02 Dec AI adoption: what are the barriers and how can we overcome them?
The road from AI goal setting, to proof-of-concept, to fully operationalised deployment is a long one. AI adoption requires investment, leadership support and adapting to new ways of working. There are few quick fixes or shortcuts. AI and machine learning is a journey. We know this because we’re on that journey ourselves.
As more organisations adopt AI, the benefits are plain for all to see. Our proof of concept AI deployment recently saved us a projected £300,000 a year in manufacturing costs. We are now applying the data platform that we created to other business challenges.
But as the saying goes, ‘a journey of a thousand miles begins with a single step.’ And often, those first steps can be some of the hardest. From legacy systems to securing investment and even suspicion of AI in the workplace, there are a number of challenges that have to be overcome at an early stage in the discovery process.
In this post, we’ll look at five AI adoption barriers and how you can overcome them.
A lot of organisations still rely on legacy infrastructure, applications or devices to deliver some of their IT operations. Upgrading everything in one go is a huge challenge. This legacy infrastructure is often seen as an impediment to adopting machine learning or AI. Thankfully, cloud computing – or more specifically, hybrid-cloud – has changed that. Hybrid cloud is when an organisation uses both on-premises and cloud-hosted IT infrastructure. This is an increasingly common situation.
Embracing AI and machine learning doesn’t mean you have to upgrade your entire IT estate. But it does require you to embrace the cloud for your data analytics and AI. Modern ‘Data Lake’ technology works well in a hybrid environment, where the cloud can be used for analytics with operational systems on-premises. The cloud analytics systems can even push data back into the on-premises operational systems to direct their operations more effectively.
The other advantage of cloud solutions is that updates and new features are rolled out automatically, which leads to fewer misconfigurations, security vulnerabilities and incompatibilities, all of which are important to stay on top of when deploying emerging technologies.
It always takes a while for the talent pool to catch up with new solutions. As a result, there is a skills shortage around AI and machine learning and this can hamper efforts to develop the capability.
We’ve worked hard to recognise and nurture talent internally. Giving people within the business (and not just IT!) the opportunity to get involved has been a great way of finding enthusiastic people who are excited about AI and want to help.
Pulling in people from a wide variety of backgrounds is essential. Multi-disciplinary teams always work best, so try to balance technical personalities and skillsets with business and non-technical backgrounds. This also helps to mitigate against inherent bias in your modelling, which is a big issue across data science. If the team and the data lack diversity, then the results can be self-fulfilling and even perpetuate that lack of diversity. The data needs to be representative – and so does the team.
Partnerships are also useful. We’ve sponsored students on Data Science degrees at a local university, which has been a good source of technical talent. We’ve also developed partnerships with organisations, such as the CBI. Getting involved in the community, attending hackathons and meeting like-minded people also helps the team to grow their network, keep pace with the latest thinking and stay passionate.
You also need to create the right environment for staff to learn and grow. We do some formal training, but mostly we learn by doing. We’ve thought a lot about how to develop and encourage an innovative culture. From making the office environment conducive to collaboration to embracing failure and making sure people have time to explore and experiment with new ideas. All of which helps attract and retain the talent we need and helps our people to do their best work.
Lack of support from senior management
We’ve always been fortunate in this regard. Ricoh’s an innovative business and that’s reflected in our leadership. That certainly made the process a lot easier. That said, we still had to make our case and convince management that our approach was sound.
One of the key things to bear in mind when pitching AI to senior stakeholders is that you need to be mindful of your audience and their priorities. Decision-makers aren’t interested in technical and operational details. They’re focused on the wider business objectives and how quickly they can achieve them. Specific and realistic use cases along with projected ROI is much more useful than a ‘big picture’ pitch. The more data and metrics you can use to back up your case, the better.
Lack of a coordinated data strategy
We focused on delivering the overall platform and data strategy first, instead of deploying a use case as quickly as possible. Our thinking was that if we can create a productionised and automated platform in the first instance, we can then extend this to other use cases. Basically, the approach would be more scalable.
What I’ve seen working with other clients is that data science projects can end up siloed. Data ingestion and training isn’t joined-up, automated or productionised. Keeping those siloed projects running and operational then becomes a chore and the projects are less likely to make it beyond the pilot stage.
Of course, we needed at least one use case and model to launch on the platform to demonstrate the benefits, but we always knew there would be time later to refine, improve and add new models. The important thing was to demonstrate a robust and scalable approach.
Once the platform is there, you can deploy new models and integrate new areas of operation faster, which speeds up the process of making AI pervasive throughout the organisation.
Lack of understanding in the wider business of what is involved and what can be achieved
The AI or machine learning team is the face of AI within the organisation. You need to be positive and you need to be visible. Getting support for things internally is about communication. We run regular presentations and workshops to explain our strategy and our objectives with others.
Most people are curious and interested in AI because of all the press attention it’s received. Our job is to explain how it works, what it can do and try our best to take people with us on our journey.
There is some concern around the effects of AI on the job market. But recent reports suggest that AI will create as many jobs as it displaces. Educating people on the reality of AI is a key part of helping people overcome their suspicions or fear of change.
Demonstrating to people that AI can help them, not replace them, is a good use case for this. The dream of a standard 4-day working week on the same salary is a good example to share. That’s something that almost everyone would want to buy into and AI can help us get there.
Are you facing any of these AI adoption barriers?
Feel free to get in touch on LinkedIn to discuss anything that we’ve covered in this article. While AI’s use cases will differ from organisation to organisation, the impediments to making AI happen are often fairly consistent. Having solved many of these things for ourselves, we’re more than happy to do the same for you. Or continue reading Implementing machine learning in the workplace: four key challenges.