topic – the step-by-step process IBM’s data science team applies when working on

a consulting project. We believe this overview can be highly beneficial for both experienced

professionals and data science beginners. We’ll explore a best-practice framework

applied by one of the pioneer and leading companies in the field. This way, you’ll

get an insider’s look at how a consulting project that involves data analysis and data

science unfolds. In addition, we’ll examine the results achieved

in IBM’s data science consulting projects with major clients from different industries.

Why is that important? Well, each of these initiatives serves as an invaluable lesson

to the rest of the companies in the respective industry. If, for example, Carrefour managed

to leverage AI to improve its supply chain processes, the rest of the global hypermarket

chains would basically be obliged to follow, if they want to keep up.

Alright. Let’s get right in and outline the five

stages of a data science consulting project. Stage one - engage the firm’s CTO.

Stage two - meet with the company’s SMEs and brainstorm.

Three – Data collection and modeling through coding sprints;

Four - Visualization and communication of findings;

And finally - Follow-up projects; Each of these steps of the process is vital,

so let me elaborate a bit further by describing them one by one in more detail.

Things start with a conversation with the firm’s Chief Technology Officer.

He needs to be sold on the project. Hopefully, this would result in him championing and endorsing

the initiative across the organization. Such buy-in enables cooperation and improves the

project’s chances of success. At this stage, the consulting team and the CTO will define

the scope of work and the ‘lowest hanging fruits’, which will give an immediate boost

in terms of bottom-line results. What we mean by ‘lowest hanging fruit’ is an opportunity

that the data science team knows is available for most companies in an industry and is easiest

to implement. For example, they have seen on a few occasions that supermarket chains

can greatly reduce food waste if they implement a predictive AI model able to adjust the timing

of deliveries. So, an absolute best practice when working on consulting projects is to

address such opportunities first, because this gives instant credibility to the project

team and wins support across the organization. Once the project scope has been identified

with the firm’s CTO, the data science consulting team will proceed to brainstorm on how AI

can be applied in the particular use cases that have been pre-selected.

To envision this a bit better, the team needs to conduct a series of interviews and meetings

with Subject Matter Experts - the people who work in the business day in and day out and

who are able to contribute greatly in terms of identifying actionable and meaningful solutions.

Also, in most cases, SMEs are the ones who have a good idea of what data is available

and can be used for the purposes of the project at hand.

The next stage consists of coding sprints. This is the main chunk of the work, so IBM’s

team organizes it in three parts. One for Collecting data and feature modeling

Data collection sounds like ‘getting the data from all places’, but it may be much

trickier. Depending on the scope of the project, the consulting company may need to first consolidate

all data in one place, called ‘a data warehouse’. In some cases, not enough data is being collected

and new data sources must be set up. Feature modeling is inside this step as features may

be chosen from the available data. Sometimes, however, very important metrics are not being

measured. The consulting firm can then suggest starting to collect data on that, thus changing

the data collection structure of the client. Another sprint for feature selection and running

the model for the first time Once data has been collected and features

have been modeled, it is time for some data science.

While features were modeled and kind of selected during the first sprint, they were never tested

in a model. So, in the second coding sprint, features are evaluated, transformed, or new

features are engineered, this time for predictive modeling purposes. Once this is done, the

first models come to life, showing the potential to the stakeholders in the client company.

And a third sprint to fine-tune the model and adjust it as per client requirements

The moment a solid model has been thought through and executed, the fine-tuning begins.

There are many ways in which a model can be improved. A 1% increase in accuracy could

imply millions of dollars in savings for the client company. Therefore, this step should

not be overlooked even if it sounds like the least exciting one.

Okay. Moving on to the fourth stage -data visualization.

Data visualization plays a critical role in most data science projects. However, please

bear in mind that the specialists who build a model are not always the ones best equipped

to work on the visualization of its findings. When presenting in front of a non-technical

business team it is much better to show Tableau or Power BI graphs rather than a Jupyter notebook.

And hence, the data science consulting team needs skills related to chart and dashboard

creation, as well as the ability to communicate in an effective way. It is not uncommon to

have a person whose job is to solely style such findings, giving the final touch to the

presentation. And this is how we reach the fifth stage,

namely, Follow-up projects As with any other type of consulting, the

secret sauce of being a successful consultant is to be able to sell the next project. And

then to sell the next one after that. And so on.

The premise is that if the consulted company sees a measurable bottom-line improvement,

they will certainly want to retain the consulting team and will be willing to purchase additional

services - from IBM in our example. This is also why consulting firms prefer to start

with low hanging fruits – this allows them to show they can create value very fast. And

hence they improve their chances of being hired again.

Alright. Now that we’ve figured out the typical cycle

of a data science consulting project, let’s take a look at some of the successful use

cases IBM’s elite data science consulting team helped with.

Starting with… Nedbank. In the case of Nedbank, a South African bank,

a model predicting ATMs’ need for repair was implemented and this led to important

efficiencies in terms of ATM reliability and maintenance timeliness.

In another project, IBM’s data science team helped JP Morgan implement a model, which

prevented the bank’s traders from engaging with trades that are not recommended by JP

Morgan’s powerful predictive models. Experian is one of the leading companies in

the information business industry. They analyze credit payments on a global scale for a number

of institutions. In this case, IBM’s team helped Experian leverage unstructured data

and combine it with structured data (that was traditionally used in Experian’s models)

to build a more comprehensive view of the businesses Experian is hired to analyze.

One can argue that data science and AI consulting is a business in its infancy. And it appears

that the most important ingredient, IBM’s team has mastered, is the combination of technical

know-how in terms of data science modeling and business understanding.

Truth is, a successful data science project needs both. This is precisely why we try to

teach you how data science can be applied in a business context in every course of the

365 Data Science program. So, if you’d like to explore this further or enroll using a

20% discount, there’s a link in the description you can check out.

We hope you found this video helpful. If you enjoyed the topic, don’t forget to press

the like button and subscribe to our channel here on YouTube. In the upcoming months, we

will prepare tons of other useful career-oriented data science videos you don’t want to miss

on. Thanks for watching!