5 Main Phases to Build/Transform Into an AI-based Company — Google Method

Soen Surya Soenaryo
11 min readJun 1, 2018

--

Artificial Intelligence, or we could abbreviate it into A.I., is a term that we often hear it nowadays in our society. In the past, we usually heard the term of A.I. in games, when we want to play alone and need to have “some partners” to play together. Basically, A.I. is not newly founded technology, it has appeared since the 1900s. However, because of several factors, especially because of our current computational capabilities, several industries besides gaming industry are getting interested in the capability of A.I.

Interesting Fact: one of the reasons why we could unleash the power of A.I. is because of the capabilities of modern GPUs, and one of the reasons why it can reach the current computational power is because of the demand from the gaming industries. So, if you are a gamer, you should be proud because you have contributed to the development of A.I. :)

Google, as we know, is one of the pioneers who have started to implement A.I.. They do not only integrate it in all of their products but also in their business process cycle. From their perspective, AI is being used as one of the ways to scale, to automate, and to personalize by using the right data. Now, they see the problem in different perspective and conduct their business process in a unique way. If you watched the recent annual conference that was held by Google, “I/O 2018”, you could see that they are able to successfully escalate their progress and lead the race in A.I. between other big competitors, such as Amazon, Microsoft, or Facebook.

Figure 1: General Feedback Loop.

Typically, when we are talking about an A.I.-based company, that company is working in the technology industry. The thing is, every company in every industry has the capability to become an A.I.-powered company. In order to build or transform your company, it is harder than just applying A.I. or Machine Learning systems on top of their existing business system. Typically, at least, it would be hard to keep iterating, especially for an old-style business process, who use abstraction and intuition in many sectors to iterate their general feedback loop like in Figure 1.

The good thing is, Google has shared their 5 major phases that they use and believe in order to transform their company into an A.I.-powered company. In order to reach the same goal, they suggest tackling these phases from the beginning.

#1 Individual Contributor

We usually meet a lot of individual contributors in our society. An attendant in the hotel, for example, has a lot of abstract tasks, from carrying a luggage, leading someone to a specific area, providing some information for the customer, and so on, whereas their main objective is to give the best service for their customer. An individual contributor can be thought of as a set of abstract tasks that could be performed by a single person. (We could also call it as informal tasks.) In order to move toward the next step, we need to identify the individual contributor in each sector of your business process first.

If we do not try to identify the individual contributor in your company, it would be hard for your company to scale up and move toward implementing A.I.. The reason is that there is a lot of abstraction in your business processes. This kind of abstraction makes tasks hard to be conducted in parallel. Furthermore, it is hard to change any incorrect assumptions that are made by decision makers. An assumption is not a bad thing, however, the activity that you conduct using that assumption would be hard to be scaled and validated. Minimizing the use of assumptions will help in reducing the abstraction of tasks.

However, if we were stuck in this phase, it will also hurt your company. Have you ever heard a story about a company who lose their ability to compete because they loss their key employee? Companies like that have a hard time scaling because of the dependence on their key employee.

#2 Delegation

After you identify the individual contributor in your company, it’s time to formalize the blueprint of that task in order to conduct that task in parallel. You could notice whether or not your company has reached this step if the particular business process becomes more and more important for the contribution of your company, and you can delegate multiple people to perform this task in parallel.

This second phase is also an important one. If you are not being forced to formalize the process, and by sticking to the abstraction in the individual contributor, you would be hard to identify the general idea of how that particular business process works. This decision also impacts the scalability of your company because that particular business could not be conducted in parallel — no scalability on here.

This phase is quite important for the further steps. If you skip this step, you will miss the opportunity to hear feedback from multiple perspectives. We could use this process as a testbed — a great product learning opportunity.

Furthermore, this delegation phase is a crucial one, especially in preparation for the Machine Learning phase. A great machine learning model will generate a lot of output that needs to be reviewed by human. It can be a good indicator of a sub par model if the output of the model is confusing to humans during the review process. This activity could be useful to measure the “false-positive” output in the performance metric by conducting random sub-sampling.

The problem is, if we are lingering too long in this phase, we need to pay a high marginal cost to serve each user. Generally, the more abstract the task is, the more important that task in the company and the pricier it will become. The longer a business is in this phase the more likely it will be that people will doubt that automation is possible. The common term for this situation is organizational lock-in.

“Use Machine Learning as a way to expand/to scale the impact of our people, not as a way of completely removing them. That’s a very high bar for a Machine Learning system to meet.”

#3 Digitization

After you start to make a transition from doing a task in an abstract way into a more parallelable way, another important thing to do is to digitalize that process. The thing is, in order to move up into the next process, we need that information — that data about how the task was conducted. The data on here is the main resource for the later stages.

By digitizing the business process, it does not only useful for the further process, but it also becomes one of the resources to enable scalability in the business. Even an IT project that was successfully tied with Machine Learning system would fail if it required some handwritten input or gap that caused a bottleneck preventing the full potential of automation.

On the other hand, if you get stuck on this step, by merely collecting the data, you could have more organizational lock-in. Nevertheless, the competitors are collecting data and tuning their system from this insights. The race in the technology field is quite fast, even in the non-technical industries. For instance, the integration of Deep Learning in Google projects from 2012–2015 is really quick like in Figure 2.

Figure 2: The Growing Use of Deep Learning at Google (Source: https://www.analyticsvidhya.com/blog/2017/04/comparison-between-deep-learning-machine-learning/)

#4 Big Data and Analytics

Here’s a simple general question: How many times have you drunk soft-drinks last month? Do you know the difference between your exercise plan A and plan B? Are you sure that there are no other things that affect your workout result, such as your motivation, intake nutrition, sleep duration, and so on? You track it in a log, but do you get some intuition from your log?

Big data and analytics are the major drivers behind a company’s ability to apply a data-driven decision. We start to use the data that we have collected and use it to create an insight. This insight could help a company to measure and achieve data-driven success about the internal operation and external user. We need data as the resource to produce insight and measure the state of company business.

Big data and Analytics phase could help your company to make it easier to review, summarize, and deep dive every aspect of their business process. Toyota Lean Manufacturing Philosophy, for instance, is a philosophy that was applied by Toyota to help them tune each little knob in their business processes in order to get a better outcome and faster car.

“It’s a great opportunity to pause and re-iterate the definition of success!”

Be careful, however, the data that you have collected before are likely messy and not in a form useful for machine learning. Data from different resources can have different structures, and if we do not process these data into a clean one, we could not use it as the input of Machine Learning system.

However, if your company stuck in this phase, this situation would limit the complexity of problems that your company can solve. Yes, the business could iterate normally and get some benefit from big data insight. However, in order to achieve a higher degree of business, you need to tackle more complexity of problems in the business process. Machine Learning is one of the possibilities to answer this issue.

Without Machine Learning, this condition will limit the speed at which your company can solve the problem. We could iterate the business by having several experts in a particular sector of the business process. Nevertheless, the speed of that process mostly depends on the expert themselves.

#5 Machine Learning

This is the phase that we want to reach, in order to turn into an AI-powered company, where the company compiles everything from the previous steps, and automatically refine it to improve the business processes’ performance. Refer back to the picture of General Feedback Loop above, the idea of this phase is to automate that iteration. This automated feedback loop can outpace human scale between measuring success and tune the business process. Conventionally, Machine Learning system could outpace the human ability to handle the number of inputs and corner cases in the real-world data.

“Machine Learning could improve 10% on top of all the human-hand tuning process.”

Machine Learning could help and solve the limitation of human recognition in analyzing, reviewing, and solving the business problem because we could get the faster answer and more nuanced treatment of details. We could describe the capability of it likes an artificial brain that learning from billions of interactions every day.

The Transformation in Business Perspective

In general, by digitizing our business processes, we could get a lot of data, which could be used to create a very deep market insight/user insight / operational insight. Machine Learning is one of the new methods to choose the operational parameters. For instance, by using a simple regression, we could regress the performance of each operational performance in a period of time by using a line or more complicated curve. It helps us to extract the cases we might not have seen it before by using data-backed validation.

On the other hand, if we skip any of the 5 main phases, there are some drawbacks, even if we could generate a better operational parameter by immediately implement Machine Learning in the company. The main problem is about how to fit back these parameters into the cycle and execution of core business process.

If we are still in the first and second stage, where there are a lot of individual contributor and abstractions in the business processes, we need to conduct a training and counseling for everybody in the company, especially for the adoption of Machine Learning and integrate it in their tasks. Furthermore, by skipping the delegation phase, we could not measure the correlation between the Machine Learning system that we create and the tasks. On the other hand, because we still include human in this complex process, we need to create a manual reference, which human can use to refer back to as needed. The problem is, it will become a bottleneck because the process of updating the manual reference will not be as fast as the evolution of the Machine Learning system.

If we have moved into the third and fourth stage, we could have a new operating parameter with a better standard and validation. In this process, every circle in the business process could make most of their decision from the insight that was drawn from the data and focus more on the decision step. For instance, in the customer relationship circle, every employee could only need to use a headphone and provide a precise information to the customer in the meanwhile they also collect the feedback from the customer. In this process, our employee only needs to focus on interacting with customers more quickly and providing more accurate information.

However, if we move into the last stage, we could automatically iterate the general feedback loop of a particular business process with less, and perhaps no, involvement by humans. The idea of Machine Learning in an A.I.-based company is to make sure that machine learning has been involved in every cycle of business processes or products on that company, which we could give a better service from segmentation into an individual person.

I was inspired to read this article from my experience when I was learning one of the courses from new Machine Learning Specialization by Google, “Machine Learning with TensorFlow on Google Cloud Platform”. I really encourage you to take this course if you are interested to have a good fundamental knowledge of using Machine Learning services that was provided by Google, either Tensorflow or ML API in Google Cloud Platform.

Surya is a Back-end Engineer in his professional life and a Technology Evangelist in his personal life. This year, he builds a resolution, called “Road to Becoming an Applied AI Scientist”. If you are interested, or want to follow his progress, you can visit him at his personal website, soensurya.com, or on LinkedIn.

--

--

Soen Surya Soenaryo

Software Engineer @Xendit, Organizer @Kaggle Days Meetup Jakarta | soensurya.com