Build Or Buy AI: Three Factors That Can Determine Your Strategy
The race is on as organizations adopt AI to improve their operations and gain a competitive advantage. All signs point to a boom, with the global market forecasted to hit $190.61 billion by 2021, as the industry shifts from early adopters to the mainstream. As the foundation for ramping up AI implementation is laid, are organizations prioritizing the right factors that will allow for flexibility and scale for the future?
Many are making decisions based on current deployment options (standardizing around one cloud provider) or available bespoke applications. Others are spending months or even years cobbling together open-source solutions to save money but leaving holes in critical parts of the end-to-end model operations (ModelOps) pipeline, such as monitoring model drift. Unfortunately, these decisions are nearsighted. They only prevent an organization from getting more models into production and generating increased value for the business.
To build the right ModelOps pipeline — one that can adapt to future needs — organizations must plan for shifting priorities and make insightful decisions that will stand the test of time. This includes allowing for multiple deployment options, managing costs and allowing teams the flexibility to incorporate new tech into the process.
Consider these three factors for building — or buying — the right ModelOps pipeline for your organization.
1. Deployment Options
While it might be easy to conceive of a reality where models are only running on one type of infrastructure or location, it’s not realistic to expect that this will extend into the future. Plan for hybrid cloud, multicloud, on-premises and edge deployments and choose tools that allow you to run models wherever you need AI capability delivered. By going with a tool that only allows you to run models built in one ecosystem (e.g., a single cloud service provider) you’re only creating more work for the inevitable future where business models and infrastructure needs change.
2. Integration Flexibility
Along the same lines, it’s important to look for tools that easily integrate with your existing tech stack, as well as allow for extensibility for the future. Rather than locking your data science or development teams into working with the tools or languages du jour, it’s important to prioritize open standards and architectures over vendor lock. Not only will you have happier and more productive teams, but you will also lay the foundation for future updates and enhancements.
3. Cost Control
Lastly, cloud consumption is an overlooked or unaccounted for cost when it comes to AI deployments — that is, until a high bill turns up and you’re the one who must answer for it. To avoid this, prioritize real-time insights into infrastructure usage and costs and the ability to set limits that prevent unexpected spikes in usage. Many options allow you to control autoscaling to ensure low latency and faster processing when you need it. By proactively setting limits and ensuring infrastructure managers can monitor and control usage, you can better optimize processing consumption.
For many, the decision to build or buy comes down to a tradeoff between potential near-term cost savings by building a ModelOps pipeline with open-source tools or the opportunity cost of using a commercial solution in the future.
Having been in the trenches consulting and helping many organizations who have pursued both paths, the long view wins out.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs, and technology executives. Do I qualify?
Source: https://www.forbes.com/sites/forbestechcouncil/2021/09/28/build-or-buy-ai-three-factors-that-can-determine-your-strategy/?sh=69fb87c768bf
Post a Comment