There is no AI without IA: How to Build Information Architecture

May 20, 2020 | 3 minute read
Silviu Teodoru
Information Architecture
Text Size 100%:

This post was previously published on LinkedIn Pulse

When we talk about enterprise adoption of artificial intelligence (AI) across large companies, we can agree that AI needs a strong information architecture (IA), which is the way data is organized and structured so that users find what they need. 

The journey towards AI business value has the following maturity levels (in an iterative manner), all being key building blocks in a modern IA process. 

Subscribe to the Oracle AI & Data Science Newsletter to get the latest posts sent to your inbox! 

 

1. Build a Strong Data Foundation

Information architecture starts with the ability to collect and store raw data from different perspectives and paradigms: batch collection and real-time (IoT, streaming data), structured and unstructured data, transactional and analytical data, relational and knowledge graphs represented data.

Converged databases and/or a big data stack are the solutions of choice to build such a strong data foundation.

Once you have the data, the question is how to make it understandable and accessible. 

 

2. Build Trust on Top of the Data Foundation

In order for organizations to benefit from the data they have, they need to understand it and trust it. Thus, data governance is key to operationalise AI in organizations with trust and transparency.

There are different means to understand and govern the data:

- Ensure enterprise data quality: Profile data; understand, monitor and improve data quality; enforce validation rules

- Design and build metadata for trusted AI: Discover and improve metadata assets (business, technical, operational), classify data assets, build a data glossary

- Prepare data for consumption: Engineer data through different transformations to be ready for business consumption; democratize access to information, put in place policies, and ensure privacy while accessing data. 

Only when the level of maturity in data governance is improved can one say which data is understood and trusted. In turn, this data is business-ready, and the company is able to make real steps towards the data-driven organization desideratum.

 

3. Empower AI for Enterprise Data Consumption and Valuation

Now is the right moment to put AI to work and to execute the full AI lifecycle: business challenge understanding; data exploration and preparation; modeling; evaluation; model deployment, monitoring and improvement.  

In addition to trusted data, the main ingredients necessary to execute this step at the right level are:

People: Skilled resources with different roles within the organization, from business owner and business analyst (to define the need), to data engineer and data scientist (to explore and prepare the data and the model), to MLOps teams (to operationalise and monitor the model in production)

Technology: Tools and techniques to automate the entire AI lifecycle. The offer is large and constantly increasing on the market, both from open source communities and large technology players

Methodology: The right approach and steps which ensures the success of AI projects. Without a rigorous framework able to connect end-to-end business needs, stakeholders' support, and AI's technical power, most AI projects will never go into production.

Thus, using AI on trusted data, organizations can obtain additional automation and optimization plus prediction capabilities.

 

4. Infuse AI Throughout the Business 

This is the level of maturity when one company can name itself a data-driven organization.

Using the power of AI, the company is able to identify and benefit from the value in data. AI infusion into business can be done from an analytical perspective (predictive or prescriptive analytics) or from an operational perspective (recommendation engines inside transactional processes).

It's key to all AI implementations to align model performance to business outcomes, and to correlate technical model metrics with business KPIs, in order to measure the business impact in an ongoing feedback loop. 

Since the only constant is the change, including in the business and data context, we have to ensure that the models are resilient to these changing situations and that they automatically adapt once model drift and/or data perturbation is detected. This is important in order for the AI model to remain relevant for business objectives.

Last but not least, businesses are executed under regulated conditions; thus, it is essential to ensure AI production monitoring for compliance. This is why companies should be able to explain and audit model decisions or to detect and mitigate model biases. There's a large focus on these aspects, both from the research communities and AI technology providers.

 

Conclusion

AI is the new normal, and the business potential as estimated by analysts is mind-blowing. For example, Gartner estimates AI augmentation will create $2.9 trillion of business value globally in 2021. But, in order to reach this magnitude of potential, modern IA is critical in order to get business value from data at enterprise scale.

Thus, IA is the lifeblood to AI. They are the two sides of the same coin...the success coin.

 

To learn how AI and data science can benefit your business, visit the Oracle AI page, and follow us on Twitter

Silviu Teodoru

Information Architecture


Previous Post

Multi-Head Self-Attention in NLP

Praphul Singh | 4 min read

Next Post


How AI, IoT, and blockchain boost efficiency

Emma Hitzke | 3 min read