From Data Entry to AI: The 10-Year Pivot

Samasource employees in Kenya

Jason Calcanis, startup angel investor in companies like Uber, interviewed me for his This Week In Startups podcast. I’ve included some highlights below.

This year, my tenth as CEO since founding Samasource, has been a time of great reflection for our entire team. What started as a small nonprofit in Kenya has grown into East Africa’s largest training data company, fueling the AI teams at the world’s top automotive and high tech firms, including Google and Microsoft. Today, over 75% of our 2,500 agents are focused on high-quality inputs to train the artificial intelligence behind many popular consumer applications.

How did we get to this point from very humble beginnings?

We hear a lot in Silicon Valley about scaling fast and pivoting on what seems like a quarterly basis, based on the brilliant insights of a founding team. But I thought I’d share the story of our 10 year pivot, based on the hard work done by our agent base, and what we learned from our evolution from a data entry company to a quality training data provider.

Moving Up the Food Chain

One of the challenging things about launching a B2B startup is that you are serving customers that have complex, multi-stakeholder decision-making processes underlying any purchasing decision (whereas a consumer might have to consult, say, his wife). Understanding this process when you’re not inside the client company is incredibly difficult. Even if you invest in great relationships and schlep to the client’s HQ as often as they’ll have you, you’re always at an informational disadvantage before you sign a deal.

The only way to fix this is after the deal is signed, by getting to know the sales cycle from the inside.

When I started Samasource, the first contract I could find was transcription work for an online library. I didn’t have the luxury of choice: it was my only sales prospect. So I signed them, for $30K, and then personally committed to doing all the QA on the process. As we did the work, we got to know the core team that managed the product, which led to improvements in our process and product, and finally an on-going contract for several years of work that evolved each year into more complex tasks at higher and higher quality levels.

We followed the same process at Google, where our first pilot contract came after I had over 40 (yes, 40!) meetings out of a spreadsheet of 70 contacts I’d amassed at networking events across Silicon Valley. From there, it took another year before we solidified a larger contract there, which we’ve grown over the last five years through a similar process.

Evolving into an AI Company

In 2011, I was included in Fast Company’s Most Creative People list. I’d been cautioned against attending press events like this (I disagree; attending industry and press events is a key tactic for growing sales if you don’t have a marketing budget and can’t get VC money, as was the case for me for many years), but I ended up meeting an AI leader at Microsoft who spoke just after me.

We shook hands, and the following year, we began training the model behind a popular Microsoft product, becoming one of the first enterprise vendors for high-quality training data at scale. Mind you, we’d never really built training data or thought of ourselves as part of the AI/ML ecosystem. At the time we just responded to a customer need for image annotation and did what we always do: listen and learn from our early pilots.

Product evolution at a B2B startup is as much about listening and learning quickly than it is about the brilliant strategic insights of a founder or founding team (as much as we’d like to pat ourselves on the back!). Our customers led us into the AI space, and have kept us here.

Standing Out from the Crowd

We’ve heard consistently, since 2008, that the crowd would come for us. That crowdsourcing firms with more nimble labor models would eat our lunch. That completely fungible workers with no living wage guarantees were more competitive for customers and for us.

But we disagreed.

From our perspective, paying workers living wages (based on the wage floor set by the Fair Wage Guide), creating a decent work environment, and reducing turnover are the best ways to ensure the highest-quality inputs for our clients. If your self-driving car or defect-detection algorithm is fed the wrong training data, there will be disastrous consequences for your business. Quality is king in our industry, and we’ve found that there are no shortcuts: the best way to guarantee quality is to maintain our own facilities, where we train and manage our own agents who have access to health insurance, subsidized transport and meals, scholarships, funding for entrepreneurial ventures, and even a nursing room. (Our Nairobi facilities are nicer than our offices in San Francisco, which reflects part of our culture — to treat our agents the way we’d like to be treated.)

All this means that we’re doing work that really can’t be done with a crowdsourcing model, because you might need somebody go through months of training. Or you might want the same person working on this work for years, because they need somebody that has subject-matter expertise in a particular type. We also have started training developers to do some basic kinds of coding on AI projects and customizations of our own platform that integrate with client software going forward. I’d love to see if we can train people to be machine learning experts and data scientists in-house.

What’s more, we’ve had to build our entire layer of middle management ourselves, because there are not a lot of people emerging from similarly-sized companies in the region that we can hire to be middle managers. That’s the challenge in a lot of emerging markets. It’s that middle management layer that can help you scale an organization. We’ve grown that internally, and it’s a vital part of what makes the agent experience at Samasource so motivating.

At our offsite this year in Northern California, we hosted two team leaders from Gulu, Uganda, who started as agents four years ago and have now managed teams of over 50 people each. The chance to move so quickly up the career ladder and learn management skills is rare in any startup, let alone one operating in Northern Uganda. And we think this kind of growth gives us access to the most motivated talent in the region.

Have any of you pivoted your company over a longer period? What have you learned along the way? Please share your experiences in the comments below.