Is certainty achievable or even desirable in business?

Eliminating risk
The second article was on a more familiar topic and describes how tech start-ups are using observation and data mining to help retailers and entertainment venues be more granular and efficient in running their businesses. In the article Ken Martin, executive director of global sales at Cisco, suggests, "data eliminates the risk." From real-time observation of shopper activity to match making retail facilities and tenants data and algorithms are being used to guide a wide variety of retail business decisions.

You can reduce risk; you cannot eliminate it
The thing that strikes me is that the two articles highlight opposing objectives.
The artist seeks to let art emerge; the outcome is uncertain. The retailers seek to control financial returns and eliminate risk. And I am pretty sure that desirable though the latter objective might be, it is also pretty much unachievable. You can reduce risk; you cannot eliminate it.
Perhaps more importantly, in business and marketing specifically, you need to be very clear about what you are trying to achieve. Do you want transformation, in which case you need a creative mindset, or do you want optimization, in which case you need an efficiency mindset?

Unreasonable promises
The article reviewing the use of data mining, machine learning, and artificial intelligence includes a comment from Mark A. Cohen, the director of retail studies at Columbia Business School, as follows,
"I'm a fan of fact-based decision-making, but there are a lot of charlatans promising things that aren't reasonable in terms of outcomes."
I too am a fan of empirically based learning, and I believe there are many problems that big data and A.I. can solve which traditional approaches cannot, but lots of data is not necessarily valuable data and A.I. is not yet omnipotent.

Good at counting and pattern matching
As of today, all the evidence I see suggests that A.I. is good at counting and pattern matching. The word counting might seem a little demeaning to sophisticated algorithms but counting can be very powerful. What might be monotonous and time consuming for humans is a breeze for an algorithm. Whether it is monitoring shopping behavior in real-time or calculating wait times, machines and algorithms can conduct an accurate census not just make an estimate. Everyone or everything is counted.

A.I. is also great at pattern matching. I have the PictureThis app on my phone to help me identify plants I find in and around my garden. PictureThis claims to use "revolutionary artificial intelligence technology" and to identify 1,000,000+ plants every day, from a database of 10,000+ plants, with 98% accuracy. The blurb in the App Store suggests 98% accuracy is "better than most human experts." Until recently I was blown away with how well the app worked. Whenever I find a new plant, I simply open the app, take a photo, and seconds later I know what I am looking at, along with all sorts of related information. Brilliant.

Even revolutionary AI is not foolproof
Or is it? On a whim I decided to test the abilities of the PictureThis app. I drew the outline of an imaginary leaf and then used the app to identify my creation. Hey presto! Without hesitation the app identified the leaf as Myioteris Gracilis or the Slender Lip Fern. Now either that is an act of extreme flattery (judge for yourself how lifelike the drawing is), or the lines that define the curve of the stem, the wavy edge, and pointed tip most closely matched Myioteris Gracilis in the database. The A.I. did its job, but any human would likely have identified the image as a product of imagination not a real plant.

Not so good at forecasting
If nothing else, the example of Zillow's recent debacle should provide a cautionary example that algorithms are not infallible when it comes to prediction. Zillow thought its A.I. based "Zestimate" guaranteed a good return on home-flipping, until, caught out by changing context, it found itself holding a $3.8 billion inventory of homes it could not sell, losing millions on the homes it did sell, and ended up closing its home-buying business. (Although to be fair, it is not like humans are much better at estimating the value of houses when faced with a hot market.)

As J. Walker Smith put it in a message to me, "My bottom line is that (the Zillow example) illustrates the limits of the current state of AI. It was a model at scale that couldn't learn, couldn't adapt, and couldn't stop in time." He believes that predicting the real estate market presents even the most sophisticated AI with some big challenges. Not only is the real estate market subject to big shifts in macroeconomic factors, but it is also incredibly local, with a host of variables determining how much people will be willing to pay for a specific property.

Changing circumstances and missing data
Zillow spokesperson Viet Shelton told CNN Business that,

"The challenge we faced in Zillow Offers was the ability to accurately forecast the future price of inventory three to six months out, in a market where there were larger and more rapid changes in home values than ever before."

But the same article suggests that even at a point in time the Zestimate have a median error of 1.9% for homes that are on-market, but 6.9% for homes off-market. That is a lot of wiggle room, which I suspect is largely down to a missing data problem. This is not just a matter of problems that come up in the inspection. How do you account for the buyers who made their purchase because, "We just fell in love with the place"?

It is easy to write off Zillow as a one-off, but I believe there is a wider lesson to be learned. Churning through huge datasets does not eliminate risk. When the context changes faster than an algorithm can learn, the probability of erroneous results increases, sometimes with dramatic consequences. And the biggest problem is that because the analysis is unsupervised and based on huge data sets, it is almost impossible to predict how, when, or why an A.I. will make mistakes. So, unless you have a human involved with the experience and ability to identify odd results, you can end up with nonsensical outcomes.

Do we really want to eliminate risk?
But then, mistakes are not a bad thing if you can learn from them. A.I.'s can learn, but within a specific context and the risk of error still remains. But maybe that risk is not all bad, provided it leads to evolution not inertia.

Agnieszka Kurant is very clear in her opinion that elimination of risk leads us, "Nowhere good." In the New York Times article, the artist defends risk by referring to evolution and listing serendipitous accidents like aspirin, X-rays, and Viagra (to which you might add Post Its, Velcro, Penicillin, the microwave, and Botox). But to my mind there is a huge difference between the trial and error of evolution and that of accidental innovation. The difference is that it takes human intelligence to recognize that a mistake could be good for something other than it was intended for. An A.I. would probably just classify it as irrelevant to the problem it was designed to solve.

Humans possess a wealth or potentially irrelevant experience and learning which means they can apply common sense and imagination, and better understand cause and effect. And those skills are incredibly important in business.

Progress in business requires transformative insights
If you work for a big company, it is likely locked in a battle for market share with look-alike competitors. Analyzing the same data, with the same analytics, and the same mindset is just going to end in stalemate. (Do not kid yourself. You are not unique in having a big customer database, deploying revolutionary A.I., or trying to analyze the crap out of everything.).

All the evidence suggests that the only way to gain ground in the battle for market share is to deploy innovation and creativity to disrupt the status quo, and that requires applying human intelligence and accepting a degree of risk. It implies finding transformative insights, not just looking for patterns in existing data. It implies empathizing with your potential customers not just observing them. And it implies making customers lives better, which means a change from the way things work today and all that lovely data you have collected and analyzed so assiduously.

Risk reduction is laudable
Just in case you think I am a complete nihilist let me state for the record that I believe risk reduction is a laudable exercise. Reducing waste - be that resources or dollars - is  good thing. 

I have no problem with Adam Henick, a founder of Current Real Estate Advisors, when he says, "If you're spending dollars, don't you want to spend them as accurately as possible? I think that's the benefit of data." Note the caveat, "as accurately as possible." Zestimate is a pretty accurate model when used as originally envisaged, but I doubt its accuracy can be improved dramatically, simply because many variables affecting the final sale price go unmeasured.

And I am not suggesting that you cannot reduce the risk involved in innovation and creativity – if nothing else, you can test your ideas with the potential customer – but you will never eliminate risk, and even if you did, the result would likely be inertia not progress. This is why I think that successful innovation and creativity requires us to relinquish our desire for certainty.

So, that is why I believe certainty is neither achievable nor necessarily desirable. But what do you think? Please share your thoughts.