What does AI-ready even mean?

Susan Payne Cook, CEO, DAS42

Hey CIOs, being a CIO right now must be incredibly frustrating in this confusing, complex, sometimes scary and always overhyped world of AI.

When your CEO demands tokenmaxx to drive AI adoption and your CISO says to lock AI down, you read a million articles written by AI, telling you how to do AI, without allowing rogue AI to go unchecked by a different AI. So, you implement agentic AI to automate processes and enable generative AI to eliminate those junior roles, all while being inundated by AI-generated slop that uses the same tired platitudes everyone on LinkedIn tells you are the right way to do AI.

And, in case you did not have enough to do, every consultant is reaching out to tell you they can get you AI-ready in a matter of weeks or maybe months if they really want to fleece you!

What does “AI-ready” even mean in this current landscape?

I will never profess to know all the answers, because no one does. But, having spent three decades in the data and analytics industry and leading a team of highly skilled data engineers in the Snowflake ecosystem, I have learned a few essential truths.

1. For the initial project, pick the right line-of-business partner, for example, we love working with CMOs because they understand data, with a real use case that will show value quickly. The problem you are trying to solve needs to be as specific and tangible as possible. You cannot be successful if the goal is a squishy objective like “implement agents.” Get very granular on what actually hurts and clearly describe what the solution actually fixes. Business teams should be stewards of the platforms that generate data that can be used for analysis and AI acceleration. Curate a culture of ownership with your partners.

2. Get the basics in place first: a modern data estate platform. We like Snowflake for many reasons, but mainly to eliminate as much complexity as possible because it’s cloud-native, SQL-centric, with tons of functions fully integrated and easy to maintain. Basic access controls, security hygiene and governance are non-negotiable. AI-ready means you've thought about your ecosystem and which capabilities and partners you want to invest in. Don't throw the net so wide that you can't become an expert at anything. Give your people guidance on what to use and how to use it. Utilize the data platform’s embedded capabilities as much as possible. For example, most enterprises won’t need a separate third-party data catalog product over what is already in your data platform. Don't increase the complexity of adding more tools until you have to, and only for reasons that are justifiable and clear to everyone.

  ​AI-ready means you've thought about your ecosystem and which capabilities and partners you want to invest in.   

3 Have a well-defined medallion data architecture and data governance framework in place. What sources need to be ingested in raw form in order to produce a transformed, clean gold layer that will address the use case? Build fault-tolerant data pipelines and consistent, centralized, performant storage formats for your data, structured and unstructured.

Regulations around AI usage and data privacy are shifting, with renewed focus on data inputs, prediction targets and data exposure for training and telemetry. Keep the logic behind predictions human-readable, and implement a robust tagging framework for PII, PHI and other sensitive data classifications.

Tracking the data that flows into predictive models and agents is a key requirement for both compliance and platform optimization reasons. Auditing and data exclusion for workloads is easy to do in a well-governed platform, but is difficult and expensive to address retroactively.

Invest upfront by building modern semantic layers, RBAC and dynamic access frameworks and automated governance and data classification tools. This bears repeating: data security best practices and governance are non-negotiable.

4. What kind of AI do you want to use for: writing marketing materials, having agents adjust campaigns in real time, complex matching to resolve customer identities, automating responses to support tickets, customer service chatbots, etcetera? Use the right model for the job. Keep your options open. LLM innovation is happening at warp speed and the right one today may not be the best one forever.

5. Deliver value, fast. Don't let the everyday distractions get you off track. You need a tangible win to eliminate that perception that the team is spending a ton of time and tokens on “science experiments”. But, most importantly, remember that AI is not set-and-forget. Models and agents drift over time, so plan to review, retrain or refine and re-attribute on a fixed schedule. When a workload or agent underperforms, treat it like a promising employee who missed a goal: adjust the model type, inputs or scale rather than scrapping it.