The relatively slow pace of adoption I talked about in part 1 is more to do with dealing with the unknown associated with AI. Simply knowing what to expect, and how and where to start, were reported as major challenges.
Look at fig. 1 below and you’ll see that working out the financial benefit of AI adoption and even the cost of implementation are high on the list of concerns.
These two factors were considered the most difficult (scoring either 4 or 5 in a 5-point scale) by 58% and 49% of respondents, respectively.
Even the core strategic issue – would AI help our business and how? – remains a difficult question for many, with “business level scoping and prioritizing” mentioned by 51% on the same ranking basis.
Other areas causing concern include the AI skills shortage (53%), plus the challenges of securing (44%) and managing, maintaining and evolving (44%) an AI platform in everyday use.
Figure 1: Unfamiliarity with AI and its relative novelty present challenges for many
One last hurdle that may not be appreciated until a project is well underway is that AI is very much a multidisciplinary team sport. While our respondents think AI projects might be driven by different groups within an organization, more than a third agree that people are not always pulling together as well as they should.
Making AI easier to consume
The uncertainty we have uncovered is clearly an area where suppliers and service providers who already possess AI expertise can do a lot to help. Suppliers have a key role in overcoming barriers and making AI more comprehensible and accessible, particularly in terms of de-risking and streamlining adoption. This comes out clearly in the survey, where the participants saw considerable appeal in optimized platforms, reference architectures, pre-integrated systems and similar approaches to de-risking such as managed services [more details to the report here].
There are many ways to deliver and consume AI capabilities, including public cloud services addressed via APIs, dedicated machine learning and deep learning software stacks running locally or on a hosted service. Other options include machine learning embedded within Software-as-a-Service (SaaS) applications or hardware devices, and several various other hardware/software combinations.
Customers’ delivery model preferences are surprisingly evenly matched – the spread across five options was only nine percent. Function-as-a-Service (FaaS), which offers discrete cloud-based AI functions accessed selectively via APIs, headed the list, at 53%.
Interestingly, at the other end of the list was Platform-as-a-Service (PaaS), which is the use of full public cloud-based platforms to provide comprehensive AI capabilities - only 44% expressed an interest. Which is in between were data center models (47%), private hosting (49%) and a hybrid-approach (45%).
The manufacturing focus is, as we also saw in part 1, somewhat different here. Levels of interest in all modes of delivery are high, but with private hosting coming out on top (62%) followed by the data center model (60%). FaaS is at 58% and, again, PaaS comes in last, this time at 48%. I believe this is primarily related to the fact that - workloads which are trained are best done closer to where the data is obtained from and hence more effective and secure to have the solution on-premises.
I was also fascinated to see that pre-integrated systems and reference architectures are seen to help accelerate, streamline and de-risk AI adoption. These are well out in front of a list of possible AI acceleration mechanisms offered to survey participants, at 63% and 57%, respectively.
How to get ahead in AI
As I mentioned in the first part of this blog, manufacturing specialists consistently show deeper commitment to the use of AI. The AI opportunity appears to be clearer and they are rolling it out faster with a sharper appreciation of specific use cases.
More broadly, when it comes to AI, uptake and acceptance are infectious. In other words, once you recognize and understand the potential via one project, you are likely to see the opportunities for broader AI usage across an organization (Figure 2).
Figure 2: AI is both broadly relevant and of broad interest
If you are looking for wider use cases, then a number of things can help.
Find a quick win application area which offers obvious positive and measurable results. For example, an area where automation clearly offers significant benefits but was previously not feasible or not cost-effective.
Second, remember that packaged and semi-packaged AI platforms can enable faster and easier implementation and reduce many of the risks involved in new technology.
Finally, work with a supplier such as Fujitsu, that understands both AI and how your industry operates.
Clearly AI adoption is complex and people are seeking help. If you have a use case that looks promising for AI, that’s a conversation where Fujitsu can offer insight and value. Contact us to find out more.