Businesses are drowning in AI instead of riding the wave!

Written by Arpit Agrawal, CEO at Elastiq

I’ve personally spoken to over 100 Executives this past year from businesses across the US. 2024 was the year of AI experimentation – a lot of POCs, Pilots, Failed attempts and some really big wins! Here’s what I’ve learned from those winning with AI.

The AI wave

Operating a business is like swimming through a rough sea. The winning strategy isn’t just staying afloat – it’s waiting for a wave and then riding it to victory. 

We’ve seen various high-impact transformation waves over the last 30 years:

• Outsourcing
• Internet
• Mobile
• Social
• Cloud
• And now – the biggest one yet – AI

History shows us that huge businesses are built when you ride a wave properly. Each wave has given us many Billion-dollar businesses. Some 10Bn businesses. A few 100Bn businesses. And even a Trillion-dollar business in some cases.

We’re at the very beginning of a massive AI wave that’s going to transform every business out there.

100% of the “Digital Transformation” conversations today are about AI. Here are some key features of the AI wave –

AI will take away jobs

AI helps in optimizing workflows. And we’re helping companies identify workflows that have tasks that can be automated using AI.

But it doesn’t have to be 100% perfect at those tasks! It has to be just good enough to speed things up. Think a lot of knowledge work – market research, content search, even writing code!

It means now you can achieve a lot more in a much shorter time period. If your business previously hired a team of 10 to do a job, the same job can now be done using 3. And at that, the output would be more consistent, free of human error etc.

For example, we partnered with a US-based media & entertainment company to streamline their RFP response process using Gen AI. This led to reducing time spent on it from days to minutes. 

In another instance, an online art marketplace, leveraged AI to automate artist biography creation, increasing speed by 100x, reducing costs by 10x, and increasing their average impressions per page by 50%

But here’s the million-dollar question: Is that a good thing?

As per Gartner’s research, almost 70% of employees don’t feel as engaged as they should be. AI would be a forcing function for us to find our calling, our purpose and to find more meaningful work. We can then delegate everything else to AI!

AI will provide way more jobs!

Technical advancement has always taken jobs! But generated way more!

Back in the day, people rode horse carriages for ground transportation. When the internal combustion engine (ICE) was invented, a lot of people (and horses) lost their jobs. This included people who took care of the horses, who sold and repaired carriages, etc. But this paved the wave for the whole automobile industry. Which is a $3.6T industry!

The mobile phone revolution took jobs from various professions. Whole businesses were shut down – Telephone Directories, Film cameras, and so much more. But at the same time, it created a whole new industry, with companies that are worth billions of dollars!

Similarly, AI will take away jobs, a lot of jobs, but then it’ll also provide for a lot more! That’s how we evolve.

Let’s look at coding as an example:

Back in the day people coded assembly language.

Then came low level languages like C and C++. These languages provided a layer of abstraction and allowed people to program faster. This also opened up software development to a much larger group of people. As a result, more people started coding.

As technology continued to evolve, languages like Python, Java, Javascript, etc. emerged. These were higher level languages. They provided a higher level of abstraction but then also opened up the world of programming to a much larger group of people.

And now AI allows a much higher level of abstraction by allowing us to code partly in English. What we’re seeing in this space is only a V1. This is going to get better over the next few years. 

English as a language provides an even higher level of abstraction as compared to Python or Java. But this will open up software development to even a larger group of people. It will help people be more productive.

Here’s the crucial point though – strong architectural best practices are needed irrespective of whether you’re coding in assembly, C, Python or English. And there’s no substitute for that. 

A human has to and will have to provide that for the foreseeable future because of how nuanced the architecture for each application, each piece of software can be.

Hire an AI employee!

People think of AI as software, I suggest you think of AI as an employee that you hire. It’s like having a high-maintenance virtual junior intern.

An LLM that has never seen your enterprise data before is like a fresh intern that gets technology, has a good understanding of the real world but knows nothing about your enterprise (yet).

If you want to use an LLM to produce SQL queries to run on your data warehouse, it’ll only be as good as your data catalog or your schema metadata. 

Generally there’s a lot of subject matter expertise that engineers who have been with the organization for many many years hold. In the real world, if you ask one of your experienced developers to write a complex business logic as SQL, they’ll only probably take a few minutes but the new intern would struggle, get stuff wrong, will miss data points that are not well documented, etc. Such is an LLM too!

Having said that, if you have a history of all of the queries that have been run on your data warehouse over time and fine-tune the LLM on that historical knowledge, you can actually have the “AI Employee” gain experience of your specific enterprise and perform better. It is like your junior intern gained a bunch of experience.

AI may not be for you – Don’t be fooled

AI opens a lot of possibilities. You’re able to solve problems that were not possible before. But you’ve to be careful where you’re using AI.

I talk to over a 100 CXOs every year, many executives come to me and ask to “add AI” to their business. Because the board is asking or because it is cool or because our competitors are using AI. All the wrong reasons! 

The real question should be: do you have a problem that needs to be solved and is that problem something that can not be solved without AI?

On the other hand, a lot of technology leaders understand the potential of AI and are trying to find where they can apply it. In many cases it is like they’re moving around with a hammer, trying to figure out where to use it without thinking whether you really have a photo to hang or is this hammer the right solution for this kind of a photo?

AI is great, but you have to understand where to use it and not be caught up in the Shiny Object Syndrome.

To do this:

First, identify what are the use-cases in your business that need to be solved? What are the bottlenecks in your business?

Establish a business case – Is this worth solving? What’s the cost of not doing this? What’s the potential upside if this is solved?

Identify the solution – what’s the best solution to the problem? Don’t default to AI for the solution, identify what’s the right solution for the use-case. 

For instance, if you want to reply with a standardized email to every “refund request” for your ecommerce business or to every “job application” for your hiring function, all you need is an email template. Using an LLM is an overkill that’ll make the process slow and way more expensive without adding any incremental value.

Even when you decide that AI is the way to go, a “Large” language model may not always be the best solution. For example – if you want to do sentiment analysis, do you need your model to know the lyrics of the latest Taylor Swift song? Probably not. In many cases you can do well with much smaller models designed for a specific purpose.

Going by my previous analogy of hiring AI employees, this is like saying you don’t need to hire the most expensive “Mr. know-it-all” if all you want this new employee to do is to bring you coffee every morning. You can hire different people for different skills.

LLMs by default are dumb

LLMs by default are naive and dumb. But they can be super smart if you give them the right tools.

As I’ve said before, hiring an AI is not that much different from hiring a human. You need to give the human the right tools to perform.

Humans by default are not fast enough, many animals (like Cheetah, Horses etc.) are much faster than humans, but humans have the ability to use tools. Give a human a car (or a plane!) and a human can be the fastest animal in the world!

Similarly, humans are not super strong, many animals (elephants, bears etc.) are much stronger. But give humans the right tools, (say a tower crane or a bulldozer) and a human can be much stronger than any animal.

Similarly LLMs need the right data and the right tool to excel.

And this is not new – any system for that matter needs the right data. Garbage-in – Garbage-out. This truth holds whether it is a deterministic system like a simple Python function or an LLM.

Image credit Eduardo Ordax

LLMs are over-confident

They hallucinate – They can be extremely wrong and extremely confident at the same time. Which can be a dangerous thing. You’ve to ask yourself, If the AI solution is going to be non-determistic, i.e., if it hallucinates, does that work for your use-case?

And this also is a big issue for the big picture. The internet is full of AI-generated content already and there’s no way for people to distinguish that content from the raw source of truth. In many cases, readers trust this convincing sounding non-sense.

Bet on AI but know how to win!

When you hire a new employee, you don’t just give it the most important task and trust it’ll always excel – you give it smaller samples, see how it performs, probably have someone oversee their work, provide feedback and then eventually have them take on more responsibility.

Similarly when using AI, start with a human in the loop. Build confidence over time.

How do you test your employees? Do you expect your employees to be 100% correct, all the time?

Know that AI can make mistakes, similar to employees. Put AI in production similar to how you’d put an employee in production. With the right tools, with the right testing, with the right oversight.

AI holds the promise to be the force that’ll transform all businesses. At the same time, businesses need to be strategic about implementation, set up the right architecture, manage expectations and set the right company culture for AI to thrive. The underlying data quality is still going to be critical to win with AI!

Key Takeaways:

2024 saw some amazing AI moments. 2025 will top that! Here are my key takeaways from 2024 that’ll fuel our strategy for 2025 –

• The AI wave is here: Rapid innovation in AI will continue in 2025, along with a reduction in cost and an increase in adoption. Over the next decade, AI will be the biggest force changing how businesses work!
• It is getting real: AI will no longer be a shiny new object, instead, AI will work in your organization as virtual agents/employees that’ll work alongside human employees. Businesses will have to create a culture where the “AI agents” can thrive! 
• It is a paradigm shift: AI will help raise human potential at large by providing extreme operational efficiency that we’ve never seen before, forcing us to find more meaningful work. Some tasks will get automated, but AI will open up new opportunities!

At Elastiq, we’re helping numerous businesses navigate the unchartered waters of this AI wave and co-create solutions that matter! I’d love to speak with you and discuss your AI strategy for 2025. Let’s talk!