Gen AI: The “Business End” of the Adoption Cycle

So, it’s now getting more into the “business end” of the Gen AI adoption cycle. Organizations have mostly moved from “toy mode” to “serious mode”. And I’m thrilled to see it. I always enjoy this phase in hype cycles, it suits product companies like us. Since January, I’ve seen customers evaluating Gen AI fall into three camps:

• Thoughtful & Forward facing: One large customer sent us a very good RFI (Request for Information). They’re approaching the market in a considered way. It’s a thoughtful RFI, emphasizing things like security, listing specific tasks they want to apply Gen AI to, asking for the financial health of the company etc. Another enterprise customer is working with us as part of our Gen AI early adopter program. They’re also looking at cost models, whether to use open-source permissive LLMs or the hyper scalers (AWS, Azure etc) offerings. The big guys have variable costs & can be expensive. No CFO likes that combo ?? (more on that topic next time)

• Don’t Touch it: Another client (who happens to be a major US gov agency) explicitly flagged to us that they don’t want any type of Gen AI turned on in our products. I expect this stance will soften over time. But for product vendors, being able to toggle Gen AI on/off is critical over certainly the next 18 months and could go longer. Of course, that’s exactly what you can do in our products. I talked in the last newsletter about the UK Government’s experience with their Gen AI experiment. Hallucinations and lack of accuracy make it currently a non-runner for them.

• Still viewing Gen AI as a “silver bullet”: Effectively, this cohort still thinks that Gen AI will solve/displace everything. They don’t understand which tasks Gen AI is good for, which it’s bad for, etc. They don’t fully understand the challenges of hallucinations, and security. They will need to get a better understanding of transformer technology (the underpinning of Gen AI) and why hybrid solutions (incorporating Gen AI and standard Computational models) make best sense. We have some educational work to do here. But this cohort will likely waste a lot of money & time going down blind alleys. They will buy-in products that demo well, but disappoint when it comes to real world scenarios.

I’ll use this edition of the Human Factor to outline the biggest risks that come with the actual adoption of Gen AI in enterprise (and, of course, how to avoid them).

So, let’s get right into it…

First off, the Build vs. Buy question

Should you build the solution yourself, or buy a ready-to-use solution from a vendor.

You might think that as a vendor I’d be banging the drum on buying products, but the reality is, there’s a time and a place for both approaches. Both approaches have their advantages and disadvantages.?

Whatever you choose, here are some things to consider:

Building Gen-AI Solutions: 2 Common Pitfalls (and how to avoid them)

From what I’ve seen, enterprise organizations building Gen AI solutions in-house make two big mistakes. As a result, they waste resources and time.

So, here’s what to avoid:

1) Too much focus on the tech, and too little on the problem

Gen AI is exciting. But that doesn’t mean you should focus exclusively on the tech. What matters is the underlying problem that you’re trying to solve.

When you wield a generative AI hammer, everything starts to look like a nail. In reality, not every business problem is an appropriate use case for generative AI.

For instance, don’t use Gen AI for tasks that require 100% accuracy and repeatability. Use it for creative tasks such as 1st draft creation, summarization, simplification.

Even though generative AI seems smart and intelligent – it’s not. The technology guesses the right answer based on its training data. And sometimes, these guesses are wrong (aka hallucinations).

So, begin by focusing on the problem, then look for the right technology choice. You’ll find that most problem spaces require a hybrid approach, comprising standard computational models with Gen AI thrown in the mix.

2) Scope Creep when you go down the “Build” approach

Many enterprise tech teams start off and create a nice LLM prototype that’s compelling enough to get people excited. But oftentimes, they forget that ultimately when you build internally, you need to consider:

• Scalability & performance requirements,?
• Uptime requirements,?
• Enterprise quality levels,
• Release & maintenance schedules,
• Support requirements,
• Training requirements, the list goes on….
• ..

And it’s not just a technology question. Business-wise, what budget pool will the spend come from? Who internally pays for support? what is the ROI? Etc. etc.

Hidden costs start to mount up. And before you know it, you become a glorified product company consuming internal tech and support resources – not a company invested in your business. This is a fairly standard pattern I’ve seen over the years, and especially in hype cycles.

Buying Gen-AI Solutions: How to evaluate vendors using a simple 5-step process

Given all the problems mentioned above, buying generative AI solutions might seem like the perfect solution. While buying definitely has its advantages, there are also risks.

Because it’s not just about deciding to buy a solution from a vendor, it’s also about choosing the right vendor (partner) for the longer term.

Here are 5 things to consider when evaluating vendors:?

1) Security

If you’re in a regulated industry (like defense, space, financial services or healthcare), you need to ensure that your data is fully locked down and isolated. This applies especially to Gen AI.

This will narrow your options, as many vendors rely on public cloud-based LLMs, which exposes proprietary data to the cloud.

So, the first step when choosing a vendor is to evaluate whether they can 100% guarantee that your data is 100% secure and locked down behind your firewall.

2) Vendor Newness

Before choosing a vendor, always ensure that they’re around for the long haul. Check whether they have industry references and have been operating for more than 3 years. If not, that’s a clear warning signal.

Now, I don’t mean to discourage young startups. But especially for enterprise organizations, the long-term perspective is oftentimes more important than short term gains.

3) Overpromising and under-delivering

Many vendors (and especially the newer ones) claim their solutions are capable of everything. They’re wielding that “Gen AI hammer” with abandon. But this should be a clear warning signal. Dig in, try it for real. Trust but verify.

When buying software from a vendor, trust is crucial. Make sure the vendor is prepared to allow you to test run the capabilities of their own software (or generative AI in general).

4) Infrastructure Support

Not as exciting a topic as sleek product demos, but more important for enterprise organizations. When evaluating a vendor, always dig deep into the underlying supported infrastructure.

Many of the newer vendors are effectively just a thin veneer around the underlying large language model. However, for enterprise use, you need to check whether they fully support your use cases and can meet the infrastructure demands of your IT colleagues.

Do they support virtual private clouds? SSO? Team and user management? Or plumbing infrastructure generally needed in the enterprise? What deployment models do they provide? Can you swap out one LLM (Large Language Model) for another? What is their InfoSec (Information Security) posture? Do they ship data beyond your firewall?

Point being – Don’t buy anything that you’re not 100% sure is a fit.

5) Cash Runway

No private company is willing to divulge their balance sheet, I know. But you need to make an assessment to evaluate whether a vendor is in a good financial position.

If they’ve raised a lot of VC money but are very new, they’re likely just cashing in on the AI hype. Many of these companies will crash and burn in a while. Ask for their monthly burn and balance sheet. They may not want to offer it up, but no harm asking ;-)?

Evaluate whether they’re an appropriate partner for you for the long run.

So, if you apply these 5 checks before buying AI software, you'll be in really good shape.

But regardless of build or buy, there’s one thing that always applies…

Every Gen-AI project should begin with the problem that you’re trying to solve. Once you have complete clarity on what you’re trying to automate, then you can think about what tech stack is the right choice for the job-to-be-done.

I hope you enjoyed this edition of The Human Factor. Follow me on LinkedIn, where I share more insights like this.

Best, Fergal

Founder & CEO

VisibleThread

PS - I publish a new edition every two weeks. Subscribe so you don’t miss any!