This article provides a digest of the "creww academy vol.06 - How to set up a data analysis team in a startup" held at "docks" last year.
*This article is reprinted from an article published in CREW MAGAZINE on October 17, 2018.
The event started off in a friendly atmosphere, although the visitors for this event seemed to be mostly male.
Three speakers today: Kimitaka Nakazawa of AnyPay Corporation, Hisato Kawaji of Moff Corporation, and Naoyuki Tomita of Ambedkar LLC.

So what aspects of data analysis do they each utilize?
AnyPay, Inc. provides a C-to-C split-the-work service called paymo. The service requires the uploading of receipts in order to make a split, and uses data analysis to determine whether the receipt is a receipt or not. In addition, the company offers a BtoC payment service called "paymo biz" and ICO consulting services, and uses data analysis for various operations, such as semi-automation of defection models and fraudulent transaction checks.
Moff Inc. provides healthcare services using IoT devices and wearable devices. Devices are attached to legs and arms, and data is stored on how exercise improves healing, and the data is analyzed in collaboration with hospitals and research institutions.
Ambedkar LLC provides deep learning solutions for the manufacturing industry. Specifically, they are working on semi-automating the operation of garbage incinerators. They are also working with an inspection machine manufacturer to develop an AI implementation to improve accuracy.
From here, Mr. Tomita moderated the panel discussion, which proceeded in the form of a panel discussion, and I will introduce the panel members according to the topics they discussed.
What is "data analysis"?

Naoyuki Tomita ("Tomita")I would first like to ask you both, what is data analysis?"
Kimitaka Nakazawa ("Mr. Nakazawa")There are two things. The first is to optimize something of the subject matter. And while output is important, I think many managers want to know the process of 'why the output came out,' and that's the second part, visualizing the content, which I think is the best part of the analysis."
Hisato KAWAJI (hereinafter referred to as "KAWAJI")I think it is a very difficult question. Personally, I think that even though we call it data analysis, what we are doing is simply quantifying the data and making it easy for people to see and understand, nothing more and nothing less. There are many sales pitches that say, "We use data to solve problems! But in order to solve a problem, you need to make a decision, right? I think data analysis can help you do that.
rice field used to grow cropsI'd like to answer your question as well. Mr. Kawaji said that it is not very useful for decision making, but my company is selling automation. In the field of imaging, I believe that most of the parts that require human intervention can be semi-automated. I think we are now in the phase of data analysis, where we are trying to utilize data to the point where it can be semi-automated, even to the point where it can be done by people.
How "data analysis teams" were organized in each company
rice field used to grow cropsI would like to ask you, Mr. Nakazawa, how was the data analysis team created?
middle seatIn talking with Kimura, the president at the time, I could clearly understand his need to 'use data to manage KPIs in more detail' and 'run a data-driven business.' The fact that top management wanted to use data to grow the business was a big reason why."
rice field used to grow cropsOne question: what exactly are the KPIs that you want to observe, Kimura-san?"
middle seatFor example, one of the KPIs we often talk about is ROI. For example, we often talk about ROI: how much time do we spend to generate sales (or distribution volume, etc.) and how much do we spend to reduce costs? I also put an ROI on all of my projects, and I try to execute on the projects that seem to have the highest impact."
rice field used to grow cropsBy the way, can you give us any examples of analyses you've conducted that you thought were 'good'?"
middle seatI can say, for example, the algorithm for receipts. Originally, we outsourced the receipt identification, which incurred costs. Now it's become so accurate that we can do it in-house instead of using a vendor, so I guess that's an example of one aspect of lowering costs."
rice field used to grow cropsThank you very much. Now then, Mr. Kawaji. Could you tell us how the data analysis team was formed at Moff Inc."
kappa (mythical water-dwelling creature)We have two data analysis teams. We believe that satisfaction is the most important thing, and without it, the service will not last long, so these teams were formed. We are quantifying and using the data to improve satisfaction."
rice field used to grow cropsHow do you measure service satisfaction?"
kappa (mythical water-dwelling creature)Basically, it's the frequency of use.
And the other is the team that analyzes the research data. We analyze and research data on "what kind of exercise improves healing". We hear that hospitals would like to do this kind of work, but there are hurdles such as the price of the devices and the high cost to introduce and operate them, and that it is difficult to find a simple solution.
So we created a service that we could provide at a relatively low cost, and we also received a bit of consulting to help people use the service. The goal was to build on the results of the use and research to make this an effective service for health care using the Moff Band device." What are the challenges of "data analysis?"
rice field used to grow cropsWhat are some of the challenges you have faced in analyzing data? And how did you overcome them?"
middle seatThe biggest challenge in data analysis is the 'lack of data.
In my current company, I can start from the data strategy of "how to use data well," which I think is an advantage of a startup. So, for example, we can think, "We want to create a machine learning algorithm, so we need to take this kind of data.
However, I felt that at my previous job, legacy databases were used, but when the data was created, the term 'data science' was not used, and the data was stored according to common methods. When you look at it, there are a lot of cases where there are no labels (objective variables), and without labels, you can't use supervised learning, for example, so the number one difficulty is that you have to figure out other methods, whether you create the data, pull it in from outside, or unsupervised."
rice field used to grow cropsIs there anything else?"
middle seatAnother thing I struggle with every day is how to prioritize projects. If you only work on projects for one department, you lose communication with people in other departments, and you may not get projects from other departments. So I do a 1-on-1 with each department every other week. I try to keep this time to a minimum, though, because it uses up the other person's time as well."

rice field used to grow cropsThank you. Now, please, Mr. Kawaji."
kappa (mythical water-dwelling creature)One is that we have data, but it is not in a form that can be used for analysis. The other is that there is no one around me who understands the data. These two things were the most difficult for me."
rice field used to grow cropsWhat exactly does it mean to say that the data is there but not in a form that can be analyzed?"
kappa (mythical water-dwelling creature)For example, the definition of 'analyzable data' may vary, but it may be in the form of numbers or categorized to some extent. But it is not even in a form in the first place, and when you open the lid, there is nothing at all that can be handled, and the hurdle is that you have to start by cleaning this up ....... I feel like a hurdle. That took up a lot of my time and delayed what I essentially wanted to do, so I struggled."
rice field used to grow cropsWhat about the people who understand?"
Kawaji: "Regarding those who understand, it is natural to assume that they have no knowledge of data science, but I think they basically report the final figures to management, and they don't understand that the parameters are ...... and so on. It was a bit of a head-scratcher to figure out how to communicate that."
rice field used to grow cropsFor example, when you're communicating a technical term, do you simply replace it or do you explain it until they understand?"
kappa (mythical water-dwelling creature)It's the former."
rice field used to grow cropsThank you."
In my case, when I was at the consulting firm, I was dealing with national clients. For example, with a staid infrastructure company such as an electric power company, we could not release the data outside the company in the first place. So internal politics would come into play. There were many cases where data was not released even after waiting six months, ten months, or a year. So I think we always have to consider that as a risk.
Trends in data collection, analysis and utilization
rice field used to grow cropsNext, I would like to ask you about trends in data collection, analysis, and utilization. Could you tell us about current trends, or rather, how AnyPay Inc. is collecting, analyzing, and utilizing data?"
middle seatThe first is deep learning algorithms. The first is deep learning algorithms, such as algorithms for images and time series. However, while they are good at handling fixed forms of data, they may not be suitable for handling table data or other data of an undefined form. In such cases, tree-based algorithms are used. I feel that it is now coming down to these two. Most of the time it's one or the other algorithm."
rice field used to grow cropsShould these trends be understood by those in management?"
middle seat'It's case by case, but in most cases you don't need to know. People in management are often most concerned about results and outcomes, and they are not interested in which algorithm you are using. However, they are interested in why you are getting those results. I think you mentioned earlier that 'it's hard to know how to report,' and I think that's true, because you have to succinctly convey advanced mathematical concepts that you can barely understand after years of study. So I would focus on using as little math as possible and just outlining the concept to management so they understand it."
rice field used to grow cropsHow about you, Mr. Kawaji?"
kappa (mythical water-dwelling creature)Mr. Nakazawa talked about algorithms, so I would like to talk about tools. The difficult aspect in a startup is the budget. The budget for data analysis is limited, and as a result, the tools are limited. Among the cheaper ones, I think the easiest to use was tableau. tableau is very intuitive to operate, so it's very easy to use."
rice field used to grow cropsIs tableau the only tool you use in your company?
kappa (mythical water-dwelling creature)"The tool is tableau. And in terms of systems, we use a combination of AWS and Google Cloud Platform from amazon."
rice field used to grow cropsWhat tools do you use, Mr. Nakazawa?"
middle seatWe basically use open-source tools without spending any money. For example, Airbnb has an open-source tool called Airflow, which is a library that manages things like running algorithms."
rice field used to grow cropsI see. Thank you very much."
What is the mindset when building a data analysis team?

rice field used to grow cropsWhat should I be prepared for when building a data analysis team?"
kappa (mythical water-dwelling creature)"When I started the data analysis team, there wasn't a single person who understood data ......, so it's important to be strong first. It is easy to stand in a position where people complain that you don't know what you are doing, and it actually depends on how you communicate, so I thought one thing I could do is not be discouraged by that.
In addition, I am currently trying to get analysis results from the members, but I am conscious of giving one quantitative evaluation for what we have done, since we basically think on a numerical basis. I may be wrong, but I try to run it once. I basically give them recognition first, and then tell them what they can do to make it better. I make a conscious effort to never tell them that their evaluation is low or that I will lower it."
rice field used to grow cropsThank you. How about you, Mr. Nakazawa?"
middle seatWhen I set up my data analysis team, I adopted a method commonly used in the U.S.," he said. This way of starting up has become somewhat of a default.
First, top management must be in a state of agreement that they properly 'want to build a data-driven or AI-driven company. After that, hire or bring in a person responsible for data analytics from another department. From there, you have to build a cross-functional company. A cross-functional company is one where the data analysis team exists in a place that is cross-functional with other departments, such as the marketing team and CS. And one thing that is often overlooked is hiring data engineers early on."
rice field used to grow cropsWhat's the difference between you and a data scientist?"
middle seatA data scientist is basically someone who does the analysis, while a data engineer is someone who creates the infrastructure for the analysis." After hiring data engineers, we then hire experts to meet the company's needs, such as deep learning experts. This is the process we use to get up and running."
rice field used to grow cropsWhat are you prepared to do?"
middle seatThe key to success is to achieve a positive impact on sales and costs. In Japan, I think there are still fewer successful cases of huge sales achieved through the use of data science than in the U.S. or China. For example, amazon's recommendation engine is said to account for 30% of their sales.
rice field used to grow cropsThank you."

So, what kind of person should we hire as a data scientist?
In response to this question, Mr. Kawaji responded, "Since the company's performance improves when its products improve, it is a prerequisite that the person be someone who can think product-centered. On top of that, someone with a high level of commitment would be good (Mr. Kawaji).
In addition, "the candidate should be someone who likes to analyze data on a hobby level. It is also important to be able to read English, although this is often overlooked. Since this is a field in which the U.S. and China are advanced, English is often the first language in which information flows, and academic conferences are basically in English. I think it is important as a basic background as a data scientist (Nakazawa).
It was an intense event where the participants were able to discuss not only what they need to be aware of in order to set up a data analysis team, but also the difficulties they face in data analysis and how to overcome them.
INNOVATIVE PORT Editors INNOVATIVE PORT" is an open innovation media operated by Creww, Inc. that connects startups, entrepreneurs, entrepreneurs in recovery, and companies aiming to create new businesses on the theme of social issues.






