With the increased adoption of cloud computing and demand for managed applications and services, we are now able to offer a new choice of Software as a service (SaaS) deployment model for customers who need it. Let’s explore the organizational components that influence whether SaaS is the right choice for your organization when it comes to data science and MLOps.
To SaaS or not to SaaS
With SaaS, the software provider manages the application, including security, availability, and performance. Similarly, an MLOps as a Service platform streamlines the management, development, deployment, and monitoring of data science models at scale for enterprises via a fully managed and hosted cloud service.
In today’s cloud era, every organization, small or large, has turned to Software as a service (SaaS) offerings by software companies they trust for many of their business applications. From customer relationship management, communication and collaboration, project management, human resources, productivity applications to data and analytics, SaaS applications are now prevalent around us. Agility, convenience, simplicity, scalability and security are the key drivers behind the adoption or migration to SaaS applications, whereas, full control and high customizability are the top reasons organizations stick to self-hosted and self-managed deployments.
When it comes to cloud computing, every organization has its own unique strategy, pace, culture, competencies, priorities and plans that influence whether SaaS is the right choice. Using this article, carefully consider where does your organization stand in relation to these criteria and factors …
1) Cloud Strategy and Adoption Maturity
Both data science and technology leaders must assess where their organization stands in terms of cloud adoption maturity, related strategic direction from executive leadership, and their personal role in influencing the company’s cloud strategy. This is important because, the farther along you are in adopting cloud tools for a range of business needs, the better you will be positioned to succeed with a SaaS MLOps strategy.
There are a number of key considerations and indicators for this assessment. For instance:
- Is your organization primarily experimenting or actively migrating or optimizing applications and services in the public cloud?
- Has your organization already moved beyond the usual SaaS applications/platforms like Zoom, Salesforce, Atlassian, Slack, Microsoft 365 or Google Workspace?
- Are you running your data and analytics platforms in the cloud, with SaaS services like Snowflake, Amazon Redshift, Google BigQuery, Tableau Online, Looker, Domo etc?
- How long has your organization been in business and acquiring IT assets? The longer your company has been in existence, the more likely it is you’ll have a number of legacy applications and data stores that are still on-premises, adding to the friction of data transfer to the cloud.
Ultimately, organizations that have adopted cloud services for their compute, storage and analytics needs are farther along in their cloud journey, and would encounter easier buy-in and less friction internally when seeking to conduct data science and MLOps in a SaaS platform, closer to their data and other applications in the cloud.
2) Resources and Core Competencies
Every organization has a finite set of human resources, skill set and budget. Hiring the right data engineers, data scientists, MLOps and infrastructure engineers is arduous. Most data analysts and engineers that I have worked with possess a strong desire to learn and grow into the data science field, and leaders need to leverage the domain expertise and deep understanding of their data assets that their teams already have. Upskilling your existing talent pool and giving them a mature MLOps platform to work within solves both problems at once.
Moreover, in larger organizations with many on-premises applications and systems, the cost of keeping the existing applications up and running (aka KTLO - keeping the lights on) is significantly high, leaving less bandwidth for new platforms and implementations. The sheer amount of work here is staggering: procuring and maintaining your own on-premises hardware, refreshing it every few years, building a scalable platform and keeping up with high processing demand including GPUs for your data and analytical needs. Moreover, custom-built homegrown platforms have very high build and maintenance costs that can only be justified if the organizational needs are so unique that commercial off-the-shelf (COTS) software cannot be adapted to them.
If you are resource-strapped or have your infrastructure and cloud engineers tied up on other initiatives, ditch the plans to manage Kubernetes clusters on your own, put your feet up and let the MLOps cloud service provider do what they do best. This way, your team can stay focused on practicing MLOps rather than building or maintaining the platform, while staying a lean team.
3) Information Security and Compliance
Many companies, especially in regulated industries, have stringent infosec processes and compliance requirements for end-to-end application architectures and data flows. Organizations with mature cloud adoption are also likely to have clearer guidelines for evaluating and procuring SaaS applications in the cloud. It’s worth consulting with your sourcing/procurement team for any such processes and guidelines to follow.
It is also important to understand the sources, volume, variety and flow of your data assets planned for use in your machine learning projects, as you’ll need to plan for the right connectivity, staging or sampling of your training data. In addition, be sure to understand the profile of your data elements, including the sensitivity of the data (e.g - Personal Identifiable Information or Protected Health Information), and how they influence the security and compliance requirements of your organization - whether industry specific (e.g - HIPAA for healthcare, insurance) or region specific (such as GDPR). Extremely sensitive data (especially hosted on-premises) may be best connected to an on-prem MLOps platform, however, SaaS providers with comprehensive security and compliance in place serve most organizations’ infosec requirements very well.
As you proceed along this journey, ask your SaaS provider what their current and future cloud security controls and compliance accreditations look like, including SOC2 and ISO certifications, along with other industry-specific ones. Additionally, understand the data and compute isolation architecture of your service provider and assess your preferences for single-tenant vs multi-tenant topology.
4) Pace and Culture
Your organization, like any other, has a culture of its own (the way things are done), and a pace and rhythm at which it moves. This is driven by various factors including:
- Whether the Technology organizations are highly engineering-driven.
- How centralized or decentralized they are.
- Whether there is a tendency to build custom/home-grown platforms versus using commercial software.
Depending on these factors, decision-making for technology and tooling may either be more centralized or decentralized across business units — especially in larger organizations — and software procurement processes could also be driven or influenced by any one particular line of business.
It’s also important for data science and platform leaders to understand their own organizational culture and align with the pace of change. If speed of innovation in data science is a key driver, then starting out with one team or a subset of the organization on a SaaS-based MLOps platform is the right move to avoid delays of bureaucracy, allowing you an opportunity to expand the footprint as and when you realize more value and traction.
Finally, it’s important to recognize that sometimes timing is everything. Data science and technology leaders can save themselves substantial headaches by recognizing the right time and use cases for ‘do-it-yourself’ implementations, given their current resources and core competencies. The fact is, there are way too many DIY projects in the technology graveyard that have been dragged through their initial hype and build excitement to their gradual maintenance-nightmare state with no love from their users and operators. Instead, consider accelerating and differentiating your data science impact by applying your business domain expertise into leveraging an existing MLOps platform that meets most of your requirements.
In summary, your choice to go with SaaS or not depends on various diverse factors:
- The farther along your organization is in its cloud journey, the easier it gets to adopt a SaaS solution.
- The more limited you are in your technology staffing and skill sets, the better the case to go with SaaS that lets you move forward fast without many of those dependencies.
- Your InfoSec and compliance requirements are dictated by your industry, region and your business data assets in use. Ensure your SaaS provider can meet these requirements.
- The size and makeup of your organizational structure along with the pace and culture influences your vendor procurement process. Collaborate with your procurement team and assess whether a plan to start on SaaS reduces the risks and friction in the process.
No matter where your organization is on the spectrum of private, public, hybrid or multi- cloud, Domino offers you the choice of operating across them all with the associated flexibility and convenience you desire. Your organization might just be more ready to adopt a SaaS deployment model than you think, and it’s always important to know the levers that drive that shift to the cloud. Understanding these levers, you could be the voice of reason in your organization to embrace MLOps as a Service in the cloud, which can accelerate the data science impact on the pace and efficiency of your business.
As you consider which MLOps option is best for you, keep in mind that Domino Data Lab recently launched Domino Cloud, a fully-managed and hosted SaaS offering that gets data science teams up and running quickly on Domino’s Enterprise MLOps platform. We’ve designed Domino Cloud as an always-on, fully-automated and scalable platform so you can focus on data science and leave the hosting, management and maintenance to us. For more information on accelerating data science innovation with Domino Cloud, visit www.dominodatalab.com/dominocloud.