It’s easy to get caught up in the excitement and promise of artificial intelligence, especially its flashier newer variations such as generative AI and agentic AI. There also are a lot of fear tactics around AI, pressuring organizations to jump on the AI bandwagon immediately or jeopardize their competitiveness and possibly their futures.
Much like the technologies that came before, the Internet, Excel, email, or smartphones, AI and ML represents a groundbreaking opportunity to transform business. Most companies and their processes remain kludgy, resource-intensive, and manual, so there is a lot of room to make workflows and jobs more efficient as well as develop new products and services.
But rushing into AI projects will not maximize ROI or results. Leaping without looking could lead to initiatives that fall short or fail, setting companies further behind while damaging workforce engagement at a time when many employees are already on edge over how AI will affect their jobs and careers. And the risks of jumping into subpar customer-facing or critical workflow AI-driven apps also could hurt internal teams and customer experience.
Moving quickly is important, but so is a disciplined, thoughtful, and strategic approach to AI. What’s needed is a framework for making, measuring, and scaling AI investments that includes rapidly culling the losers. Here are five guidelines to navigate this tricky challenge:
1. Establish AI within operational and strategic goals.
As with any technology, no matter how pioneering, AI is a tool, not a strategy. Companies should start with their three-year strategic plan for growth and innovation and work from there to identify potential opportunities that AI can help support or accelerate corporate goals.
At a minimum, AI should first act as an aggregator and interpreter of information across disparate corporate data sources. In this role, it would translate data from all sources into actionable insights and enable streamlined business operations, such as increasing automation, making data-driven decisions, optimizing resource allocation and even influencing pricing strategies. This use will also give you invaluable insights of where your business needs AI the most.
2. Identify and prioritize operational use cases.
Develop a list of possible use cases, with each possibility meeting at least one of these three criteria for the use of AI:
- Enhances the clarity and quality of the work, brings visibility
- Lifts work off people and systems
- Reduces friction and increases speed and efficiency
It’s important to not take on too much at once, so determine the use cases that could offer the greatest impact and focus on those to start.
3. Customize existing LLMs vs. building or buying.
Few companies have the expertise, capital, or time to build large language models (LLMs), key building blocks of generative and agentic AI, from scratch. But open-source and off-the-shelf LLMs have been trained on broad data sets that aren’t customized to meet a company’s strategic or operational needs.
The best way forward is to leverage open-source LLMs trained on data most analogous to your industry, then locally host the model to prevent leakage of proprietary data and build in a context layer with your industry and company data. Creating AI solutions with a custom context layer in a secure environment has substantial benefits, including:
- Intellectual property protection
- Increased accuracy for company use cases
- Explainability of results
- Audit trail for compliance and legal needs
- Mitigated compliance risks
4. Fast fail to curate and validate AI outputs.
In software development, fast failing entails quickly delivering increments or prototypes that can be used to assess project feasibility and benefits before too much time and energy is invested. It’s part of a continuous improvement approach, focusing on experimentation instead of perfection and prioritizing learning from mistakes versus avoiding them. This approach is particularly effective with pioneering AI.
For AI projects, it means setting short evaluation cycles and phases, which allow for course corrections or jettisoning what isn’t working and thus saving time and resources. They also require specialized scrutiny: Even though using customized versus public data reduces the chances of ‘hallucinations’ that produce inaccurate, irrelevant or fabricated information as facts, they remain a possibility. Building corporate AI tools with a “human in the loop” model to continue reviewing and refining the context layer is essential. These projects will need experts and human judgment at every step– reviewing and refining results, adjusting weights, and ensuring AI-generated results are useful and accurate.
5. Refine and iterate.
Build on results from the first AI use cases to determine and implement the best next AI phase.
In some ways, AI is a problem of abundance. There are so many paths to take, how will any organization know that what is best today will still be viable tomorrow?
It will take courage, knowledge, discipline and a lot of motivation to move organizations into an AI-centric future. Success hinges on treating AI as an operational initiative, with guidance and sponsorship from business leaders, aligned with technology teams.
Finding the right individuals to focus on AI for your company will be a challenge, so there will be no fast moves in aligning your leaders to choosing new additions to the team that embody AI experience, both operational and technical. You may also look to educating from within, finding those leaders and team members that are best equipped to learn and expand their own skills with AI. All businesses will need to consider how AI skills will fit in the future and start adjusting now.
We’re not alone, we are all facing this exciting time together so navigating to an AI-centric world will be fun if we stay calm and consistent.
Stay tuned for more insights and practical advice about AI in customer communications and payments.
