Fast-paced technology like AI can have barriers to entry. Whether infrastructure, data limitations, talent gaps or complying with rapidly changing regulations. Organizations from health care to manufacturing and the public sector are often stymied by these obstacles that can slow AI adoption and use.

Udo Sglavo

Ready-made AI models can bypass the challenges in a containerized manner and are easily accessible and usable. Metaphorically, they provide the house without having to build – saving valuable time and resources – and keeping up with the blistering pace of AI.

Udo Sglavo, vice president of applied AI and modeling R&D at SAS, shares how democratizing access to ready-made models is an AI accelerator with less risk.

Q: What are ready-made AI models and what are the market drivers for this approach?

Sglavo: Ready-made AI models are pre-packaged, industry-tailored AI solution components for quick deployment. They are API-enabled for seamless integration into existing IT ecosystems and cloud-ready for scalability. One of the key market drivers for this approach is the ongoing scarcity of AI talent. Many organizations have AI needs but lack the specialized expertise to develop and implement models from scratch. Ready-made models provide a low-risk, high-ROI alternative, delivering targeted solutions without requiring an entire AI platform investment. These models benefit companies without in-house data scientists and even those with AI teams that lack domain-specific knowledge.

For example, a health care organization may need an AI model for supply chain optimization or customer intelligence – areas outside its core expertise. With ready-made models, businesses can quickly test and deploy AI-driven solutions to address specific challenges without the overhead of custom development.

Q: How can this approach solve common issues like model drift or decay?

Sglavo: That is a good question. Take fraud detection, for example. AI models identify suspicious activity based on historical patterns, but bad actors constantly adapt to evade detection. If a model relies on outdated patterns, its accuracy declines – a phenomenon known as model drift. With ready-made AI models, continuous monitoring detects performance degradation early. When drift is identified, the model is automatically retrained with new data, eliminating the need for a complete rebuild. This approach saves time and resources while keeping the model effective against evolving threats.

More broadly, all AI models have a shelf life due to changes in underlying data. Ongoing monitoring and maintenance ensure they remain high performing. Our ready-made models come with built-in model management, providing a streamlined way to address drift and decay without requiring extensive AI expertise.

Q: How are ready-made AI models trained?

Sglavo: We offer two approaches for delivering ready-made AI models. Let us start with fully pre-trained models. In this case, we handle everything – data collection, model training, and parameter estimation – so organizations can apply the model to new data without additional setup. The alternative is to deliver customizable pipelines: some organizations prefer to train models on their own data, or regulatory requirements demand it. We provide AI models as pipelines for these cases, allowing businesses to run their data through the pipeline and estimate parameters based on their specific needs.

When real-world data is scarce, such as in fraud detection, we can use synthetic data to train and validate the model, ensuring it can detect emerging threats. Both approaches are valid, and the choice depends on an organization’s AI maturity and required customization level.

Q: Are there productivity gains from using ready-made AI models?

Sglavo: Absolutely. Ready-made AI models can significantly boost productivity by eliminating the time-intensive steps of data collection, model development and business alignment. The models are containerized – therefore organizations integrate them, feed in data, and deploy them with minimal effort. If implemented correctly, a ready-made model can move into production in as little as a week, dramatically accelerating AI adoption while reducing operational overhead.

Q: How can an organization use prompts with ready-made AI models?

Sglavo: We are designing some of our ready-made AI models to support prompts, making integration with chatbots seamless and user-friendly. This allows organizations to interact with complex models using natural language without requiring deep AI expertise.

For example, a large consumer-packaged-goods company might want to give truck drivers access to inventory optimization models without requiring them to understand AI. Using our prompt framework, they could build a simple chatbot that enables drivers to query the model in plain language, making AI insights accessible and actionable. This approach lowers the barrier to AI adoption, empowering organizations to deploy sophisticated models in practical, real-world applications quickly and efficiently.

Going forward, these conversational interfaces will evolve to enable more collaborative use of AI models. Rather than simply responding to queries, the system will allow users to define a goal, and the model will autonomously determine the best approach to achieve it. This shift from reactive to goal-driven AI will further enhance accessibility and decision-making, making AI an even more valuable partner in business operations.

Q: Many organizations struggle with data quality issues because of duplicates, inconsistencies and poor data management. How can an entity resolution model help?

Sglavo: Entity resolution streamlines data integration by identifying and merging records that refer to the same entity, even when unique identifiers are missing. It not only helps combine tables without key fields but also consolidates duplicate records within a single dataset, ensuring individuals assigned multiple IDs are correctly recognized as one.

This capability is crucial for customer intelligence, fraud detection, regulatory compliance, and public safety – where reliable, accurate and unified data is essential for decision-making.

Q: Regulations often change and can sometimes slow progress. Can you explain how ready-made AI models can help comply with regulatory bodies?

Sglavo: Imagine you are a company starting with AI and facing regulatory challenges. Lacking experience, you might be unsure about compliance. Not to mention, regulations are constantly evolving, and navigating compliance can be as significant an effort as developing the AI model itself. The time and expense required to satisfy regulatory concerns often match or even exceed the effort of model creation, particularly in highly regulated jurisdictions like the European Union.

Ready-made AI models help organizations meet regulatory requirements without the burden of developing compliance frameworks from scratch. We use our expertise and intellectual property to ensure our models not only solve industry-specific problems but also align with the latest regulatory standards. When companies adopt a ready-made model, they can trust that it meets compliance requirements, reducing legal and operational risks.

Additionally, some of our models include data to populate model cards: comprehensive documentation that provides essential metadata about the model, including its purpose, performance metrics and ethical considerations. This ensures transparency, explainability, and trust – making it easier for organizations to demonstrate compliance to regulatory bodies.

Q: Regulations also vary by country and region, which can be challenging to navigate. Is there a way to regionalize models based on the nuances of laws?

Sglavo: Yes, absolutely. Regionalizing AI models for compliance saves time and cost by ensuring adherence to varying regulations. A model acceptable in the United States may not meet Germany’s legal standards, such as GDPR’s strict data privacy rules. By integrating compliance requirements during training, we use our intellectual property and expertise to build region-specific models. This approach minimizes risk, accelerates adoption, and enables confident deployment across markets without regulatory hurdles.

Q: What is next for ready-made AI models?

Sglavo: AI technology is evolving rapidly and ready-made models will continue to drive adoption across industries. These models are already accelerating use cases in fraud detection, supply chain optimization, entity management, document conversion and health care payment integrity. As organizations seek faster time-to-value and struggle with integrating AI into legacy systems, ready-made models provide a scalable solution to keep business moving.

Looking ahead, we will continue to innovate with our AI modeling team, using our deep industry expertise and intellectual property to expand AI’s impact across both public and private sectors. The future of AI is about making innovative technology more accessible, and ready-made models are at the forefront of that transformation.

Read more about applied AI modeling from SAS bloggers

Share

About Author

Lindsey Coombs

Senior Editor, Data and AI

Lindsey Coombs is a Senior Editor for data and AI at SAS. She researches and writes on topics covering advanced analytics and evolving tech like generative AI. Lindsey is a seasoned communicator with more than 18 years of experience writing content for a broad range of industries and audiences. She is passionate about the safe and ethical use of technology that benefits humanity.

Leave A Reply

Back to Top