Materials and Chemicals Informatics (MI) has gained traction over the past several years as both academic groups and commercial companies have demonstrated success. In the early days of Citrine, many of our prospective customers considered whether they should use AI in product development. Now, the majority consider how to apply AI. Some companies run pilot projects themselves, see the scientific & business value, and are now deciding whether to build an informatics platform themselves or work with Citrine to roll out “AI at scale” across their global operations. This post lays out the key considerations for the “build vs. buy” decision.
Time and Resources
Building an enterprise-ready MI platform from scratch will take 1-2 years. It will depend on how much resource you put into the project, but we estimate it takes 6-12 months to get a first proof of value deployment up and running. (In contrast, first proof of value is usually delivered 4-6 weeks after buying the Citrine Platform.) But that first system will not be an enterprise-ready platform that can support a global roll-out. We estimate that will take at least another year of full-time work.
Resourcing Requirements to Build an MI Platform
Recommended resources to build a minimum viable product (MVP) MI platform in-house
- Executive sponsor (.25 FTE for 12 months)
- Project/product manager (1 FTE for 2 years)
- Data architect (1 FTE for 2 years)
- Data scientist (1 FTE 2 years)
- Data engineer (1 FTE 2 years)
- Full stack developer ( 1 FTE, 2 years)
- Infrastructure/cloud engineer (1 FTE, 2 years)
- DevOps engineer (1 FTE, 2 years)
- Materials scientist (.25 FTE, 2 years)
These estimates include a V1 platform build including structured data management, the ability to predict properties of known materials, and cursory design of new materials. This does not include ongoing resources such as maintenance, scalability, or upgrades.
The impact of delay
Let’s assume you understand how Materials and Chemicals Informatics can provide value to your business by increasing agility, improving efficiency, and helping your team focus on projects likely to succeed. And also, let’s assume you know rival companies in your market vertical are also exploring MI. Then it follows that delaying the full realization of the benefits of MI by 1-2 years will significantly affect the bottom line.
THE DIFFERENCE BETWEEN A PILOT PROJECT AND AN ENTERPRISE READY PLATFORM
There are plenty of open-source or low cost ML libraries, databases, and data pipeline tools that can be connected to make a prototype MI system, deploy your first AI Model, and complete a proof-of-concept or pilot project. However, we’ve seen customers run into trouble when they attempt to apply this to additional projects.
An Enterprise-ready software MI Platform needs:
Good User Experience
If you want your whole team to be able to leverage the platform and input their domain knowledge, then you need an easy-to-use user interface. The spaghetti of open-source tools should be hidden under a well-designed, intuitive UI.
Security, Availability and Access Control
It is vital security is built in from the ground up to prevent exposure of sensitive data, and therefore loss of competitive advantage. Automating back-ups, monitoring performance, and preventing system downtime are equally important. As users increase, your platform needs to scale, with a code structure that can grab compute power as it needs it. As the platform is rolled out internationally across business units and used with subsidiaries and joint venture partners, it becomes even more important access control is suitable to ensure employees and partner companies can access only what they need to, and any data covered by ITAR regulations are kept on US servers.
Part of the value gained from a Materials Informatics platform is dependent on the reuse of digital assets (data sets, search spaces and AI models). Methods therefore need to be in place for researchers to find and plug these assets into their projects. Deploying AI models needs to be an automated process, requiring little, if any, code to run.
Finally, going forward, you will need a team of software engineers to not only add new features, but also to maintain the database and code. Changing data schemas as needed and refactoring the code base as new software languages become the fashion, extends the longevity of the platform. To do this though, your team will need systems for version control, regression testing and continuous integration.
RISK OF FAILURE
The risk of delay has been covered above, but what about the risk of failure? Setting up a secure data management platform with proper access controls and well thought through ISO 27001 standard security protocols is not easy, and if you get it wrong, your valuable IP will become public, decimating your competitive advantage. You might also get to a point where you realize after lots of sunk costs, your home-spun system just isn’t performing as well as you’d hoped.
THE BENEFIT OF PEOPLE DOING WHAT THEY ARE GOOD AT
Many management books suggest companies should stick to what they are good at, their core competencies. This doesn’t mean that a company is not full of smart people, who can do anything if they set their minds to it. It is just others can do it quicker and cheaper because they are already set up to do it. Citrine Informatics, established in 2013, has a team of 100+ employees that have created the world’s leading Materials and Chemicals Informatics platform. And as we move forward, we will learn from all our customers and continue to develop the Citrine Platform, adding new features you don’t know you need yet. Focus your team on what they are good at, materials and chemicals development. Let Citrine support you.
Key Questions to consider
- What impact would it have on your business for a competitor to adopt an AI solution faster than you?
- Is the cost of licensing from a 3rd party worth the delayed deployment timeline and internal resource commitment to build your own solution?
- Do you have the team in place to commit to platform development, maintenance, technical support, and scalability?
- How will you maintain the security of an internal system across teams and across geographies?
- Can your team efficiently build and deploy machine learning models across R&D projects and Business units?
Find Out More
You can request a demo of the Citrine platform here.