Why Implementing AI is Hard - A Guide for Non-Technical Execs

Guide to Implementing AI for Non-Technical Execs

DATA

November 6, 2024

6 min read

Teo Gonzalez

Summarize with AI:

Why Implementing AI is Hard - A Guide for Non-Technical Execs

It’s probably the buzzword that you haven’t been able to shake since 2022: Artificial Intelligence (AI). You see it in countless videos, articles, and blogposts. Yes, I see the irony considering this is a blogpost about AI. You’re also probably being told that you need to find ways to implement AI at your company, whether from your CEO or your Board. There’s a problem with this directive: implementing Generative AI into an organization is pretty freaking hard.

As with any major change, there’s bureaucracy you’ll need to navigate. However, with implementing AI, there are also some unique roadblocks you’ll encounter. We’ll discuss some of the major ones, like a noisy market landscape, disconnects between technology and business, and inaccessible data. We’ll also include next steps so you can make sure you have the best chances at success for deploying AI at scale.

Information overload. Where do I even start?

Probably one of the most daunting graphics you’ll ever look at is a landscape map by Matt Turck.

It’s awesome for someone in the data space, but total chaos. However anxiety-inducing the market map is, it actually makes sense. Within data infrastructure, there are so many things you can do/work with to have the optimal experience.

Now see that section in Matt’s map that says AI? It’s exploded with more options since it was published. AI is a new technology sector (remember when you added ecommerce or a mobile app to your business?). AI needs all new infrastructure, just like those initiatives. Check out Sonya Huang’s market map in Generative AI Act II which is overflowing and now a year old. If you are just shopping for a solution to tell your stakeholders that you’re making an AI investment, good luck.

Having a successful AI project requires understanding your business and the technology itself. Ask yourself the following questions:

What are the tedious manual tasks our people are doing over and over?
Where are we currently experiencing bottlenecks in our operations?
What processes are currently experiencing highly inconsistent outcomes?
What tasks rely on repetitive decision-making that could benefit from standardization or automation?

If your answers to these questions have parts of your org or function that overlap, congrats! You’ve likely found your heading for where to implement AI. Now, how? I’m not even going to broach the buy vs build conversation yet. Your next step needs to be connecting with your internal team, which introduces challenge #2.

Business & technology often speak different languages.

You’ve identified a high-impact area for AI, but now you need buy-in from a critical stakeholder group—your technical team. It’s tempting to assume that getting them on board is as simple as saying, “Let’s use AI to improve efficiency.” However, moving forward requires something more nuanced: defining a clear vision, understanding the technical expectations, and bridging gaps in communication to ensure that everyone shares the same understanding of goals and feasibility.

One major challenge lies in creating a shared vision that aligns business and technical objectives. For your technical team, success in AI projects isn’t always immediately apparent and often requires setting specific, measurable outcomes. Without this alignment, there’s a risk of pursuing broad, ambiguous goals that ultimately lead to frustration and misunderstandings.

The next hurdle is the language barrier between business and tech teams. When translating business goals into technical requirements, differing assumptions and terminology can create friction. For instance, consider the term “real-time” data: business leaders may assume it means daily updates for reporting, while the technical team sees it as a demand for continuous streaming infrastructure, updating on the order of seconds. The latter is a much more complex requirement. These disconnects can turn straightforward projects into costly, complicated efforts if not addressed early.

Finally, bridging the gap between what’s desired and what’s realistically achievable is essential but difficult. Determining whether your AI project needs a fully trained model or a more manageable Retrieval-Augmented Generation (RAG) solution often depends on your specific needs and resources. Many organizations find the best fit somewhere between these approaches. I recommend showing the value of AI through an 80/20 approach. In all cases, without a unified understanding, it’s easy for expectations to drift from what’s technically feasible.

That brings us to #3, the final boss you need to overcome to achieve it all - your data silos.

Your data isn’t where you need it to be.

Data is the lifeblood of AI, yet for many organizations, it’s not readily accessible or structured in a usable way. Rapid growth often leads to an accumulation of tools and platforms, each of which may house valuable data but also contribute to a disjointed data landscape. This data sprawl can make it difficult to find, consolidate, and prepare data in a way that meets the standards required for AI.

Scaling AI initiatives brings its own complexity as you work to integrate data across multiple databases and APIs. While companies assess whether to bring AI into their internal or customer-facing workflows, there are many questions to answer. What is an acceptable UX for the audience? What is the minimum accuracy required? There’s a lot here that depends on robust infrastructure. Without it, even the best algorithms can struggle to deliver meaningful insights.

A major step before launching any AI project should be to address these data challenges. Establishing a data governance strategy and investing in internal infrastructure might seem less glamorous, but it’s a foundational piece for scaling AI. Properly organized data will enhance the value of AI by ensuring that it can operate on a solid base, unimpeded by issues of accessibility and quality. As we always say, “garbage in, garbage out.”

How you can get to where you want to be in AI.

Achieving AI success, including the use of tools like an AI detector, requires a mix of strategic planning, team alignment, and a commitment to foundational steps. Begin by setting a detailed project plan that clearly defines your AI project’s scope and measurable goals. Visual aids, such as process diagrams, can help stakeholders see the complete picture and understand the specific outcomes you’re targeting.

Bring this vision to your technical team, not only to secure their buy-in but to gain their insights on what’s feasible within your current infrastructure. Advocating for your technical team’s perspective will reveal where you might need to make adjustments, bridging gaps between business and technology. Diving under the hood of your technical systems might feel intimidating, but understanding the strengths and limitations of your infrastructure can pay dividends as you work toward scalable AI solutions.

Setting a strong foundation through clear scope definition and cross-team collaboration will position your organization for success. AI is a journey, not a one-off project; adaptability and a commitment to continuous improvement are key to thriving in the evolving AI landscape.

Looking to speak with our AI team about what you're building with AI? Just fill out this form.

About the Author

Teo Gonzalez

Teo leads GTM for AI at Airbyte.

Try Airbyte Agents

Be among the first to explore our new platform and get access to our latest features.

Try it free

Why Implementing AI is Hard - A Guide for Non-Technical Execs

Try Airbyte Agents

Related posts