Data & B2B : How to launch a high impact ML project when your core product is not AI centric ?
TL;DR : In order to prevent your Data team from only handling support tasks and to enable them to develop high value ML* projects, you will need to build a certain environment where research and new ideas are thriving. As a rocket needs a well-defined trajectory, an advanced technology and a fuel, you will need at the very least: business insights, R&D approach and a source of data.
The use of the machine learning can be quite challenging these days. Some are claiming the death of data science as a non valuable asset for companies and it is becoming hard to clearly foresee the interest of such a field in a business (Cosley, 2021).
Two types of decision have been made by companies.
- Either they go crazy to the new AI world and try to develop new scientific breakthroughs, hoping that at some points they would become profitable.
- Either they use their data to only gain some basic analytic insights.
A lack of data, the high cost of an R&D team, and the growing development of open source tools all make the decision even harder to consider for non-data-centric companies.
At Ovrsea, we think that we can become a key player in our business by empowering AI at the right level of R&D to solve concrete business cases. Through this article, we want to reconcile and moderate these two visions by providing a way to develop high potential ML projects even in an environment that is not necessarily suited for it. With that in mind, we will take B2B companies as a characteristic example of a non data-centric company.
*In the whole article, we use the ML acronym to designate the concept of machine learning.
Issues encounter with B2B
You might think saying B2B is not a data-centric business seem silly… But if you take a closer look, you will see that the most important fact about B2B is that both the number and the type of customers change. While in B2C you would face a human being responding to a law of large numbers’ logic, you will face a much different perspective when clients happen to be companies. Their number may be less, their power of negotiation increases and finally client account managers may be needed to respond properly and individually to each one of them (Georgiev et al., 2021).
This is often when small B2B companies stop thinking in terms of data and start working for their customers with a human professional approach. In this kind of business your client may never face an AI result directly. For example, putting in place A/B testing on your web app won’t be suitable. The number of users may not be large enough or their responses may not be relevant to develop the best long term company vision.
Thus all this environment will prevent you from creating a data driven strategy. To avoid coming back to the HIPPO — Highest Paid Person Opinion — strategy, some companies will still use analytic data to gain insights on their business. While being a commendable approach, this may come down to use only a fragment of the data power.
So, how can you trigger and develop a high-impact ML project?
Project launch
In order to prevent your Data team from only handling support tasks for your B2B business and to enable them to develop high value ML projects, you will need to build a certain environment where research and new ideas are thriving. As a rocket needs a well-defined trajectory, an advanced technology and a fuel, you will need at the very least: business insights, R&D approach and a source of data.
1. Business insights
In a B2B business, a Data team can’t decide on its own where they want to go. Short term as well as long term business vision need to be incorporated to the project to consider the possible revenue generated by it. This is often achieved by structured and formal interactions with other teams to consider their problems or ambitions (Ononiwu Frankline, 2021).
However, increasing informal communication can be a good way to overcome the problem of understanding the work between each team. If people don’t know exactly what you are capable of, they might never reach out to you and the company could miss a very valuable idea.
Moreover, the need of a project and the accuracy of its potential outcome inescapably need to be tested by the final users once deployed. In B2B this is even more true, as your Data team might not be able to sufficiently understand the customer needs on their own. Will your service be deployed for only some customers, why, for which need, and which ROI? All these answers must be addressed hand in hand with the business teams.
So be sure that you Data team is connected to concrete world’s challenges!
2. R&D approach
When it comes to bringing accuracy as well as new algorithms or external software, you need to have a Data team that is listening and monitoring new technologies. And this is not simply about accuracy or mathematical discovery, this is about qualifying possible external products or ML software, that may boost your ROI (Ransbotham et al., 2020).
And here is the key to succeed: define the right level of research and development.
In AI, the real value comes from knowing what you can do better than others, and accepting to deploy solutions developed by teams better than you on your targeted topic. AI research requires a lot of time and money and being at Alphabet, Amazon or Meta’s level does not need to be a goal. Your precious added value will be to challenge the solutions and adapt them efficiently to your business case by understanding them and use them at their most.
3. Sourcing data
Let’s talk now about your fuel here: data.
B2B approach to data is specific. Internal data might often seem limited, but the first step is always to see and consider all the data you have at your disposal. Data mining of your own company might seem useless, but it defines a strong part of your own competitive advantage (Jasmin, 2021). You might gain a lot to crop it well, clean it and monitor it to be able to use it!
However, sourcing external data to enrich yours could result in an amazing growth factor. Sometimes, you need to seek for data, either free or not.
It is always hard to consider buying external dataset, as your competitor could do exactly the same. But contrary to your own data, this means that this time your added value will reside in processing and using the data in an innovative way. Data doesn’t need to be client focus directly it could be anything that brings added value and at this stage being open minded is the key.
So try to test your idea on a small sample and consider step by the step the potential outcome that might deliver value to your customer. It will prevent you from buying too many sets of data at each trial and still enable you to test and fail … or succeed!
Use case : Price prediction at Ovrsea
At Ovrsea, many ML projects have been impactful, so let’s illustrate our approach by one concrete use case: the price prediction algorithm. As a freight forwarder, at Ovrsea we quote shipments for our customers. This task is very personalized and complex to put in place, however a quick and reliable answer is needed for the customer to be able to plan its supply chain. To do this, the Data and Tech teams have developed over the last 3 years an algorithm that enables the prediction of the price when it is not given by freight companies. We named it Kronos!
As stated above, this project is combining the strengths of the Data team to create a powerful tool.
- We have considered the business need. Here, by lowering the quote answer time and the price risk for our customers, we are increasing our quality of service and enabling our customers to have a reliable and quick answer, which is an interesting step forward.
- Once we had the business part, we needed to consider the ML algorithm that we were going to implement. Do we need to build one or to use cutting edge technologies for the pleasure of buzzword? Well here a traditional and yet efficient algorithm was sufficient enough to deliver value: XGBoost.
- Finally we used both types of data. Internal recent data could be useful. At the same time we completed it by connecting our stack to price APIs or Databases of supplier companies.
This project has been made possible by the efficient interactions and collaborations between the Product, the Pricing, the Tech & Data teams. This prediction software is a powerful tool that is now an essential part of Ovrsea’s value proposition, while helping on 30% of all our quotations!
Conclusion
At Ovrsea we cherish the balance between AI and business needs that provides unbelievably high value projects. Even in the B2B freight forwarder business, we have been able to implement a Data team dynamic that enables us to develop high potential ML projects with significative business impact. While monitoring new ML technologies, we still primarily focus on the concrete help that an ML project is bringing to our internal users and our customers, and use all relevant sources of data to achieve it.
This is how we are revolutionizing the freight forwarding experience! If you are craving for AI and want to be part of this terrific journey, do reach out!
Bibliography
Cosley, B. (2021) Is Data Science Truly Dead?. From Data Scientist to Ai Practitioner, Towards Data Science. Available at: https://towardsdatascience.com/is-data-science-truly-dead-b11cd76eafc8 (Accessed: March 29, 2022).
Georgiev, S. et al. (2021) The data gambit: How large B2B companies can outmaneuver start-ups, McKinsey. Available at: https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/the-data-gambit-how-large-companies-can-outmaneuver-startups (Accessed: March 30, 2022).
Jasmin (2021) What are the advantages of machine learning projects for businesses?, Nextlystics. Available at: https://www.nextlytics.com/blog/advantages-of-machine-learning-projects (Accessed: March 30, 2022).
Ononiwu Frankline (2021) How to scope a Machine Learning project, Medium. Available at: https://medium.com/codex/how-to-scope-a-machine-learning-project-d74d4025e04c (Accessed: March 30, 2022).
Ransbotham, S. et al. (2020) Are You Making the Most of Your Relationship with AI?, BCG. Available at: https://www.bcg.com/en-in/publications/2020/is-your-company-embracing-full-potential-of-artificial-intelligence (Accessed: March 30, 2022).
