Implementing AI seems to be simple at first glance:

1. Get a dataset, run a regression or neural network on it, and voila, you have insights
2. Send your prompts to a generative AI API like AWS or Open AI, and receive excellent content (or code) back

However, beyond the innovation lab, adopting enterprise AI is complex to say the least.

In fact, one of the top challenges cited by most business leaders is about putting the AI insights to use in a systematic fashion.

In this article, I’ll outline some of these needs/challenges and highlight how traditional digital transformation planning can/should be used. (the connected model is a good strategy reference)

1. Technology Strategy

We often don’t think of technology when discussing enterprise AI.

After all, AI is the cool new kid on the block. But all kids need help to meet their potential.

That scaffolding for AI is provided by good old technology architecture.

1. AI operationalization: Once the model is ready, how will it be accessed by our business applications (e.g. website, contact center etc.). It’s a fact that there are often a complex set of batch processes pushing the output of AI models between various databases.

2. Closed loop feedback: How will the business applications send data back to our data stores used by the data scientists

3. If we use 3rd party software packages (e.g. CRM, CMS, ERP etc.), how will they be integrated?

2. Data Strategy

Our enterprise AI data strategy dictates the availability and quality of your data for very practical purposes.

1. Lack of sufficient data will affect how well we can satisfy the business needs through AI. After all, if data scientists don’t have what they need to predict something, they won’t be able to deliver very reliable models.
2. The speed at which the new data is available (latency) will affect how well we can update the models based on new data available. This is generally not that big of a problem.
3. The location and quality of where the data is will affect our bottom line. More effort will be needed to aggregate the data in a way that is usable by the data scientists. In fact, it is estimated that 50-60% of total analytics effort is spent aggregating and cleaning the data.

3. Defining Business Priorities

AI is used to:

1. Improve revenue and CX
2. Reduce costs / improve operations
3. Be future ready

Each of these buckets can have many initiatives defined under them. And we can’t do all of them at the same time.

So the first thing to do is to define objectives in the order of priority. For example, what seems to be the most pressing problems we want to solve, followed by others?

4. Enterprise AI Governance Needs

The pesky aspects of governance always seem to slow things down. But they need to be solved for before we can roll out anything to the market.

For example:

1. Can we explain how the AI models reached their conclusions?
2. Do we have an underfitting or overfitting model that is also likely to be erroneous in the real world? Underfitting is also technically called “bias“. For example: using an AI model trained on one race of people for a more mixed population. (note: Entrepreneurial type people heard the words “race” and “bias” and started making very good money on the topics of racism and equity in AI)
3. Privacy & security: Is the data being sent to an external service or cloud? Is it masked properly?

Path Forward

It seems that defining an enterprise AI strategy needs things which are very much like defining a digital transformation plan.

Even if we are thinking of implementing AI modeling locally for single use case or one business function, it does need a holistic view of the following to make it work.

1. Business needs prioritization
2. Governance
3. Data strategy
4. Tech scaffolding
5. Digital CX definition
6. Automation

Boring, but the recipe is still the same to make exciting things come to life.

So, I recommend starting with an enterprise AI Maturity Model that measures your readiness and gaps in each of the above dimensions. Here’s how you can create one.

Evaluating each of these aspects will help you plan the investment roadmap to meet your stated business objectives. And your enterprise AI projects will flow much faster.

To save time, we can optimize in many ways. For example, we can take short term actions like proceeding with AI model definition/testing by making data quickly/manually available. While that is going on we can plan out the change management plan.

As with almost all things AI (especially gen AI), there is also a high error rate (called hallucinations in gen AI). So, creating the right scaffolding needs a combination of techniques – after all, these are predictions, not facts.

Hopefully, by thinking of our AI project more holistically as an AI strategy/approach to AI, we will be able to get to implementing enterprise AI much more efficiently.

PS: No AI was used in the production of this blog. I’ll probably use it next time so blogs read better.

If you’d like me to help you build an Enterprise AI Maturity Model, get in touch.