When Revolution Hits Evolution: Building An AI Startup In The MENA
Grow Your Business, Not Your Inbox
You're reading Entrepreneur Middle East, an international franchise of Entrepreneur Media.
Founded by Sahiqa Bennett and I, Searchie is a technology company, focusing on human resources (HR), which uses artificial intelligence (AI) and live video streaming to pre-screen and shortlist candidates.
Our proprietary machine learning software can accurately profile candidates, in real time, based on competencies, behavior and cultural fit. The feedback from the majority of people who view our platform is that Searchie will revolutionize HR; however, the challenges we have faced is how to simplify complexity into a market that may be looking for evolution, versus a revolution. The first challenge we have dealt with is around the whole understanding of AI. A lot of people are unfamiliar with AI, and quickly come to the conclusion that AI is all about removing humans from the work landscape. Now, Searchie can do a lot of exciting and time-saving things at scale that directly impact the role of a recruiter, from interviewing candidates while the recruiter is sleeping, through developing a profile, through analyzing data which explains how well candidates fits a role. However, what we can’t do is replace the role of people in the recruiting process.
Nevertheless, the most common feedback we received from prospective clients who pushed back were on things like, “We prefer the human touch,” or “Human relationships are important to us,” or “What if the candidate doesn’t do well in a video interview?” Therefore, the challenge to us was how to change our communications so that we could emphasize the role of Searchie in providing a tool to assist in making better and quicker decision, thereby allowing recruiters and senior managers more time to focus on improving talent, instead of spending hours per day reading through CVs and trying to determine the fit of new employees.
The challenge of education is arguably ongoing is likely to be a hallmark of companies pioneering this new technology across the world. We realized that the depth of communication we needed to have was heavily restricted by our available human resources, and this limited our ability to scale.
Enter distribution partners. An inflection point for our business occurred when we realized we could leverage channel companies that shared synergies with us to spread our message, and communicate the value of our product, which then drove sales. This isn’t easy. AI technology is complicated, and when you layer in organizational psychology, it’s very easy to overwhelm even the sharpest people in the room. So, we began exploring ways to explain topics at scale. LinkedIn Live has become a key method for keeping partners engaged in the conversations and keeping them up to date with the technology, science, maths, and industry considerations.
As sales have become more frequent, this in turn has caused a requirement to accelerate product development. Building an AI company is arguably more complex, more time-consuming, and more expensive than a typical software business. For starters, pre-launch, there is an extraordinary amount of R&D required to build, test, validate, explore, and productize. In our case, it took roughly two years of development work to get to market- this was so much longer than we originally expected.
Beyond the R&D, data science, organizational psychology, and maths resources involved in the team, there is an entirely new cost center that a typical software business doesn’t have to carry: data cleansing and data labeling. In order for AI technology to work, you have to feed the model an overwhelming amount of data. Usually, that data comes in two parts: one, the source, and two, the label (the ground truth). The de facto method for labelling data is to outsource micro tasks to a third party- alternatively, you can build your own pool of micro-taskers in house and the infrastructure to manage them, assure the outputs’ quality, share new jobs, and extract, transform, and load, etc. This cost is unavoidable. You have no model with- out this data. While the cost of a single task is inexpensive, the volume required is so great that it becomes one of the most expensive line items of your profit and loss statement.
To make matters even more challenging, there is no data labelling platform operating from the GCC or the wider MENA region. A quick search for data labelling in Dubai returns 1.8 million results for companies offering barcodes, stickers, nu- tritional labelling for food products, and other physical labelling equipment. We eventually settled on a data labelling plat- form based in the US, with micro-taskers all over the world. Updating, testing our labeling plans and instructions, or sharing feedback is obviously impaired by the time zone and availability issues, which are intrinsic in working with companies on the other side of the world. Bias influences everything around us, all the time. In some part, it is due to our neural programming and pattern recog- nition abilities. In other circumstances, we may fool ourselves into believing something that is inaccurate or not based in fact. Trying to color the lens through which we see the world as fairly and evenly as possible is difficult, particularly when your use case is subjective. We’ve taken steps at a technical, human, and behavioral level to reduce our exposure to group think.
Our team is comprised of 55% female and 45% male employees, hailing from 16 different nationalities, and ages ranging from 20 to the mid-40s. We also perform internal cultural analy- sis with our own technology to review behavioral diversity, and ensure we don’t submit to group think or conformity. All this complexity has had an impact on how we sell our product. We had been targeting annual subscription packages to large clients for a month, when the market signaled it wanted something more trans- actional, and to “test our technology.” We got to work and quickly integrated with a payment gateway that allowed us to pivot to a pay-per-interview business model.
Within a few weeks, this solution was in production. Which leads to my last secret: funding. The range of experience in the venture capital sector, combined with the depth of experience in AI, can create some friction in the evaluation process. In meetings with investors, we’ve fielded questions running from, “What if a candidate doesn’t like video cameras?” to “How do you imagine the prediction engine applying across different industries outside recruitment?” Eventually, they all converge on: “How much revenue are you generating?” and “How did you arrive at the valuation?” Both are seemingly good questions, except that our sales strategy starts with a free trial, which is usually tied to a typical recruitment cycle of approximately three months. The true indicator of growth in our case is the delta between converted and unconverted trials over three months. Interestingly, we are seldom asked, “How do you measure growth?”
After all, a key requirement for AI to work is data acquisition, and so, unsurprisingly, a key measure for growth for us is the number of video interviews we perform. Additionally, trials precede sales. Another question that should be asked is about our growth key performance indicators. We’ve completed 40,000 video interviews in half a year, and that’s growing at about 35% per months. With respect to repeat customers, our model is designed to be transactional, and as such, we’re creating an “always on” approach to recruitment.
The conclusion to all of this is that our journey with Searchie has been harder and longer than we expected. Listening and responding to the market has meant we’re regularly reviewing our priorities, as well as our product roadmap. At the end of the day, we continue to be driven by customer feedback, and the benefits they are experiencing from being an early adopter of technology.