Artificial intelligence (AI) improves prediction. AI is computer software that learns, finding complex patterns in massive amounts of data. Improving predictions is obviously very useful for most businesses, with a wide range of applications. For example, AI is helping industries from manufacturing to air travel predict when key equipment or airplane parts will fail, reducing downtime. A servicer of printers and copiers can determine when they will run out of toner, based on how many and what types of print jobs are done. The machines will automatically order more just-in-time toner, preventing waste from replacing toner cartridges too early. Google Translator AI “predicts” (recognizes) equivalent speech in different languages. In all of these examples, the AI gets better as it takes in more data, or “learns.”
But businesses can throw a lot of money at AI simply because it’s the “in” thing. So, owners must improve the odds of success.
First, figure out what problem you want AI to solve. What mission critical information are you dying to know? The problem may not require a solution as complex as AI. Basic snags could simply require statistical analysis, such as linear regression. Another key question to ask yourself: Has AI ever been used before for that application? If not, then being the first may be a tad costly. Run small experiments to determine usefulness before committing to a massive project.
Second, do you have good data to work with? Remember the GIGO principle: If the data are old, unstructured or disjointed, then you’re not going to get good results, period. And you need a lot of data—AI is only useful when it has much to work with because its whole purpose is identifying complex patterns in large data sets. Thousands of data points are a minimum, and millions are best. Also, your data needs to be labelled so a machine can understand what it is. Provide the context that is creating the data from the start so that the AI doesn’t make a boneheaded mistake, such as treating Christmas-time sales as the normal monthly level.
Third, AI learns or evolves with experience in real world operations. It will be wrong at first. So, does the application allow room for growth, or is it so important that it has to be right the first time? AI cannot be perfected solely based on in-house training data.
Fourth, AI requires development—there is no AI “plug and play.” It is unlikely that a company has the expertise to do this in-house. So, finding the right contractor that has the specific knowledge for a particular problem is important. Trusting the contractor is key: AI programs are not very good at explaining how they got their results, so you’ll have to take the contractor’s word for it. Have your experts examine vendors’ results and evaluate both their accuracy and usefulness. Today’s main AI platforms are IBM’s Watson, Amazon Machine Learning (part of Amazon Web Services), Microsoft’s Cortana Intelligence Suite, and Google Cloud Platform. (A website listing a plethora of specialized AI vendors follows below.)
Finally, don’t relegate AI to the IT department. AI will boost worker productivity, but it can also shed light on what lines of business strategy to choose, so CEOs should be fully informed.
Some AI vendors worth a look:
Health care: https://www.techemergence.com/machine-learning-healthcare-applications/
Advertising and marketing: https://www.techemergence.com/artificial-intelligence-in-marketing-and-advertising-5-examples-of-real-traction/