More than 1 trillion dollars.
That’s how much tech companies are expected to spend to introduce and mature their AI capabilities over the next few years. This investment has sparked debate about whether the excessive hype surrounding AI has led to it being overvalued.
If expectations for AI have been inflated, some organizations may discover that their investments in it will not be sustainable in the long term. This, coupled with how long it will take — in some cases, years — for some AI technologies to generate a return on investment, is driving analysts’ concerns that an AI bubble has begun.
Although fears about an inflated AI market hold some merit, the current narrative that “AI solves everything” has become a temporary pressure release. It allows tech companies to buy time as they work toward delivering tangible solutions. This claim also forces companies to overpromise, however, leading to potential missteps when AI doesn’t meet the market’s lofty expectations. This misalignment often results in confusion around what AI, especially generative AI-powered language models, can and cannot do.
In this article, we’ll explore three key strategies to help organizations maximize their AI investments: equipping their teams with large language models (LLMs) or small language models (SLMs), properly structuring and labeling data and developing an AI center of excellence.
3 Strategies to Maximize the ROI on Your AI Investment
- Equip everyone in your organization with an LLM.
- Structure data for ease of AI consumption.
- Establish an AI Center of Excellence.
Equip Everyone in Your Organization With an LLM
When introducing AI into your organization, the first step should be to equip every employee with an LLM. Most organizations will likely benefit by having an enterprise license for LLMs like ChatGPT, Gemini, or Claude. Even in the hands of someone new to using them, these AI models can perform various tasks, including content creation, research assistance, programming and coding concepts and behavioral data analysis. Making an LLM available to everyone and encouraging them to use it will make your organization more efficient.
Likely the largest benefit of putting an LLM chat interface in front of all your employees is that working with an LLM (and getting an LLM to work for you) is nuanced. Understanding the “uncanny valley” of talking to an LLM has a learning curve. It’s similar to, but not the same as, talking to a person. These nuances are hard to articulate but are certainly real. The best way to scale that hill is to get everyone climbing it together at the same time.
With the rise of SLMs, however, which require less data to automate a more limited set of day-to-day tasks, many are asking: Which language model will benefit my company the most? The key considerations here are who will use it, what they’ll do with it, its capabilities and how easily it can be integrated into existing systems and processes.
LLMs typically align best with the business goals of companies that will use them for customer support, content generation, and data analysis. They have billions of parameters, are trained on extensive data sets, and their size allows them to capture complex patterns in language using nuanced understanding and contextual reasoning to perform tasks.
SLMs, on the other hand, are trained on more focused data sets and are effective in simpler applications and specific domains. Whereas LLMs are trained to respond to natural human language, this may not be the case for SLMs, and, as such, they require highly specific prompts to create accurate responses. For this reason, SLMs are often the better choice for organizations that are more mature in their AI journeys and have the technical resources and data scientists to support them.
Structure Data for Ease of AI Consumption
An AI project’s effectiveness hinges directly on the quality of the data it can access for analysis and information output. Today, we’re seeing big tech companies using large data sets to train their models because they have the financial resources to do so, but the ultimate determinant of success lies in the quality of the data, not in the volume. The maxim ”quality over quantity” rings especially true here. Ensuring the use of high-quality data enhances the accuracy and reliability of AI-driven insights and creates trust among users and stakeholders, shaping the trajectory of the project’s impact and adoption.
Inadequate data collection processes contribute heavily to low-quality data. Companies need to capture and process data in a way that removes bias and formats it to be easily consumed by AI. Unfortunately, most companies still rely on instrumented data collection methods. This approach means picking and choosing certain data points to collect, which leads to incomplete data. It’s expensive because it requires an engineer’s time. Furthermore, instrumented data collection can overlook non-interactive behaviors, rely on technical accuracy, and require ongoing maintenance and configuration to ensure effective real-time analysis.
Companies that rely on providing exceptional customer experiences, especially online, can and should use behavioral data to raise the quality of data supporting AI systems. This data source includes the user’s interactions with digital environments and reveals detailed preferences, patterns and sentiments behind their clicks. It provides insight into metrics such as session duration, active time on site, number of error messages, or dead clicks to deliver a clear picture of user engagement, preferences and frustration points.
As mentioned above, however, many companies still rely on instrumented data collection methods for gathering this kind of data, which is fundamentally flawed. By using advanced autocapture/autostructure methods, companies can collect, translate, and discover critical customer behaviors — the obvious, the invisible, and everything between the clicks — to obtain the most precise and comprehensive behavioral data possible. Autocapture/autostructure automatically logs all digital interactions that users or customers have with a website or app across every visit and organizes the structure of data using ML algorithms to analyze and extract meaningful insights without manual data cleaning or labeling.
Advanced autocapture/autostructure interprets a user’s online actions at a hyper-granular level and takes all the related context — every scroll, every click target, interaction properties about business context, etc. — to structure this data and turn it into tabular, labeled, congruent inputs. Using all this contextual information, AI can then turn interaction into intention.
Establish an AI Center of Excellence
Establishing an AI Center of Excellence (CoE), which is a specialized team or framework that consolidates AI knowledge, tools and governance, is essential to fully maximize the potential of AI across your organization. This dedicated team should serve as the backbone of your AI initiatives, guiding the implementation and utilization of LLMs and SLMs within various departments. The CoE can oversee the selection of the most appropriate AI tools for the organization and focus on training and supporting employees in integrating them into their workflows.
To establish an AI Center of Excellence, an organization should define its role and objectives. The CoE should focus on an organization’s business strategy, technology, AI development processes, operational integration, and governance. Some of the key roles to include in a CoE team are AI strategists, business analysts, data scientists, AI engineers, data engineers, an ethics officer and a compliance officer.
A critical function of the CoE is identifying specific roles and functions within the organization poised to benefit most from LLM/SLM assistance. For instance, sales teams may require training on best practices for AI-generated communications, while developers might need assistance in automating code generation and QA testing. The CoE should work closely with different teams to understand their unique workflows and demonstrate how AI can address their specific challenges. This approach enhances user adoption and empowers employees to embrace AI as a valuable resource to improve job performance and overall satisfaction.
AI has emerged as a transformative force, but simply adopting AI technologies is not enough. To truly maximize their investment in AI, companies must determine how the technology aligns with business objectives, how resources will be allocated to use it, and what the key performance indicators are to measure its success.
Start with a team’s core KPIs and work backward to identify how AI can make a measurable impact. For instance, if your support organization adopts an SLM-powered chatbot trained on your company’s knowledge base or equips support agents with an LLM assistant, you might track metrics like Customer Effort Score or First Contact Resolution rate. Additionally, monitoring weekly tickets closed per agent could provide valuable insights. The key is to ensure that AI implementation is directly tied to improving the metrics and KPIs that matter most to your teams, reinforcing its role as a tool to enhance — not replace — human performance.
Although the hype around AI has raised concerns about sustainability and overvaluation, companies can still derive significant value from strategic investments. By equipping teams with the right AI tools, such as large or small language models, ensuring high-quality and well-structured data, and establishing an AI Center of Excellence to guide implementation, organizations can optimize their AI initiatives. These steps will enhance operational efficiency and position companies to navigate the evolving AI landscape effectively, turning potential challenges into opportunities for innovation and growth.
