Guide to Enterprise AI Platform Selection

This is an article from DZone’s 2022 Enterprise AI Trend Report.

For more:

Read the Report

The use of artificial intelligence (AI) in companies is becoming more and more widespread, possibly even trending toward industrialization. Whether this is done on the basis of an existing data science platform or not, using all-in-one tools or not, or hosting data in the cloud or on-premise, there is a large number of software solutions available and, therefore, choices. This profusion of choices should not make us forget the raison d’être of any IT solution! Between those who promise you the moon and those who want the ultimate (and unfeasible) solution, you must stick to your needs more strongly than ever.

Part I: Assessing AI and Business Needs

Don’t Buy Into Buzzwords!

The first thing to know about assessing AI is not to buy into buzzwords. You know, terms like big data, IoT, blockchain, and so many others where “the revolution expected” was not the reality. I’ve seen several instances where IT were told that they had to implement a specific new technology without thinking about the business need, defined here as the needs of the team and organization. So more than ever, focusing on the need is the very first step to success. Sometimes, AI is used for cases that can be solved by a simple “if-then-else” statement, when AI is actually the most useful for problems that are difficult to solve by simple algorithms.

What Is the Need?

Of course, this means asking why, and more importantly, what objective you are trying to achieve. Often, when needs are formulated by management, they are not necessarily complete. For example, if you are asked to set up an AI platform for a company, and the company’s shareholders are asking for profits to double next year, you need to take that into account. You must be aware of the company’s objectives, not only the objectives from your management, but also the organization’s needs and its consequences.

But let’s get back to the business needs. Of course, it is necessary to make the needs explicit, but it is always a good idea to master the use cases that have already been identified. This requires not just competitive intelligence (has my competitor implemented a relevant use case?) but also meeting vendors, visiting trade shows, and, of course, knowing your company’s processes.

Which AI Use Cases?

AI use cases are endless, but some are relatively recurrent. Here are some that come up often across multiple industries:

  • Marketing automation and definition
  • Sales forecasting, lead generation, and analytics-based training
  • AI in fraud detection (but can be achieved, at least partly, by CEP platforms)
  • Customization of services
  • Inventory management
  • Administrative tasks such as automated mail, file processing, and decision support
  • Decision automation (especially in legal and insurance)
  • Predictive maintenance

What Are Some Anti-AI Use Cases?

When wondering whether a use case should or should not use AI, it is worth asking whether a use case should be computerized. The big questions to ask are:

  • What are the consequences if an AI solution is wrong?
  • What are the implications if an AI solution suffers from bias?
  • Can a decision made by an AI project have legal ramifications?
  • Does it risk dehumanizing the customer relationship?
  • Will it bring real help in a use case where a human remains indispensable?

Part II: Build vs. Buy

When wondering whether to build a platform in-house or buy one externally, you need to answer a few more questions, starting with, “Is your need very specific or small?” If your answer to that is “no,” then you should be ready to buy! Here is a more extended checklist:

  • Compare the business plan between build and buy
  • If your need is a little specific, does the market contain AI solutions for it?
  • Are there already solutions proposed by vendors for your use case?
  • Is this vendor at significant risk of failure within four years?
  • Could you obliterate the competition with your ideas and own way to use AI?
  • Do you have a critical need that requires you to be fully independent of a vendor?

Part III: AI Enterprise Platforms

AI Capacities Checklist

Here is a list of capacities you must look into and the needs that an AI platform should fulfill:

  • Data integration
  • Data governance
  • Experimentation and development
  • Deployment and monitoring
  • Intelligence engine (ML programs, libraries, etc.)
  • Optimization capacities
  • Collaboration capacities
  • Visualization

List of Vendor Types

There are many vendors in the market, so it’s up to you to determine your needs. Here are two general categories of vendors you will encounter and some key differences between them:

Generic, pure AI platform Cloud providers’ solutions
Completion of offer Often complete and generally derived from data science; very knowledgeable about the subject. They have comprehensive, high-quality offerings.
Data integration They have greater facilities to manage the integration of data from the outside. Integrate well with their other cloud services. But integrating data that does not come from their own cloud offering is more complex, making multi- and hybrid cloud patterns a bit more difficult to implement.
Type of target user This often covers less experienced developers and citizen users. They are perhaps more oriented toward experienced developers and not citizen users.
Future Because this “specialized” activity is in a phase of concentration, it is necessary to be vigilant on the health and roadmap of the vendor. Roadmaps can differ from different cloud providers.

Alternatives to Enterprise AI Platforms

Enterprise AI platforms are not the only solutions for the use cases discussed. Two types of platforms may be relevant, depending on whether your use cases are simple or redundant in your industry — “business-oriented solutions” and robotic process automation (RPA).

“Business-Dedicated” Platforms

In some fields, you can have “old” vendors that sell solutions focused on one kind of subject. Especially in manufacturing, you have some historic vendors who embraced AI and offer ready-to-use AI solutions to help manage a factory, enable predictive maintenance, etc. These solutions can sometimes be straightforward to use and cover some of your use cases.

Robotic Process Automation

RPA is a bundled solution that tries to “robotize” human gestures. These solutions are complemented with OCR solutions, but they also can write and send responses via email to cover a number of AI use cases. The ROI of this kind of solution can be exceptional. Develop, managing dependencies between RPA and manipulated applications can be very difficult. Ideally, RPA should be considered if your business software rarely evolves.

Conclusion

Hopefully, these insights will help prepare you for when your boss says, “We need AI for these operations!” There is a big gap in understanding what AI can do and what we would like it to do. From assessing business needs to focus on the right direction, building it yourself or buying from a vendor, and managing on-premise or in the cloud, you now have the tools necessary to make the right choice for your business use case.

This is an article from DZone’s 2022 Enterprise AI Trend Report.

For more:

Read the Report

.

Leave a Comment