Doing AI, Using AI and Mistakes Impacting
Your ROI
We speak to countless business leaders who are excited — but also bewildered — at the promise of disruptive technologies like AI. Recently, we’ve seen major light bulbs go off during conversations about putting AI adoption into two main camps: “doing AI” and “using AI.”
Each method caters to specific requirements and resources. Both offer unique advantages and use cases. In what’s ahead, you’ll conquer two overlapping challenges: 1. Knowing the differences between “doing AI” and “using AI” and how your organization can use each approach to your advantage.2. Avoiding the common mistakes in AI that prevent you from optimizing your results — and how to successfully avert them.By the time you’ve finished reading, you’ll be ideally positioned to use and do AI. But, in ways that avoid common pitfalls and catapult your business ahead of the competition.
The “doing AI” approach involves a more comprehensive development of an AI system. This method focuses on creating an in-house AI infrastructure — one that is fine-tuned and meticulously crafted to address needs across the organization.
The key characteristics of the “doing AI” approach include:
Large, well-manicured data sets: Successful AI implementation requires access to extensive and reliable data sets. In the “doing AI” camp, organizations invest significant efforts in gathering, cleaning and curating data to train their AI models effectively.
Expert team: Building AI models from scratch demands a multifaceted team. One that can include highly skilled data scientists, data architects, Ph.D. researchers and other subject matter experts. Their collective expertise ensures the AI system is optimized for the given domain.
Substantial resources: Implementing AI from the ground up is a resource-intensive endeavor. Organizations need considerable financial investments to support the infrastructure, compute resources and staff required for the project.
Time intensive: Doing AI often involves extensive experimentation, tuning and optimization to achieve the desired level of performance. Converging on an ideal model is often measured in months, if not years.
Machine learning brings new hope for better patient outcomes. See how this pioneering health system is applying AI for better, data-driven care. Read the story.
The “using AI” approach focuses on incorporating pre-existing AI solutions or services into a workflow. You might consider key players such as Microsoft, Anthropic and OpenAI leveraging their capabilities as a feature to drive improvement within your organization. That said, you don’t necessarily need to lean on a provider that only does generative AI. The provider could be a security vendor that’s built an AI model into its solution. There are several products that use AI models internally to drive better outcomes in their offerings.For example, there is an entire category of AIOps tools. These improve operations by using AI technologies to predict failure, correlate events and even optimize your environment.
The mindset of a “using AI” approach includes the following:
AI as features: Integrating AI capabilities into existing products, solutions and workflows helps you add AI functionalities seamlessly.
Pretrained models: The “using AI” camp benefits from pretrained AI models that are already capable of performing specific tasks. These models can be leveraged for inference, generation or other relevant purposes.
Fine tuning for customization: Organizations can modify pretrained models with their own data to tailor the AI system to their unique needs. This allows them to achieve better accuracy and relevance within their specific domain.
10 AI use cases. Take a glimpse of AI in action — from length-of-stay monitoring systems in healthcare to digital twins of factory floors in manufacturing. View the infographic.
Consider an oil and gas company. Suppose the company is working with their extensive, geo-seismic data that they have collected through expensive processes. This company would likely prefer the “doing AI” approach to develop their own AI system, ensuring complete control over their data. That is the organization’s competitive advantage.
Within that same company, maybe their development arm has a goal to improve coding efficiency. In this case, they may want to opt for the “using AI” approach.
They might use a service like Microsoft Copilot that integrates AI to assist with code development. The organization can enhance their coding capabilities without investing in extensive AI model training and building.
There’s no way around it: Rationalizing adoption is a huge part of how successful your organization will be as it adapts to new technologies.
At a high level, these doing and using approaches to AI have been a highly effective way to rationalize AI adoption for Insight’s clients.
As you think about all the factors that can guide your journey — data sensitivity, expertise, available resources and your desired level of customization — remember that embracing the right approach can be a game changer for your organization.But a game changer can turn into game over if you aren’t careful. Which brings us to the caution you must take to get the most out of AI.
An annual survey conducted by Foundry and commissioned by Insight found that while 85% of companies already use AI to drive business insights, 57% have only just started their journey or are looking to mature their practices.
Whether you’re using AI or doing AI, knowing these common AI mistakes will make it much easier to avoid them and optimize your results.
Sometimes an AI journey can get off the ground before the purpose is fully defined. This can lead to insufficient results that are not aligned with the business need. Even if an AI project has already begun, it’s never too late to revisit or recalibrate the purpose to make sure you are asking the right question. In fact, the data science lifecycle encourages frequent hypothesis testing, reinforcement learning and adaptability. Part of this evaluation should also be to ensure that the data you have and want to use is adequate: Do you have the type, scale and quality of data needed for this AI project to be successful?
Your questions may need nuance. Aim to get the outcome that will either answer your question or provide a better understanding of how to refine your pursuit of the answer. A company may be looking for deeper knowledge into their customer’s behaviors — but it’s important to define which behaviors they’re trying to understand. If the company wants to dig into the buying behavior of customers to determine store hot spots, that will look different than if they wanted to analyze the behavior differences between new and returning customers.
Let’s say you have a good understanding of the questions that need answers. Even then, results from AI techniques must still be interpreted in the context of the model’s ability to provide an answer that is unequivocal. Essentially, in the world of algorithmic learning and intelligence, probabilities and correlations are rarely 100%.Evaluate the conditions around the project: Is there confidence in the model that was created? If the model is based on low-quality or outdated data — or the wrong types of data — then the result might not be answering the question at all.
This is a chance to evaluate if the model and results might be biased by incomplete data or presumptions. At the end of the day, AI can be a transformational tool for your business. But it’s not magic. We can instead think of AI as a powerful tool — one that should be wielded by domain experts who understand its strengths and can augment the “business intelligence” of the organization.
One of the key lessons of AI is that it’s a continuous process. Implementing AI is not a “set it and forget it” technology. It is most successful as part of an iterative process. One of the best ways to ensure you’re asking good questions and coming to correct conclusions is to maintain an active feedback loop for your AI project to re-engage and recalibrate as necessary.
Besides the fact that you may have been asking the wrong question or coming to an incorrect conclusion, you may just want to completely shift the purpose of your AI.
Recall the customer behavior example. After some time, you may observe a different or more pressing customer behavior you want to learn about and adjust accordingly.
It can be frustrating when AI consequences do not meet business expectations. It might be a strain on financial or infrastructure resources, or critical business insights can be missed, thus hurting ROI. It’s crucial that organizations aren’t just doing AI or using AI — they must embrace AI. A great example of this is a national convenience store chain that wanted to get business insights from its security cameras and opted to use computer vision technology for the solution.
By using computer vision, the client gained confidence in the mature model as it provided the needed visibility into inventory and loss prevention. The project’s success caused the chain to look toward future applications of the technology to answer other questions about the business. This example demonstrates the power of purposeful AI and the value that can be gained even from existing assets.
Use it or do it. We’ll help you harness the power of purposeful AI to create impact, intelligence and innovation to deliver real value for your organization.
Learn more
