Video: Developing an AI Strategy - 4 Fundamental Pillars

In this video, Graeme Cox, a principal AI consultant introduces the 4 pillars that are fundamental to developing a successful AI Strategy. This is an excerpt from a more complete presentation on developing an AI strategy, which Graeme is presenting on Friday 15 March (9.30 am to 10.30 am - UK).

Friday Briefing (15 March)

Submit this form, if you would like to attend the Friday Briefing on Developing an AI Strategy, which Graeme is presenting on 15 March from 9.30 am to 10.30 am (UK Time)


So when we're undertaking, strategic evaluations of, AI strategies and roadmaps within organizations, helping organizations understand how to tackle AI, how to build AI deeply into their organization.

We look across four key pillars.

They are they start with value first and foremost.

What is the measurable ROI of an AI project?

We see a lot of projects in organizations that sitting labs and benches that look really, really exciting that have been driven by technology evangelists and from the software engineering teams that look really cool, but actually there's no way of understanding the the practical benefits they will deliver to a company. There are various measures of ROI that we'll go into later.

But it is very unimportant to understand what they are and how the ROI over a in a great period of time can, beat the cost of capital that is involved in in in executing them as any project should The second of those pillars is feasibility.

Now that starts with, a data audit and understanding what the data is capable of delivering. Data is, of course, the oil, the drives, so much of our business and AI, first and foremost, is driven by data.

We look across three key, areas here. The first is accessibility. Is the data integrated? Is it in a data warehouse?

Is it, is it accessible to a centralized AI system, or does it exist in in multiple physical locations? What is the data data engineering scale required to get value from these data sources.

The second is the harmonization of the data how complete and consistent is it? Our do primary keys exist across the data that allow data to be correlated between datasets.

Is it clean?

How many gaps are there in it? And what again, what work needs to be done to put it in a state where it can be used for training and driving AI models.

And the third is actual value to AI.

Many datasets look really promising, but when you drill into them, they don't deliver the value that you're hoping and vice versa, we sometimes find that data sources are being overlooked where real strong potential value from AI systems can be found.

The third of the pillars is the human one capability.

AI maturity in an organization is a is a soft concept, but it's so important.

When we're talking about driving AI into an organization, we're fundamentally talking about changing processes and ways of working. It's very easy to take on a co pilot system for helping develop software, but the natural impact of that over the next two years is likely to be either a reduction in headcount or at very least a slowdown of hiring into that team as automation starts to take over. And that's true across the business. As it is true as you roll out AI into customer facing processes as well.

Top down understanding buy in and belief in AI really matters and selling that within an organization based on, first and foremost, ROI, and risk.

Is critical.

And the final pillar here is advisability. Should you do this thing? Five key presets to look at in there is can a lay person understand it? Is it explainable?

Does it produce unintentional bias and discriminate against certain, groups of people, fairness?

Will it withstand cybersecurity attacks? How strong is the infrastructure?

How how traducible is the model, that's robustness, an approach to explaining what it is you're doing Why you're doing it, the strengths and limitations of your approach through transparency is incredibly important to get user buy in, again, whether that's employees or customers, And finally, a very strong approach to data privacy, and personal data privacy, particularly matters as it does in all other parts of our business, but even more so in AI where inferences can be made across big data sets that might identify individuals and exposed data about them that, standard IT, undertakings might not.

Okay. Let's go on and have a look a little bit more at ROI.