The business world is beginning to adopt AI. Research by McKinsey shows that 47% of organizations have implemented it in at least one function in their business processes, compared to 20% in the previous year. Another 30% of respondents said they are piloting AI. The primary sectors that have adopted it so far include advertising, marketing, and media & entertainment companies. And the next sectors are just around the corner.
The popularity of AI is growing exponentially, and so is the demand for data science specialists. According to a report by LinkedIn from August 2018, there was a shortage of over 150,000 people with data science skills in the U.S alone, and according to IBM, this demand is supposed to grow by almost 30% in 2020. No wonder the Harvard Business Review called the Data Scientist role the “sexiest job of the 21st century”!
It’s clear this creates a major talent challenge in the labour market, making it even harder to find the right people with the right set of skills. Building a data science team in an insufficient talent pool is difficult and expensive, and as such - only affordable for big players in the market. And let’s face it: there aren’t many companies that can compete with Facebook or Google.
Added to this talent challenge is the risky nature of sourcing AI specialists to push through adoption. What if the market hypothesis turns out to be wrong? What if the new solutions don’t bring enough profit to cover the initial expenses? How long will the process be and how fast can you expect to see the return of investment? These questions are arising and there is no simple answer to any of them.
It’s not surprising that companies that want to adopt AI often seek the help of external agencies specializing in this kind of service in order to minimize risk. In this blog we look at the challenges of building a data science team and how you can overcome them to increase the success of outsourced AI projects.
Stage 1: Preparation
The foundations of AI project success lie at the very beginning of the project and are typically built in the planning phase. At this stage, it is important to get ready for AI adoption, plan the process carefully, and eliminate as many potential risks as possible.
But just how do you do this?
Based on the assumption you have your data ready (even the best AI development team can’t do much if they don’t have the required data!) and your team is prepared for the process, let’s go through some of dos and don’ts of working with an outsourced Artificial Intelligence development team:
...and with the lowest cost possible. Implementing a company-wide AI strategy takes time, effort, and money. There is no point building a complex product from the very beginning, without proving the efficiency of the models first! Instead, choose one segment to test AI in the business and start with a proof-of-concept to validate the idea.
Your outsourced Artificial Intelligence development team should be able to advise you on that area and help to choose the right segment for the experiment. This approach presents a much smaller risk of failure as it leaves room for improvement in each iteration. At the same time, you can quickly see what AI can bring.
Plan goals of the project
Know what you want to achieve with AI and what your success and failure criteria are. Although with that in mind, saying “I want to reduce churn” is not specific enough and without narrowing it down to numbers, it will be hard to tell if the achieved results (reducing churn by 3%, let’s say) are satisfactory or not.
Again, a good outsourced AI development team should be able to help you plan these goals: both the main objective and the little wins, and monitor the progress towards achieving them throughout the whole development process.
Don’t try to skip the workshop phase!
2 days of workshops before the actual development can save you months of work (not to mention a lot of money)! Save your calendar and make the most of that time. Share your domain expertise, tell your outsourced AI development team as much as possible about the process that you are trying to optimize with AI, its strong and weak points, and all the pains related to it. There’s an important thing to note here: if the outsourced team doesn’t express a need to gain a deep understanding of your business and domain expertise, it’s not a good team to begin with. You’re not in for the tech, you’re in to grow your business.
During workshops, you should be able to discuss your long-term goal, processes that can be improved with AI, the data you have and the data to be predicted. The goal is to select predictive models that can give you “quick wins”, ensuring the areas with the most value potential are prioritised for improvement with AI.
Stage 2: Development
Once everything is planned, the strategy is created, the data is delivered, and the team is ready to start, it’s time for the actual development. It’s the time when your outsourced artificial intelligence development team builds a proof-of-concept that helps you validate all the early assumptions regarding implementing AI into your processes.
This part usually consists of a few steps. The development team needs to clear the data (if necessary - and it usually is) and combine multiple data sources into one vector, build a model as close to the established parameters as possible, publish the model and prepare instructions that tell you how to use the model in your everyday work, and how to interpret the results. It may also be necessary to create integrations, such as weekly emails with predictions sent to a given group of people or the preparation of a file with predictions included.
Share your know-how
No matter how good your outsourced Artificial Intelligence development team is, there is something they will always lack: the domain expertise. Luckily, they can learn it - to the necessary extent - from you and your team! Do your best to answer all the questions that may arise. Share everything they need to know about how you currently solve the problem that they’re aiming to solve with AI, and tell them what they need to keep in mind at different stages of the process (e.g. sales process if they aim to increase sales). Your contribution is invaluable!
Don’t hesitate to ask questions and share your concerns
Not sure if the project is going in the right direction? Share your concerns!
Also, mind that just like the development team needs to understand your problem and goals in order to deliver an effective solution, you need to understand their solution in order to use it effectively. If AI is expected to gain your and your internal team’s trust, you need to be able to understand the decisions that AI makes and explain them to the people that are going to follow these decisions. Making AI interpretable through communicating the results in a clear and understandable way can foster trust and make the insights produced by the models more actionable.
One of the ideas standing behind the “start small” approach - besides the ability to verify the assumptions as fast and with as low a cost as possible - is that working that way, you can easily iterate. Working in Scrum, you can see the results of the development with every sprint. Moreover, working on a solution may bring additional new ideas to the table. Make sure to attend the demos, watch the progress, and discuss the ideas that will arise.
Stage 3: The Finish Line
At the end of the development phase, it’s usually time to present the results and think about the next steps. In this stage, you should get some recommendations regarding subsequent models and implementation of AI in your company. It may be recommended that your company should start collecting data or guide you on what data should be purchased to improve the models, what models should be created next when you already have more data, and how to combine the models to get even better results.
Patience is a virtue!
Keeping the goal that you set at the beginning in mind, think about how close you are to achieving it. Starting small, validating your assumptions in some small part of your business with a proof-of-concept, you can expect fast results - compared to building a complex product. But even with that little part of your process that you decide to optimize, the models require some time to process the data, learn, and start bringing the desired results. It is possible that the first results will not be fully satisfying. Their accuracy will improve over time and with every iteration, but time is the key.
If you’re happy with the results - and AI has proved to be worth implementing in your business, then you need to take the evidence and present to convince your board or investors to increase the innovation budget for the next steps of implementing AI. It is at this stage, you need to think about how to scale it in the most efficient way. Should you feed the models with more data regarding more products or users from other locations? Once you know what can actually be achieved with AI models, you can test more assumptions.
Is It All Worth It?
The big asset of artificial intelligence is that by improving different processes, it often contributes to an enhanced customer experience. The clients experience better customer service thanks to intelligent chatbots available 24/7, better product recommendations based on recommender systems analyzing historical data, higher efficiency of different processes (e.g. faster delivery thanks to better product demand prognosis) and better personalization.
New solutions, especially when they improve customer experience, quickly become the new standard expected by clients. More and more companies are now in a difficult position. Either they will adopt AI, or they will fall behind their competitors who are already embracing it.
Editor's note: This is a guest post from Neoteric. Neoteric is a software house specializing in intelligent systems around predictive analytics and recommendation engines. The solutions they develop help companies gain a better understanding of their customers, reduce and prevent churn, benefit from dynamic pricing, and discover best learning opportunities for their employees.