Sergey Nivens - Fotolia

Software creation is not for every startup

While it may be beneficial in the long run, creating AI software requires patience, investment and the right cast of employees. For one company, the investment paid off.

It's a decision all startups and young companies need to address: whether to license software or whether to build it in-house.

Software creation is an expensive time-consuming proposition, but the end results could outweigh the cost, depending on the goals for the company.

For a company like ReplyYes, based in Madrona Venture Labs, a startup incubator funded by Seattle-based Madrona Venture Group, the end goals of curated discovery and easy transactions for its customers meant it was more logical to build out the software in-house, rather than licensing existing products.

"There are plenty of tools that make a ton of sense to take advantage of, but for us the question was 'What do we want to be great at?'" said Dave Cotter, CEO of ReplyYes. "For us to own the things that light up curated discovery and easy transactions is super critical."

In its second year, ReplyYes is a commerce service that sells vinyl records and graphic novels through its two "stores," The Edit and Origin Bound. Users interact via Facebook Messenger or SMS, first curating their taste in music or graphic novels before ReplyYes begins its daily recommendation. After having a basis for your specific tastes, the company will message you once per day with a recommendation, and you can choose to message that you like or dislike the recommendation, or message back that you already own it or you wish to buy it. Each option helps influence ReplyYes' future recommendations, further tailoring the process to your taste.

If we get a piece of content that can't be interpreted by the AI, it goes to a human who interacts with the customer.
Dave CotterCEO, ReplyYes

"We realized early on that if you keep the experience simple and easy to digest, we could create a personalization engine so that instead of sending out the same album to all users, we can send out specific albums based on feedback," Cotter said. "The way we approached our AI, we kept the components very narrow so customers wouldn't run into a dead end."

'You own everything'

Personalization engines for e-commerce are of growing importance, according to Gartner. The research firm predicts that by 2018, companies that have invested in digital personalization are expected to outperform companies that haven't by 30%.

If all things were equal, most companies would probably choose software creation over buying software. Beyond saving money with licensing fees, the proprietary aspect of owning the software would pay off in the long run. The issue with this avenue, however, is that every company isn't in a position to build something like AI software from the ground up.

"Just because you can doesn't mean you should," said Kenneth Sanford, U.S. lead analytics architect for New York-based Dataiku, a data science software company. "It takes so much time and expertise. And the biggest reason this is challenging is hiring the correct talent is extremely difficult."

Cotter said ReplyYes would have gone a different path if it didn't have its current engineering team in place, which Cotter said had the experience and background to create personalization engines and other software.

"You can stitch tools together and get to market quickly at 60%, but that last 40% is about what unique thing you bring to the table," Cotter said. "That takes longer than off-the-shelf software. The pro is you own everything. The con to that approach is it takes much longer to get right."

The cost of software creation can also play a factor, as the talent needed and time invested can add up quickly, but it can pay off in the long term.

"You can buy something off the shelf and off you go pretty quickly," Cotter said. "When building it yourself, it's more expensive to hire the people and you have to incur the market wait period. But in the long run, it's exponentially more valuable because we own these experiences."

Cotter added that maintenance of its software is easier and quicker to manage, because ReplyYes doesn't have to reach out to a third party when issues arise or updates are needed.

"That is really critical when companies are going from startup to scale," Cotter said.

'The process is very iterative'

Software creation from the ground up, if affordable and your company has the capable talent, can be very beneficial in accomplishing your company's specific goal. 

"Perhaps the most important aspect of building a service like this, not only is it quickly prototyped, the process is very iterative," Sanford said. "No one ever knows the transformation they are going to get with data until they run a model. The tweaking is insight that you're getting from the data and it's getting fed back into the modeling. Doing it yourself lets you incorporate that insight into the production process."

ReplyYes has worked on this continued process of feeding information back to its system, even using humans to augment its personalization engine.

"If we get a piece of content that can't be interpreted by the AI, it goes to a human who interacts with the customer," Cotter said. "As these interactions happen, we queue up the phrase or word that wasn't interpreted and feed that back to the AI."

Since its inception in October 2015, ReplyYes' record store The Edit has sold more than 100,000 albums and has 70,000 users receiving vinyl recommendations every day. As it continues developing and fine-tuning its AI offering, the uniqueness around selling an antiquated technology in vinyl record albums via a modern e-commerce and AI platform isn't lost on Cotter.

"I think it's one of the more interesting revelations," he said. "Using cutting edge AI engines paired with vinyl albums that essentially went dark until a recent renaissance is really interesting."

Next Steps

Is e-commerce right for small business?

Online music startup picks a personalization engine

Customer insight at center of customer experience

Dig Deeper on CRM tools and strategy

Content Management
Unified Communications
Data Management
Enterprise AI