Spare5: Humans Helping Machines Become Intelligent


Spare5 has devised a business model that draws on the power of crowdsourcing and the “sharing economy.” One of their principal functions is to help train AI (artificial intelligence) models – or simply put, to use human skills to help machines become more intelligent.

We recently had the opportunity to speak with CEO Matt Bencke and Andy Ganse, Principal Data Scientist, for Spare5.

Driving Crowdsourcing and “Sharing Economy” Innovation

Spare5’s business model involves a two-sided marketing proposition. On the one hand, the company creates groups, even virtual armies, of people with skills to help filter, clean and organize massive amounts of unstructured data. On the other hand, it attracts commercial clients who rely on using various types of unstructured data and are often overwhelmed by it.

The company started about one-and-a-half years ago. They saw an opportunity to draw upon the population of Seattle, which Bencke describes as including “thousands of intelligent people with insights, smartphones and free time.”

The basic concept was to offer these people an opportunity to earn extra money at their personal convenience – the essence of the sharing economy (think Uber) – and to draw on their individual skills and experience.

The company developed its Intelligent Crowdsourcing Platform. The platform utilizes crowdsourced individuals, who the company refers to as “Fives.” It carefully analyzes the skills of these people. In any given job Spare5 divides the work into micro tasks for assignment among the group working on the data.

The company has developed its own machine learning algorithms to rate each individual worker on specific tasks.

Business Focus: Types of Services and Customers

Spare5 points to forecasts that estimate that by 2022, 93% of all data will be unstructured (IDC). Principal sources of this unstructured information include: images, video, social media content and text messages.

Their clients are generally overwhelmed by vast amounts of unstructured data. Typically, clients attempt to deal with cleaning up the data themselves. In fact, the company states that its biggest competitive challenge is not other crowdsourcers, such as Amazon’s Mechanical Turk (MTurk), but rather companies’ trying to solve the data problem themselves.

The customer base, Bencke describes as, mostly “product owners,” specifically companies selling products online. They emphasize three areas of concentration: 1) training AI models, 2) improving Search, e.g., better aggregation, more descriptive terminology, 3) cleaning up data, especially meta data.

Spare5 filters task results “through a rigorous quality assurance process” that includes the company’s own proprietary machine learning algorithms. As customers use the platform, the company asserts, “the process becomes faster, smarter and better.” Human insights combined with machine learning provide customers with “clean, enriched, labeled data.”

Retail as an Example

In the retail area, Spare5 explains that it can enhance a retailer’s search capability, for example. Using the best terminology and putting it in the optimum place in the navigation system of a web retailer’s site can enhance the customer experience and lead to a higher conversion rate, i.e., more sales.

Spare5 executes such projects by assigning people who have particular experience and aptitude in using retail sites and can express a judgment on exactly what works, and what doesn’t. These are individuals who resemble the customer profile that the retailer is trying to reach.

The company states that it has undertaken a variety of tasks in this area, including: product keywording, product classifying, product matching, outfit creation, media categorizing, image editing, content rating, content comparing, price estimating, and sentiment gauging.

One of the keys to Spare5’s success is its skill in breaking down assignments into micro tasks so that the best-suited individuals can be used to focus on these. Examples of tasks include: writing image titles and descriptions, researching via phone or web to find missing info, matching similar or complementary products, rating help articles, ranking photos.

They describe their key selling point as “the right human in the right loop.”

IBM Watson Deal – AI

In October 2015 Spare5 announced its relationship with IBM Watson. The parties demonstrated a golf instructional app that was developed using Spare5’s crowdsourced data evaluation techniques. Spare5 stated that it expected to be working with IBM on “applications across a broad range of industries including sports, retail, healthcare, and life sciences.”

IBM Watson is both a partner and a customer. Bencke explains that Watson is using Spare5 for reliable data to train some of their domains, one being golf. Watson also resells Spare5.

Spare5 sees a significant market opportunity as various offshoots of AI are adopted in a wide range of industries. These systems typically must be “trained.” One aspect of the training involves feeding sets of data into the system, which are designed to address a desired outcome. Spare5 contributes to the validity and accuracy of this input data. Among the micro tasks that Spare5 undertakes for AI clients are: tagging, rating, classifying, titling, associating, comparing, annotating, and taking surveys.

Working with a Vast Array of Crowdsourced Talent

We asked how does Spare5 secure engagement, or commitment, on the part of the Fives. Bencke points out three approaches. One is simply the added income for the work, which is helpful even though a large number of their Fives have good paying jobs.

The second is Spare5’s attention to making the work “fun.” This is achieved through various gaming approaches. The third is simply individuals’ satisfaction at doing something that they know they are good at. In addition, he mentions that the company makes it easy for these workers to contribute to charities, if they wish.

They have been successfully scaling their approach and currently maintain a force of 45,000 Fives. “Jobs typically require hundreds of workers,” Bencke explains, “but we have some that go into the thousands.”

Business Growth Strategy

We were interested in the question of whether most data crowdsourcing jobs were simply one-offs, i.e., once you deliver cleaned up input, is that the end of the relationship? Bencke stated emphatically that they had established a platform, rather than a mere consultancy type of approach. He explained that while relationships start with a specific app that the client has in mind, they have found that there are ongoing needs to help with data ingestion issues. He also stated that the training models for AI systems need to be “re-trained” over time, since human judgments change.

The company has published some of its case studies. In one example, attorney-locator and advisory site Avvo has used Spare5 for data tasks that relate both to identifying and verifying information about attorneys who want to be listed on the system, as well as tasks that relate to classifying questions and issues that come in from users.

Spare5 currently has about 25 employees. About a quarter of these are data scientists and designers and another quarter are game engineers, with the remainder being in other business operations. It raised $10 million in 2015, from institutional investors. They currently offer an iOS-based app and a desktop-based one for their crowdsourced workers. They expect to offer an Android app during 2016. The service is sold primarily as a subscription.

Bencke also looks forward to growth in the area of helping customers with online directories.

Spare5 management are frankly complimentary to Amazon in the area of crowdsourcing. Regarding Amazon and its MTurk service, Bencke acknowledges, “They invented crowdsourcing.” But he adds, “We reinvented it.”

Our Take

Spare5 appears to have a strong foothold in the crowdsourcing business. They are proving out their ability to scale and manage a very large, and obviously diverse body of workers. Among their other assets are their own computer skills in using machine learning to evaluate this workforce and using gaming techniques to make otherwise rote micro tasks more enjoyable and interesting to the workers.

The sales cycle for their product seems to us to be the biggest challenge. As we have written in a number of articles about mobile cloud retail apps for example, there is a lot of resistance that has to be overcome to gain multi-billion dollar, omnichannel retail organizations as clients for new technology and new approaches.

Spare5 has a key factor in its favor, which is the powerful trend to create more and more unstructured data that must be dealt with, that all organizations are facing. The amount of pain from this unwieldy growth should only escalate, helping the company demonstrate the value of its services.

We find the company very interesting from another point of view, which is that they raise questions about the intricacies of the ongoing relationship of human and machine intelligence as artificial intelligence and cognitive technologies take over more and more of the computing universe – a subject we intend to highlight in future articles.

Visit their website: