top of page

The right way to put AI in business

Many companies are in a hurry to cash in on the AI gold rush, but data-intelligence startup Adatos sets itself apart by only focusing on profitable use cases, says its Managing Director Mr Drew Perez.

Interest in artificial intelligence (AI) is at fever pitch, prompting some venture capitalists to invest in AI companies that have little more than an idea, and large corporations to buy over AI startups that don’t even have a functional product.

“As transformative as AI undoubtedly is and will be, many people do not realise that not every problem needs to be answered with AI,” says Mr Drew Perez, Founder and CTO of Adatos, a company founded in 2015 to provide data intelligence solutions.

As psychologist Abraham Maslow once said: if you only have a hammer, there is a great temptation to treat everything as if it were a nail.

The feasibility barrier

According to Mr Perez, a former US intelligence analyst, there are four main constraints that an AI solution has to overcome before it can be commercially viable. First, training machines require large amounts of the right type of data—data that is actually insightful.

“Without data, you can’t send your machine child to a school to learn. But many industries don’t have what we call indicative data,” Mr Perez says.

For example, he explains, retailers might be interested to do market segmentation and customer profiling, but only have data from traditional sources such as point-of-sale systems or loyalty programmes. While these data sources might capture simple demographic information like age, gender or marital status, machine learning has shown that these variables may not actually indicate whether a given customer is likely to spend more or be loyal to the brand.

“When retailers find that the type of data they are capturing is not actionable, they try to force the machine to classify customers by their existing categories rather than using what the machines infer,” Mr Perez continues. “In those cases, the probability of efficiently training AI on that data is slim.”

The second barrier to the feasibility of AI solutions is the prohibitively high cost of hiring trained data scientists. The dearth of AI experts coupled with a global boom in demand for talent means that people who list machine learning on their resumes can now command sky-high salaries.

“Over 99 percent of the AI companies operating today are dependent on humans to build and tune models. The scarcity of talent is the number one reason why AI projects fail to bring in a return on investment,” Mr Perez says.

Making AI that is acceptable

Even if companies are able to develop feasible AI projects, they may find themselves stymied by the next two constraints: regulations and resistance.

“AI works and it works today; the question is whether or not we humans will let it work,” Mr Perez says. “Technology-wise, autonomous cars are already feasible; it’s just that the authorities have not approved them for use. And if a technology is not approved, you cannot realise its commercial value.”

In the case of medical imaging, the technology to make diagnoses is not only mature but has been empirically shown to be faster and more accurate than human doctors, he continues. However, AI has not been widely adopted in the field of medicine because of resistance from the medical profession. Similarly, the possibility of job losses is frequently raised as an objection to AI.

“The final constraint, which may be the most important one determining the success of a company, is whether or not the use case is profitable,” Mr Perez says.

After all, the ultimate goal of any business is to stay in business, and if AI fails to bring in more money than it costs for any of the reasons above, then it simply does not make business sense to deploy AI solutions.

“The majority of AI use cases are exploratory and don’t realise a return on investment,” Mr Perez says, adding that Adatos is therefore highly selective when it comes to deciding what use cases to work on. “We only focus on profitable use cases. We’ve done over 122 use cases so far and the majority of them are not profitable.”

Building AI that builds AI

One of the practical use cases Adatos has been involved in was a project to use AI to analyse satellite images of oil palm plantations. Although such imagery has been available for several decades, the dependence on human analysis has been a major bottleneck.

“Without AI, you may have a billion-dollar satellite but still require a human with a one-dollar magnifying glass to make sense of the data,” Mr Perez says.

Adatos helped a client with three years of backlogged data to use AI to scan through hundreds of thousands of hectares of oil palms, reducing the analysis time to a matter of hours. The analysis helped the company make better decisions on their use of fertiliser, which accounts for 50-70 percent of their operational cost, translating to significant savings and a reduction in their carbon footprint.

Apart from making sure that the use cases are profitable, Adatos also targets the technology side of the equation. Instead of relying on expensive data scientists to manually tune the company’s models, Adatos reduces costs by building AI that is capable of generating its own AI.

“A country with a smaller population like Singapore should embrace AI, as opposed to a place like China which has a lot of human capital,” Mr Perez says. “We have an advantage here with a favourable investment and innovation climate, and SGInnovate has been a great vehicle for us to start understanding some of the opportunities and challenges of doing AI in Singapore, the region and beyond.”

Read more from Asian Scientist Magazine at:

bottom of page