We talked to Ben Saunders, Chief Customer Officer at MESH AI, about the use of AI in business and the company’s report, The State of AI in the Enterprise.
With AI developing leaps and bounds daily, one would be safe to assume that businesses would readily adopt it into their structures. Matt Brundrett asked Ben Saunders about the benefits of AI in business and why it might not be easy for a business to utilise AI immediately.
What would you say are the key potential benefits of the use of AI in business? And what about more public sectors such as healthcare?
At Mesh-AI, we see a full spectrum of challenges that we help customers to comprehend and understand. The first area organisations typically look at when deploying AI capabilities is using it to identify operational efficiencies and streamline the more manual or cumbersome tasks that are typically human-intensive.
Automation is often seen as an easy entry point to help them overcome some of those challenges. But when we speak to customers, we say there are three brackets of opportunity with AI:
Using AI to make money, save money, or reduce risk. We typically find that saving money is the easiest area to explore. Generally, that comes within the realms of operational efficiencies and streamlining business processes, looking at ways in which you may be able to strip out duplication of effort across an end-to-end business process.
This essentially means simplification, standardisation, and cost efficiency across the board. Those operational processes are things that would typically occur within the business itself rather than being customer-impacting because doing the former first enables organisations to build the required muscle memory in their business – safely and securely scaling AI capabilities within their own organisation before they start to move it into a more customer-impacting capacity, which is still a foggy area for many businesses.
Particularly with the explosion of generative AI and various other capabilities that have come to market in the last year. Considering this, reducing costs is a massive focus point for organisations.
Still, you can’t cut costs if you’re not engaging with customers, so when we speak about Machine Learning and AI, there are opportunities to understand your customer better and perform ‘360° accurate’ activities, cluster the targets and prospects that you want to go after, and better understand what the customer looks for.
What we saw in the research is this amplification of generative AI adoption specifically. We also see several opportunities across the generative AI landscape. You can use generative AI to develop new business hypotheses and opportunities to engage customers. Then there’s using generative AI to build those services and service customers once you’ve built those things.
Ultimately, you can use generative AI in the back end to provide operational controls around the whole spectrum of capability, i.e. using robotics, Machine Learning, and advanced analytics and AI-enabled software to run businesses much more efficiently. This means we can build 1,000 messages for 1,000 customers instead of one for 1,000 customers.
In the public sector, thinking about life sciences, health care, and the National Health Service, there are already great examples of Artificial Intelligence and more sophisticated analytics being used. Cancer, as an example, and looking at ways in which we can use Artificial Intelligence to better determine the early onset of signs of cancer. We can start to use a wealth of data gathered by the NHS and other public health bodies worldwide to better identify the habits, patterns, and medical insights into patient health.
All that data already exists when looking at X-rays, MRI scans, CT scans, etc. We just need to harness it so that Machine Learning and AI can better identify those health challenges for patients and then make the right assertions concerning the clinical interventions the healthcare service should recommend to patients.
Another example is HMRC. If you were to look across public servicing bodies such as councils, there are opportunities where people have previously filled in forms and then waited eight to ten weeks for their local council to get back to them. We can use AI generative capabilities to provide taxpayers with the right information. And they can get a much more tailored and direct response to their issue instead of a vacuous experience.
In your report, you state that 30-49% of respondents say that 40% of their AI projects failed; what would you say is the reason for such a relatively high failure rate?
There are two ways to look at that. Either the failure rate is high, or there is a high rate of pulling the plug on projects that aren’t going to work. Several organisations have a challenge when thinking about business strategy, especially when launching new products because your data and AI strategy need to be harmonious.
There is often a discrepancy between the business strategy and what technology thinks the business needs. This is where there is a clear requirement for organisations to bring together business data and AI capabilities through cross-functional teams, re-devising an operating model, and fundamentally thinking differently about how AI can maximise the reimagining of the business.
The other thing that we need to mention is accountability and ownership. There are challenges within AI that mean it has the potential to cause harm to the firm, to the customer, or even the wider market.For example, if it behaves erroneously and spins out all sorts of spurious assertions.
This makes it important to ensure an end-to-end understanding of who, what, why, when, and how with respect to Machine Learning and those Artificial Intelligence capabilities coming into a production setting. The biggest problem is that organisations aren’t mapping this to compelling business drivers.
It’s important not only to ask but understand: why do you want to do AI? It needs a particular outcome, thinking about those earlier points of making money, saving money, and reducing risk.
If you take those three categories, you could spin out all manner of assertions, business interventions, challenges, and opportunities you want to drive. But unless you understand that fundamental reason for implementation, you’re ultimately just creating a science project that isn’t going to have the desired outcome and drive the right change behaviours in your organisation. It becomes a leadership, cultural, and education requirement.
The report also suggests that while various issues faced adoption, the most prominent issues are data quality and accessibility. What are the specifics about that, and what will the possible solutions to that be for businesses?
Let’s say you’re someone in financial services who makes asset management and investment recommendations for your customers. If you deploy a large language model over your data and then make a recommendation to that customer, you need a strong assertion on the data that you’ve trained that model on. Otherwise, you are running the very real risk of a PII scandal where the mis-selling of financial products is significantly challenged.
When thinking about who, what, why, when, and how, we must ensure that the quality of data we train these large language models on is sufficient. We need a coarse-grained understanding of where that data has come from, how it’s been curated, and who owns it. Then, when we start to put a large language model or Machine Learning capability over the top, we want to understand the origin of those things. In an enterprise, we must ask how we remove this veil over the data where people are churning out spreadsheets, sending them via email, via real-time collaboration, or wherever else, and make a point to keep this data in systems of record.
We then build what we call a metadata management framework over the top of the data, essentially data about data. Who owns it, where it came from, what record systems it is associated with, the type of data, etc. In doing this, you will immediately know that it is legitimate. This creates a catalogue of data that is easily accessible within the enterprise. This not only puts the right controls around the data, but it enables you to move away from this position where people are throwing more unreliable things around and curating the data, cataloguing it, and making it accessible to organisations or parts of the organisation that wouldn’t necessarily have been able to get their hands on the data.
In addition, you need to consider how the public cloud will play a role in this. This is where we talk about immutable infrastructure components that will allow you to run Machine Learning capabilities over the top of that data. This is a self-service infrastructure, so you can get the data to a curated state, build a model over the top of that, and then look at how you would build those types of models into your AI-enabled software.
Going back to when we talked about a dynamic change within the operating model, breaking down the silos between business and technology, establishing cross-functional teams that are aligned to real business outcomes for your customers and for your organisation, and finally looking at ways in which you can introduce highly automated ways of working, such as DevOps, continuous delivery, and others so that you can bring all of this together to build a more iterative operating model around your AI adoption strategy.
What would you advise businesses that want to adopt AI into their models?
I recommend not obsessing about what this could look like for months on end because the opportunity window is in front of you right now. What I would encourage organisations to do is ensure that they can get some form of proof-of-value instantiated in a two-to-three-month timeframe. Something with demonstrable business value that either tackles challenges for your internal customers or can add massive value to your actual customers.
But what you can also do is flesh out your regulatory posture and how you would demonstrate compliance with increasingly complex regulations. Ensure you aren’t introducing bias and the AI is essentially being aligned with your organisation’s values. The aim should be to bring all those things together in an accelerated timeframe.
Furthermore, investment in people is fundamental. Ideally, undertake a two-to-three-month lighthouse project that gives you an opportunity to reskill your workforce. It enables you to better understand how AI can be applied in your business and gives you some initial trajectory to demonstrate something impacting the organisation. You can then use that as a catalyst to talk about future initiatives and scale further opportunities for the business thereafter. We would always tell customers to try to crowdsource their ideas and build a cross-functional working group of varied and diverse people across their business to curate those opportunities.
The people who are on the front line of the business are more likely to understand the challenges for customers. Building a cross-functional team of representatives from your customer services team, projects and solutions team, technology teams, and HR and finance teams enables you to identify a whole raft of opportunities.
This essentially comes back to efficacy because, from the beginning of building that AI strategy, you can say from the outset that you have a diverse workforce that has enabled you to build this and strengthen the lineage. That’s probably something we don’t particularly discuss in the research paper, but I think that’s an important consideration for enterprise organisations. To really consider over the span of the project that diversity of thinking in how AI can be applied, and furthermore the diversity and how you govern AI end-to-end as well.
Please note, this article will also appear in the sixteenth edition of our quarterly publication.
