How To Tell The Difference Between An AI Opportunity And An AI Money Pit

I’ve built a business model around natural language systems, industry first features with computer vision, and improved margins with dynamic pricing models. There are huge opportunities in AI and equally large money pits.

Knowing the difference between what Stanford is developing and what your competitors are developing takes leadership. 95% of what is out there is purely academic. The last 5% is gold but only applicable to certain use cases and can only be implemented by teams with relevant capabilities.

You can see there are 3 dimensions to every successful implementation: production ready technology applied to a problem it can solve by people who have done it before. The wrong technology for the business, the wrong technology for the use case, or the wrong people for the technology equals a money pit.

The 3 most important technical questions for senior leaders to ask are:

  • Is this technology production ready?
  • Is this the right technology for the business problem?
  • Are our people skilled with this technology and have they used it to solve this challenge before?

Companies save themselves from multi-million-dollar rabbit holes by asking the right questions and knowing what the answers should look like. I am going to give you a couple of examples starting with…


What do we want? CHATBOTS! When do we want them? I’m sorry, I didn’t understand that.

Chatbots are a great example of the thin margin between gold and money pit. There are three tracks to chatbots:

Simple – No need for any AI components.

Intermediate – Simple data science algorithms with support from 3rd party software.

Advanced – Complex machine learning approaches and large, labeled datasets.

For many projects sold as AI, there is no substance behind the label. That is the case for a simple chatbot. Many projects associated with the AI label have simpler implementations that use traditional programming approaches.

An important but often overlooked question is, ‘Do we really need to use machine learning?’ What is the alternative and how does that alternative perform? Data science and machine learning teams need guidance to avoid over engineering a solution.

On the other end of the spectrum is the advanced chatbot. Most data scientists who have never worked on an advanced chatbot, assume it’s just a bit more complex than the intermediate. It’s sold as almost the same thing. That assumption can waste a year of R&D effort to realize the business needs a more experienced data scientist.

‘Do we have someone who’s taken a solution to that level before?’ This question is important for any project, even ones that seem simple. With machine learning, a couple of features can take a project from simple to advanced. It’s important to gauge where the line is and whether those features are worth the extra expense.

Data science and machine learning teams need guidance to avoid overestimating their own capabilities or overlooking the complexity of a specific feature.

Streaming, Realtime, Scale

Those 3 words are expensive in data science and machine learning but they don’t have to be. You will often hear Hadoop, Spark, or any number of other solutions to manage that complexity. Those are the right technologies for the problem. Are they the right technologies for the business?

Many projects get hung up on selecting a solution that’s right for the problem without considering if that solution also fits the business model and core competencies. There’s a difference between solving a technical problem and solving a business problem. Data science and machine learning teams need guidance to navigate that difference.

Streaming and realtime start small but get larger quickly. Is the business similar to Netflix or Twitter? If the technology isn’t going to become a core component of the business model, it’s better handled externally.

That’s a hard call to make early on when the projects are small. The question doesn’t change. Does this technology align with the business model or is it a tangent? Tangents are difficult to justify managing internally because the team that supports it becomes a ballooning cost center almost immediately.

Technical projects never get simpler with time. In place teams become a temptation for feature creep. If a team is there, why not use them? The technology is in place and meant to scale, so why not take advantage of that? External teams define the cost of new features up front and that is a powerful deterrent to feature creep. Internal teams often don’t factor in the costs of scaling properly.

Conference Driven Development and Paper-ware

The line between academia and business is blurred for AI. So much of what we use comes straight out of research papers. That doesn’t mean that everything published is productizable. That’s a problem companies like Google and Facebook are struggling with. They spend on research but don’t often see a return on that investment.

For them, that disparity is a business necessity. They need to be on the bleeding edge. Does your business model support that same need? Are the business’s goals dependent on increments in accuracy or performance? Does the strategy call for first to market features? There are times when the business should say yes to innovative approaches.

Anything theoretical can be built, but can it be productized or applied? Data science and machine learning teams need guidance to insure projects create tangible results aligned with business goals.

Is this technology production ready? The team needs some evidence that what’s been published or presented at conference is something they can turn into a product. The team needs to connect the dots between business needs and a novel solution.


You can see senior leadership’s role in monetizing data science and machine learning. Most teams are left to themselves with little oversight. That doesn’t work. The successful oversight role asks the right questions to keep the team’s work aligned with the business’s goals.

AI is one of the few technologies that can live up to its hype. There are a lot of projects senior leaders should give the green light to. Having the resources to execute on those projects requires the business to say no a lot. The role of senior leadership is to sift through what’s possible to find what’s profitable.


I’m a top applied data scientist. I’m called a thought leader by IBM, Intel, SAP, and many others. I’m a contributor at Fast Company and Silicon Republic. I teach senior leaders and executives about what AI can REALLY do so they can monetize it. Right now, not 5 years from now. With your data, not Google’s. I speak from the perspective of someone who has built AI systems; making their potential relevant to real world business needs. You can connect with me on Twitter: @v_vashishta LinkedIn: and email:

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *