2018 was an enlightening and tough year in delivery of AI to enterprise clients. I feel that we are at the crux, at the place where enterprises will really start to adopt AI, or will let it pass as a ‘latest fad’ and will either die as their competitors march ahead, or less likely survive as their AI-adopting competitors burn money and get zero in return!

Against this backdrop, here are 7 things I’ve learned from delivering enterprise AI projects in 2018:

“You don’t need to be at the cutting edge to deliver value”

In the past 12 months I’ve worked on, and been aware of, projects that have taken the very latest products from some of the smartest minds out there, and many projects that have leveraged AI, ML and NLP technology that has been around for many years. The level of complexity or how advanced the technology was has not been a key critical success factor in the project. What has been critical, and this is the same with any successful technology project, is the measure of how well the technology answered the business problem.

“It’s not all about AI”

Often a project will start out with a view that AI can make it better, and in reality that is probably true. However, it is not often the best place to start. A business process can be improved through many other means, for example a more intuitive user interface, removing redundant parts of the process, using robotic process automation for the repetitive, structured and non-subjective parts of a process.

“Its as much about analysing unstructured data as automating the process”

Sometimes what the business needs is the ability to get meaning from unstructured data, and determine correlations between disparate unstructured and structured data sources (think images of houses and neighborhoods correlated with house prices). The downstream business process may be entirely, or largely, manual. Yet the emergence of advanced data processing techniques opens up new business possibilities.

“Don’t run before you can walk”

It’s often tempting the start with a complex case, build something that changes the world (or the organisation!), but for many companies this is the wrong thing to do. Imagine learning to mountain climb on Everest. Organisations that do well start with small use cases and build momentum. They work with experienced people and partners who’ve been there, done it, and have the scars to prove it. In some cases, walking could be leveraging RPA to automate the more simple parts of a process and gain confidence.

“It’s not just about the AI engineer and data scientist”

It’s tempting to build a team of data scientists and AI engineers and go tackle the world. However, if you think of AI as part of an enterprise project, what other skills will you need? Often you’ll need a UI for exposing the inner workings of your model to your staff for validation or processing. The model’s output will likely need to go to a downstream enterprise system (potentially with significant change management, security and data integrity controls around it!). You’ll also want to ensure that features are built based on business needs not technology interest.

“Ensure you have stakeholders engaged”

This is certainly up there as a critical factor in enterprise adoption of AI. In large organisations this can mean multiple stakeholders, some holding the budget and others owning the technology vision. This is good old technology delivery, so make sure you have people on the team able to manage these relationships, or your data scientists will get very bored with lack of progress very quickly and are likely to leave.

“Don’t forget your users”

At the end of the day, there will be users that rely on the output of any AI solution that you implement, and their ability or willingness to use it is critical. You may have just implemented the most advanced AI solution ever and be well on your way to creating the singularity, but if the organisation can’t or won’t use it, or external customers don’t like the experience, then you’ve wasted your time. I always advocate, as for any project, that an understanding of the users’ needs is utmost here. Echoing a point from earlier in the article, you’re better off delivering a solution that fully meets 20% of a users needs than somewhat delivering a solution that meets 60% but leaves them with a bad experience (of course the percentages are indicative!)

So what about 2019…

While I fully expect the large technology companies such as Google and Facebook to push ahead with ever more mouth-watering technology announcements, for the majority of organisations I see the following trends for 2019:

1. Starting out, or slowly building momentum, by tackling ever more increasing use cases for automation and AI
2. Starting to see value in the disparate, unstructured data sources within the organisation, and for those more advanced, how these can line up with open data
3. An understanding of ethics and model bias, particularly where AI is used to make business decisions
4. A wider use of SaaS and off-the-shelf products, such as those provided by IBM, Microsoft, Google and Amazon.
5. A focus on measuring business results, and a drive to predict/measure ROI ahead of project delivery
6. A focus on user experience, and hopefully a few less bad chatbots!
7. And no AI winter this year!