Before Google, I was a banker… And that’s where I got involved in AI.
Banks have a lot of data. They store a lot of information about customers, but at that time it was disconnected from anything meaningful. The data was completely unstructured without a comprehensive view of any specific customer. However, the banks also have the AI models already in place in some parts of the organization, but they were simply looking at human behavior in a linear fashion.

Analytical Models

No two parts of the bank were at the same level of analytical maturity, some were at a descriptive model stage and others more advanced were at a predictive model stage. In construction terms one set of models was descriptive – looking for idle excavators on site. While predictive analytic models – look for the idle excavators and also predict when the idle times occur.

The next level of model maturity is prescriptive analytics which not only looks and predicts but also takes action. Therefore, you need to know where the organization is in the analytical maturity ladder. You can only use the analytical models that the available technology supports. Value is only meaningful if everyone acknowledges it.

Think big, start small – deliver early wins to build support and demonstrate value.

AI requires constant model production – one-offs usually ultimately fail. Experience informs me that a lot of organizations are failing to bring AI models to production. Use case: At Google, AI is used to scan Gmail spam folders to classify the contents at a rate of 100 million mails per hour. That level of data generates highly accurate results. Most do not have that level of data throughput, and do not need it.

Get started

AI models need wide access to data whether that is mechanical analytics, telematics, site plan, weather, local traffic, etc. to enable models that offer real value. Cloud IT infrastructure now offers reasonably priced, secure and accessible processing that previously was unaffordable to most organizations, today you pay for the technology processing you are using.

Cloud IT infrastructure is affordable and scalable, so use it for AI models.

Turan Bulmus

We can all access high performance computing without owning the servers, so your data scientists and IT staff are focused on delivering value and not IT services. Democratize the AI Capability – knowledge transfer is important to gain more value.

It’s a long journey, you are not alone, so reach out to your technology partners.

Watch the full key note here: