How enterprises determine whether to buy or build AI models
From the USPS to appliance company Conair, organizations employing machine learning technology sometimes need to determine whether it's better to buy or build capabilities.
NEW YORK -- Enterprises are using AI systems at an accelerated rate across their business.
From chatbots to virtual assistants, the ai market will grow from $387 billion in 2022 to $1.3 trillion in 2029, according to Fortune Business Insights.
While most large enterprises buy AI software and systems from vendors such as Google, Microsoft and AWS, others pull in internal resources to build their capabilities.
However, the question of building or buying depends on the application, according to business leaders at The 2022 AI Summit conference.
USPS
One of those leaders is the CIO for the U.S. Postal Service, Pritha Mehra.
Mehra oversees the USPS' drive toward large-scale modernization and transformation across services such as logistics and customer engagement platforms using machine learning.
For example, for customer engagement and to improve customer experience, the postal service developed ML algorithms to predict when mail workers will deliver packages. The government agency also uses conversational AI technologies, such as virtual agents, to assist with customer inquiries.
The postal service has both custom-built ML models that it created and other models that come from vendors, Mehra said.
A determining factor is whether the capability or models needed are available in the market or not, Mehra added.
"I do not want to build something that's already out there that's excelling in its space," she said during a discussion at the conference on Dec. 7.
In its quest to modernize its IT ecosystem, USPS examines capabilities available in the market. Then it determines what fits its architecture and what it hopes to accomplish both in the present and the future.
However, even when buying technology from technology vendors, Mehra said the mission and objectives the company tries to achieve are in mind.
USPS CIO gives presents at AI Summit in New York.
Unilever and prioritizing
Unilever, a multinational consumer goods company, buys about 80% of its AI technology from vendors and builds the remaining 20%, said chief enterprise and technology officer Steve McCrystal.
"We've believed for a long time that if we get the right partners and we connect our engineers with their engineers, we can influence product better than if we try to build it on our own," McCrystal told TechTarget.
I do not want to build something that's already out there that's excelling in its space.
Pritha MehraCIO, USPS
However, due to the size of Unilever and the multiple brands the conglomerate works with, there are times when AI capabilities aren't commercially available to address a particular problem or need.
"Those are areas where we'll apply our own thinking to it," he said.
The build or buy decision also depends on timing, said Sol Rashidi, chief analytics officer at cosmetics products giant Estée Lauder, during a breakout session.
If time is of the essence, then buying AI tools is better than building.
"I always say leverage what's already been solved for," Rashidi said. "The technology itself is not the novel application. It is the aggregation of the technology that creates the novelty because you've officially solved the problem. And if you can do it in four months versus 14 months, guess what? You'll be a rock star."
When building AI capabilities makes sense
However, building may sometimes be the better route.
"When you go into a buy, sometimes not all the elements that are out there in a buy are available to you," said Dara Meath, chief information officer of Conair LLC, an appliance manufacturer, said in an interview.
And in-house talent might be needed help use what's in the buy.
"Sometimes it's easier for you to build because you have the tech staff in house, and you also have the models in house too," she said. "Sometimes it's faster to go to market with that."
However, enterprises sometimes can also buy and educate in-house staff on how to use AI models.
"It's kind of like looking at it and targeting it out," Meath said. "Planning efficiently is the key to it and then figuring out, 'Okay, what's going to be the best?'"
