Your Anti-Whiplash Guide to AI Implementation

“We tend to overestimate the impact of technology in the short term and underestimate the impact in the long term.” – Amara’s Law

The AI hype cycle is non-stop, and we see nine-figure investments on a weekly basis. What does this mean for you and your company? The good news is that the hype cycle is completely disconnected from the value of AI in a corporate environment. The bad news is that if you’re waiting for AI to go away, it won’t. We’re seeing intentional early- and over-investment in a new and promising technology.

There are three types of companies right now:

1. The Resistors: Companies that are digging in their heels and fighting AI. Think: fighting mobile devices in mid 2000’s. People were scared to check their bank accounts on a mobile device. Now, those devices are probably the most secure computer that you use in your daily life. Advice: Stop waiting for AI to go away. It’s not going away. Get “on the bus.”
2. On the Bus: Companies integrating ready-built AI solutions into daily operations through their software supply chain. Checking the “now with AI” box. This is the vast majority of companies. They are spending time understanding AI strengths and weaknesses, cleaning up their data, and identifying impactful use cases. Advice: Keep going and ignore the hype cycle.
3. Driving the Bus: Companies at the forefront of AI development, building anything from AI features to training new foundational models. A small number of companies can bring true business value by building and training their own models. Think: OpenAI, Microsoft, Facebook, etc. Advice: Don’t listen to me.

This article is an anti-whiplash guide to getting value out of AI implementation in your company for The Resistors and those companies who have already hopped aboard The AI Bus.

As a passenger on The AI Bus, you have two options available to you: revenue growth or efficiency gain. Every company has opportunities for internal efficiency gains, and some have potential for external revenue growth through AI innovation. Internal efficiency gain may lack the glamour of revenue growth, but remember, revenue is vanity, profit is sanity, (cash is king). Saving a dollar is usually as good (or better) than earning one.

Existing AI models from any mature provider are more than capable of carrying out your internal tasks to save you time. Why isn’t the world already run on AI? It’s the user interface.

The Grandma Test. The “world wide web” existed in one form or another for decades before Netscape Navigator was released in 1994. Netscape provided a user-friendly interface to servers hosting websites just as the content available on the web was becoming interesting to your average household. Perfect timing.

Netscape did for the web what Windows 95 did for personal computing, Facebook for social media, Amazon for online shopping, the App Store for mobile apps, and MakerBot for 3D printing. The interface made it accessible. Grandma could figure it out (even if she still calls you for help with the “woozywig whatchamacallit”).

When OpenAI released ChatGPT in November 2022, they made it possible (and easy!) for grandma to use the latest generative AI model for free. They had 1 million users in five days, and 180 million by mid-2024. The second huge interface win: Microsoft pushing out OpenAI’s capabilities through integrating it with their Office Suite and existing software stack user interfaces.

Steps to Make (or Save) Money with AI

“I skate to where the puck is going to be, not where it’s been.” – Wayne Gretsky

Just two years ago, many people would have considered widespread AI use in the enterprise 10+ years away. Now, many companies and employees use AI daily.

Do:

1. Evaluate your business processes. Identify areas where AI can streamline operations and improve efficiency. You do not have to build a model for it to be personal to your organization. Retrieval Augmented Generation (RAG) and customized initial prompts can get organization-specific responses from existing pre-trained models.
2. Review your data. Determine the quality and quantity of your data, including knowledge base articles, customer interactions, and operational data. If you’re using privileged data and different user roles, you’ll need a data governance solution across your data. No one wants employees querying HR data through an AI app.
3. Think through the interface (UI). Do you have the internal expertise to architect an enterprise-ready AI application? Will you use partner-led development, use an existing AI data platform, or build your own? Assess your team's ability to develop AI solutions. Consider existing workflows and how employees will interface with the application.
4. Build prototypes. Build the framework for internal ideation to identify use-cases, build prototypes, and fail fast or scale. Spread your investment across several pilot projects to see what gives the best ROI. Align AI initiatives with long term business goals. Measure success with clear metrics and drive continuous improvement.

Cautions:

1. Avoid Oversimplified Solutions. Don’t throw a chatbot on the internet, point it to your knowledge base articles, and call it a day. One of AI’s core “features” is hallucination. A Canadian tribunal court recently ruled that Air Canada was responsible for the (mis-)information that its chatbot put out: it told a customer traveling to a funeral to go ahead and book the fare and submit after travel for bereavement pricing (directly counter to Air Canada’s published policies).
2. Solve your own problems, not someone else’s. Focus on the problems specific to your organization and know where the technology will catch up. As an example: don’t re-build your chatbot to make it faster. That’s where the puck is right now! Yeah, it’s kind of slow. Remember when internet on phones was kind of slow? Instagram still prioritized user growth and now you can scroll as fast as you want on mobile 5G internet. Every week, models are getting faster and cheaper by huge amounts.
3. Separate investment from hype. Don’t confuse your investment in AI with the hype cycle. The Gartner hype cycle is over-hyped. We’re seeing a lot of massive “better early than late” investment. Mark Zuckerberg said on the August 2024 Meta earnings call: “…but at this point I'd rather risk building capacity before it is needed, rather than too late” after investing billions in AI model development. He’s not alone – NVIDIA, Apple, Microsoft, venture capital firms, and others are doing the same thing.

Over half of the CIOs that we surveyed in Q2 2024 are getting moderate to intense pressure from their Board and/or CEO to adopt AI solutions and executives overwhelmingly feel unprepared for the challenges (known and unknown) around AI adoption. We’re in the early innings – big players are making enormous early investments and models are rapidly improving. You’ll get value out of AI at your company through a methodical approach to AI use-cases and interfaces, disconnected from the investments from those driving The AI Bus.