In uncertain times, business executives are starting to focus their video-chat enhanced abilities on driving efficiency in their organizations. For many of them it is – all of the sudden – do or die time. Many have been hearing about Machine Learning (ML) based AI in recent years, but simply haven’t had the time to explore the topic. Now, for many execs, AI is part of their contingency plan going forward.

But how do organizations best apply AI? Is it like butter, just spread it on?

Not really. In fact, there have been many well-documented enterprise AI projects that have failed miserably to net results and were also quite expensive. But it doesn’t have to be that way. A bit of forethought and sound choices before embarking on AI projects can prevent disasters. Many times, you don’t have to start from scratch – there are pre-done packages for AI.

In fact, one of the biggest decisions you may want to drive your team towards is “do we need to develop the AI ourselves from scratch or are there pre-done solutions we can more easily apply to our business?”

The breakdown of those two extremes will be reviewed here.

1. Custom developed AI projects for business problems

Even if a company is willing and able to hire an expensive Data Scientist to analyze the business problems and develop Machine Learning Models, getting the ML Models into production is non-trivial. A Machine Learning Model is a combination of the program coded algorithms and the data-set that is used to train the model to perform a certain way when used on other data. An example could be a model pointed at existing HR data trained to find the characteristics of high performing employees. Hopefully a model that will continue to learn as new data comes in.

That new Data Scientist you just hired will probably have success in getting to know your company and finding potential business problems to solve. Any of them worth their salt will also be able to develop and train a model that runs successfully for a given solution, on his or her laptop, within a few months. But after that, things often fall down. When the model is put into production on company servers, it rarely performs the same way as that stand-alone pilot. Often this is due to other constraints and interactions in the computing environment or just because the company does not have enough of the right kind of data to prove the model out in the real world. The amount of relevant data and data cleanliness (the right, usable formats) are the most often cited reasons AI projects fail.

Another consideration for implementing AI is where it will be deployed. Computer hardware architectures for AI are different than traditional Database or ERP hardware solutions. Computer servers with Graphical Processing Units (GPU’s) deployed with MANY servers (nodes) are often required to run these deep learning models in training and production mode. An alternate to buying hardware that may save money is “renting” AI enabled resources in the cloud, from places like Amazon (AWS) or Google (GCP). But as AI becomes a core part of your company’s efficiency (hopefully it will) your rental costs can get quite high as well.

When you get through those challenges, many companies realize too late that AI models will have to be carefully maintained, which is difficult. Even if the AI problem-solving model works well for a month, that doesn’t mean it will continue to perform well. Environments change, your original problem elements change. Enterprises job titles that maintain developed computer programing code are known as Developer Operations – or Dev Ops folks. The trouble is most Dev Ops people are not trained on ML Models and their maintenance – which is quite different than standard software development environments.

This is why Machine Learning Operations – or ML Ops – is a very new, very expensive and a very highly in-demand employee position for you to try to hire. Between acquiring Data Scientists, expensive computer hardware or cloud time and ML Ops people – this can all truly amount to multiple millions of dollars spent.

Given the above, you may be surprised to learn that in some circumstances, Custom Developed AI is actually an acceptable and required risk. But only embark on Custom Developed AI if your business’s entire usefulness is driven by the AI, i.e. if AIis your main competitive advantage. Uber or Lyft come to mind. Those two companies are more than just about the recent innovative concept of riding in a stranger’s car, they are about unique custom AI capability. Or, if AI is not your core business, but you still need unique AI solutions that your competitors cannot easily buy off the shelf – Custom AI may be acceptable. There are ways to manage projects to avoid failure and minimize the highest costs. Custom AI can succeed with the proper planning and budget.

2.  Pre-done AI Solutions – there are good ones that are simple and lower cost

It is extremely important to first understand what your goals are in applying AI to your business – what outcome do you want? Have other companies solved this problem with AI? Is there anything on the market that already fills this need, either within an existing software suite you already license or with an AI niche company?

The easiest way is to implement AI, if you can, is a pre-done, packaged service from a focused company or an extension of your existing enterprise software. An example of an existing solution suite that has added AI recently is the Einstein modules from Salesforce.com (See previous Blog).
Licensing a cloud-based, off the shelf AI solution is often simple, comparatively inexpensive and very common now. Company departments are able to budget for this on their own as of late. (Make sure your IT department approves the security and privacy standards, of course). This is exactly analogous to the way software procurement has always been. Buying a packaged software solution was almost always preferable to creating your own custom program, except for the above stated exceptions.

There are many business functions in many business horizontal activity areas that can benefit from AI.
Taking one corporate department as an example – a company’s Sales Dept – here are examples of cloud-based pre-done AI capabilities you may be able to make use of now:

• AI powered Account Based Marketing (ABM) platform
• Smart calendar scheduling
• AI filling in CRM actions for you while you are on the run
• AI that helps targeting prospect companies
• Tracking all communications and the relationship strength in your account interactions
• AI capturing and outlining meeting notes without you typing
• Automating lead generation and scheduling follow-ups with AI-as-service
• AI platforms for data-driven customer engagement using deep machine learning
• Sales Acceleration by speeding up dialed customer phone calls
• AI accessing the right people at the right accounts, at the right time, with the right messages
• Website messaging using AI to scale conversations for more Sales Accepted Leads (SALs)
• AI finding Ideal Customer Profiles (ICP’s) and Accounts
• Ai driven Sales Playbooks and the execution of them. Like a remote AI SDR
• Lead Nurturing via Email, Web Chat and SMS
• AI generating personalized marketing messages in email, social accounts, advertisements and other channels
• Finding hard-to-find B2B Data with AI. Penetrating the right target accounts with differentiated data
• Multi-Industry Assistants using voice, text and any channel to keep the service interaction going

For more detail see AI for Sales

Conclusion

Artificial Intelligence assists humans with business in ways that can surprise and delight us. Putting some thought into the Make or Buy AI decision will save money and accelerate time-to-benefit. Before considering the tougher road of custom AI development, first exhaust looking for the low hanging fruit by considering pre-done, packaged AI services. There are a number of pre-done AI services for many business horizontals including (but not exclusive of) the Sales Department.