A reader wrote recently that he believes
Shutterstock’s efforts to add huge quantities of image to its collection, as I discussed in a recent
article, is a “calculated plan” to eventually eliminate the need for photographers.
He believes that by “feeding these images into their AI machines they will be able to learn exactly what their customers want.” Then, it is his contention, that Shutterstock will be able to “deliver custom made images (which Shutterstock would create) to clients.” He argues that ”this would just be machine work and there would no longer be a need for image makers or even computer graphic designers.”
He compares the gathering of customer data to what UBER has been doing as it moves toward autonomous cars. In the early stages of their business UBER has used drivers to build a customer base, but eventually they hope to eliminate all that driver costs and keep 100% of what they earn from every ride.
It should be noted that
Yahoo Finance recently reported “In 2018, Uber (UBER) reported an operating loss of $3 billion – that’s on top of losing more than $4 billion the previous year. Still, it didn’t scare away investors. Some analysts predict the ride-hailing app leader will turn a profit – eventually (2030).”
Setting aside UBER as an example of the value of gathering such data, I think there are several reasons why more and more data from such a huge and ever growing collection may not benefit Shutterstock in the long run.
Reasons It Won’t Work
1 – To begin, knowing exactly what customers will want in the future, assumes you can then produce the imagery with machines instead of people at less cost than you can currently obtain images from human producers.
This seem highly unlikely. Currently, Shutterstock gets all its images at virtually no cost to the company. Image creators absorb all the costs involved in creating the images and supplying the information (keywords) that will make the images findable by customers.
Shutterstock has no production costs.
If Shutterstock were to produce custom machines that could create visual imagery to customer specifications, they would certainly have some costs. At a minimum, it seems likely that some in-house staff and maybe graphic designers would be required.
(UBER may be able to get rid of its human drivers, but it will have to invest in huge fleets of vehicles that will be located in all the right places to transport their paying customers.)
2 – About two years ago, Shutterstock developed a line of business called “Flashstock” or
“Shutterstock Custom” that enables customers to request custom shoots by human photographers. They spent $50 million to acquire Flashstock Technologies. There is no indication that Shutterstock Custom is generating much revenue at all.
There were two problems. (1) The photographers weren’t paid enough to make it worthwhile participating. (2) The customer needed to clearly specify what they wanted and then wait, at least a few days, for the imagery to be available. The images weren’t instantly available for download as is the case with already produced stock photos. Then, there was always the chance that the images the photographer actually produced wouldn’t satisfy the customer’s need and the customer would have to do something else anyway.
3 – Basic sales records do help understand what customers want. We know that many customers are happy to use the same images other customers have used in the past. To satisfy these customers all that is needed is to show them everything others have previously purchased. If you have a large enough cross section of previously used images, somewhere in that group should be the right image for the next customer.
If we go back to the early 1990s Tony Stone used to say that he could supply all the needs of every customer with only 4,000 perfect images that covered the whole range of concepts customers ever requested. Each image would be used multiple times. Tony built his catalogs by carefully selecting only those images that would better fulfill future customer needs than anything he had published previously. However, he kept publishing catalogs with even newer images built around the same concepts.
However, some customers want an image no one else has ever used. It is unclear what percentage of total images used are ones that were previously licensed compared to first-time licenses. We know that many image are licensed hundreds and even thousands of times by different users. My guess is that over 50% of total licenses fall into this category, but I doubt if in recent years anyone has really tried to track how often this happens. If they have they are not sharing that information.
That still doesn’t tell us how many new, different images will be needed in the future and where the agency will get them. What is really unclear how many more images that no one wants should be continually added to a collection in order to determine what customers
don’t want to use?
4 – It is highly doubtful that AI will generate enough specific data to anticipate each customer’s future needs and enable Shutterstock to produce such images, and have them readily available when the customer actually needs them. The machines would also have to be able to produce the images at less cost than the minimal royalty Shutterstock is currently paying creators when an image is licensed.
5 – It is interesting to consider how much additional revenue “more images” actually generate. At the end of Q2 2015 Shutterstock had 55.5 million images and Adobe Stock had 40 million. By the end of Q3 2018 Shutterstock had grown their collection to 233 million images, more than 4 times what it was in 2015. Gross revenue for each of these quarters was $104.4 million in 2015 and $149.25 in 2018, a 40% increase over more than 3 years.
Meanwhile, Adobe had grown its collection to 100 million images by Q3 2018, only 2.5 times what it was in was in 2015, but Adobe’s revenue has been growing at a faster pace than Shutterstocks. AdobeStock saw 25% growth in revenue in 2019 alone. (See this
story.)
6 – Probably the biggest problem with the Shutterstock philosophy is that a huge percentage of the images they offer are seldom, if ever, seen by anyone. AI only offers reliable data when all the potential users consider all the options. But, that doesn’t happen because there are way too many options for any human customer to consider. (See
here.)
As a result, each customer considers a different set of options. I estimate that less than 10% of the images in the current Shutterstock collection have ever been licensed by anyone. That percentage has declined from about 80% four years ago as they add more and more images.
Thus, each customer is shown a selection of imagery that includes some popular imagery and some new, never used imagery. Every day there is an increasingly high chance that no customer will be shown an increasingly large number of images that are buried somewhere in the collection. In some cases, a buried image might be perfect for a particular customer’s need that day, but the customer didn’t know it existed. The odds of this happening increase with every new image added to the collection.
Solving this problem is not easy, but finding the right image was easier for the customer when there was more human judgment involved in what to show customers. This is part of the reason why many customers still turn to smaller, human curated collection rather than collections that in theory offer everything that no customer has time to review. Where is Tony Stone?