Crossing the chasm for Industry use-case specific AI Applications

AI is a pretty funny term. It’s supposed to be the silver bullet that cures us of all problems business and personal.

But then, at least in our lifetime, Strong AI is not going to be mainstream. The next 30 years are going to be about weak AI or rather AI built for specific use cases.

Think of this era in this manner: Between the 60s to the early 90s we had a separate device for capturing photos, doing phone calls, connecting with the internet, notetaking, listening to music, etc. But now since atleast 2007 we use a single device to do most of these things.

We are not in the “Smartphone era” of AI. Rather we are in the disparate devices creation era. The smartphone era will come but most probably not in our lifetime.

Industry use-case specific AI Applications – The next 10 years

So now coming to feasibility, AI has hemmed and hawed over the last 50 years. It has suffered many dark ages. But this time we finally have the technology to match the vision of AI.

That being said, let us understand that the major problem with AI is that in many cases it expects a complete change in end-user behavior in order to use the solution. It has long learning curves and disruptions in the current process

Or in most cases, it’s a technology searching for a use-case. This is a pet caveat I have esp. considering the number of AI and computer vision platform businesses that provide you building blocks to create your own solutions. This is something similar to RPA (Robotic Process Automation)

But the problem here is that most models have not matured enough to reach the plug-and-play status. Also, in the current scenario, also knowing most engineers and their behavior, they find it easier to build their own applications rather than using the so-called platforms.

The only way these platforms will gain mass usage is when they bypass the engineers and go directly to the end-users there way RPA has done.

That being said, it’s my opinion that during the next 10 years adoption of AI will be done for specific Industry use-cases. There are many reasons behind this. Some include:

• The problem statment of most major business use cases has been clearly defined. This means you dont need to pitch technology, you pitch solutions
• All business process timelines have been clearly documented and thus any productivity improvement claims can be easily tested
• Budget allocation and stakehlder uyinwill be easy since you will pitching tech but direct ROI

Now, the above is mostly from the business side of things. But knowing what to solve is 90% of the solution. With that in mind and knowing the adoption curve. Let’s ask the next pertinent question?

How to achieve mass adoption of Industry use-case specific AI applications?

Now the question arises, how do you cross the chasm and achieve mass adoption of your Industry use-case specific AI application. In this part of my article, I’ll refer a lot to the book Crossing the chasm by Geoffery Moore.

Also, I’ll mostly focus on the journey to the first $10M which is currently the phase I am in. Now, having done your research and being reasonably sure of your problem statement you will dive right into development. It is here you will find your first chasm

Chasm #1: Data for PoC development

Unless, you are an established firm lack of data is going to be your biggest problem. Well, there are multiple ways to cross the chasm here. You can either buy he data or create your own. But what if you have identified some niche use case for which you can neither buy data nor create one on your own. In this scenario, the best way is to identify an adjacent use case which can help you enter your target customer and once in you can pitch your killer use case and build a PoC with them.

This method of penetrating the market can be done with both the Innovators and early adopters as per Geoffery Moore as they are more liable to take risks. Whilst the Innovators will work with you directly on your killer use case, you need to build trust with the early adopters and this can be done along with

Also, remember whatever be the case, always have a machine learning architecture with a feedback loop from the real time, continuous data that flows into your application as that’s the only way accuracy can improve.

Chasm #2: End-users not using the application

The application which solves the problem should be easy to use and integrate easily with the current business process. My rule of thumb esp. if you are a startup is that your product should be integrated and start providing value in 3 months and should have ROI calculated and case study ready in the next three months.

In this way not only will you get social proof which can get you further customers, it will also give you insights into wether end-users are actually using your product.

That’s a major problem with AI based applications; Selling it is only 10% of the game, making sure the end-users can use it is 90% of the game.

The best way to cross the chasm is by having low to nil learning curve. If your user manual is more than 2 pages long you are doing something wrong.

Also, remember if the innovators and early adopters are not using your product no way in hell is anyone else going to use it.

Chasm #3: Distribution Channel

The distribution channel is important esp. while selling to the early adopters and early majority. Let’s ask a direct question here “Why should your potential customers trust you?”

Considering the fact you are peddling a new kind of technology which in most cases is unproven and since you yourself are unproven, it’s better to focus on going through the traditional distribution channel which your end customers are actually comfortable with.

There is only one new variable you can enter into an human interaction at any point if you want to do a successful behaviour change. In this case it is your product.

Get into the traditional distribution channels as soon as possible and you will stand a good chance of crossing the chasm.

Chasm #4: What’s the action?

AI is well known to go throw vast quantities of data and give insights. But you can;t eat lunch with insights. Creating AI solutions which drive you towards assisted decision making or automated decision making which can then do actions that saves human labour and time has the chance to become a successful AI solution.

Let’s get it straight. AI is not going to automate decision making. The logic which you will build on top of the AI models to do a certain action is going to automate decision making.

That doesn’t mean AI solutions which give insights will not cross the chasm? – There are many data points which are currently very difficult to gather and make sense of. If your AI is providing insights then make sure it is incredibly difficult to get that insight and most importantly AI should not fall into ” I already know that” trap.

All things held equal, it’s easier to sell AI products/solutions which automate decision making and does actions than an AI solution which provides insights.

Conclusion

At the end of the day any application you build depends on what problem it is going to solve. As my boss is fond of saying, “If you are first in the market, the customers will forgive a crappy product.”

Right now though, AI is being adopted. In most sub-fields of AI esp. in places like computer vision it is still the wild west out there.

If the last two decades of the 20th century was defined by the promise of one PC in every home the third and fourth decades of the 21st century is going to be about incorporating 100 AI models for each human working for them.