easier to diagnose what is driving the bias. Once explainable AI methods show how AI is arriving at biased outputs, intervention must be swift, repeatable, and scalable to avoid further negative consequences. 

Getting AI right 

When leveraged correctly, AI can be an indispensable asset, with benefits ranging from high return on investment to delighted customers. 

Rushing AI deployment without following guidelines can produce results that run counter to business success.

About the Author

Justin Silver is Senior Manager of Data Science at PROS. PROS (NYSE: PRO) provides AI-based solutions that power commerce in the digital economy. Using artificial intelligence, PROS accelerates customers’ ability to embrace digital selling and eCommerce channels. With predictive and prescriptive guidance, companies are enabled to dynamically price, configure and sell their products and services across all channels with speed, precision, and consistency. PROS customers, who are leaders in their markets, benefit from decades of data science expertise infused into our industry solutions. To learn more, visit pros.com

Featured image: ©Green Winds