Banks and financial firms are being inundated with stories heralding the benefits of AI and its close cousins, machine learning and deep learning. And those benefits are real. In fact, McKinsey estimates that artificial intelligence can generate up to $1 trillion additional value for the global banking industry annually.¹ And Autonomous Research predicts that by 2030, AI will allow financial institutions to reduce operational costs by 22%. Achieving enterprise-level success and ROI, though, requires the right foundation.²
The tremendous potential of AI, along with advances in compute and storage resources, abundant data sources, and flexible development tools, means we can now safely say that artificial intelligence has transitioned from cool to vital. Where previously, most organizations used AI and ML for peripheral functions such as chatbots, they’re now key to mainstream productivity and revenue drivers, especially as the financial industry becomes more real-time, more global, and much more digital.
Whether you’re looking to implement AI for fraud protection or better customer insights or to improve efficiency with hyperautomation, which Gartner identifies as a top strategic technology trend for 2022, the test will come in transitioning from the proof of concept to a measurable return on investment.
Now let’s talk about practical lessons for applying your AI strategy.
With their enormous data sets and extensive experience with analytics tools, financial firms have an advantage when it comes to AI and ML. But with challenges around compliance, legacy systems, and siloed data, what is needed is a practical approach to implementing AI across the enterprise that allows it to become a core competency.
So let’s look at six steps for turning your AI into ROI.
1. Do AI with a purpose.
This is arguably the most important step. Every AI project must have a clear rationale or a “why” that can be communicated to stakeholders, whether it’s to:
- Generate new insights into markets and customers
- Leverage new or alternative data in decision-making
- Improve efficiency by automating workflows
- Reduce time to market for new products and services
- Enhance risk management and improve accuracy
One obvious “why” for AI is automating repetitive, low-value tasks to reduce human error and enable resources to focus on more strategic work. In wealth management, for example, automation can streamline client onboarding, including KYC requirements, thereby improving customer experience, reducing errors, and freeing staff from box-ticking, allowing them to put their critical skills toward more valuable work.
Another important “why” might be using AI and ML to quickly identify fraud patterns, weed out false positives, and block malicious activity before it impacts the business. An effective ML strategy automatically detects hidden fraud by focusing on subtle pattern changes and unlike rules-based processes, enables algorithms to become more efficient and effective as data sets increase.
Once you understand your “why,” communicate it to your stakeholders. The long-term success of your AI projects will require the ability to move beyond short-term financial gains to a wider understanding of AI’s role in transforming the organization and enabling its longer-term strategic goals. Without setting expectations at the outset, stakeholders may have very optimistic expectations for artificial intelligence as a plug-and-play solution, as opposed to an iterative and fresh way of working.
2. You have the “why,” now what’s the “what”?
What’s the business problem you’re looking to fix and what solutions can you expect? How you frame the business problem can make it a great fit or a bad fit for AI.
First, you need to look at algorithm feasibility: Is this a problem that lends itself to an algo? Not everything does. It could either be because of the type or quantity of data needed or the nature of the events to be analyzed; artificial intelligence can’t solve all issues.
Furthermore, you should also look into whether there are pre-existing algorithms that fit with your problem. If that’s not the case, you must factor in time to build, test, and tune your algos to address it.
Of course, you also need to assess the potential impact of the project. Will solving this problem have a valuable impact on the business? Will the end result improve revenue or customer experience, or reduce costs or risk?
A related question is recurrence: Will your AI project address a recurring issue or is this a one-off situation? How often does your current problem come up and how often will this solution need to be utilized?
Lastly, you need to look at the data. Do you have, and can you access the correct type and amount of data to enable the model to be trained to be accurate? Rely on your data scientists to determine the best approach to applying AI and machine learning and the tools needed to process, extract, transform, and filter the data.
3. Build a winning team.
AI is a team sport. Of course, you’ll need good data scientists who can translate business problems into machine learning models. But if your data scientists are worrying about infrastructure or acting as data herders your team can’t be efficient. The most effective teams have a balance of skills and personalities, as well as a team culture, that enables them to deliver results.
Organizations that successfully make AI an integral part of their business strategy leverage an AIOps/MLOps approach that delivers a flatter, more agile workflow that emphasizes teamwork rather than vertical handoffs. Repeated iterations—test, tune, train, rinse, and repeat—are needed to arrive at usable systems, so the team needs to be in near-constant contact to experiment, fail fast, and learn in a series of short, repeatable steps
4. Get your (data) house in order.
So now that you have buy-in and budget from the C-suite, you’ve determined you can access the data you need and you’ve put your team together, it’s time to get your data house in order.
In surveying recent articles on AI, not surprisingly, the most common topic was data. And by 2024, Gartner predicts that over half of finance organizations will encounter scaling problems with their AI solutions.³
The right infrastructure is therefore critical to move from a promising AI idea to better business outcomes. The ability to access and handle massive amounts of structured and unstructured data is necessary to feed your AI apps with the data they need to generate ROI.
While most process automation up until now has focused on the simpler-to-use structured data, unstructured data is exploding along with the digital economy. In fact, Gartner estimates that unstructured data represents an incredible 80% to 90% of all new enterprise data, and it’s growing 3 times faster than structured data.⁴
What is needed is an infrastructure that addresses four key challenges:
- Consolidation: Whether you’re leveraging text and images for market analysis or earnings transcripts and financial filings to generate complex trading algorithms, you need storage that can effectively consolidate all of your data.
- Performance: AI/ML workflows require powerful compute, fast storage and massive throughput to process vast quantities of data and increasingly complex algorithms.
- Integration: In order to deliver value, AI/ML models must be integrated into existing systems and applications. This can be especially challenging in financial firms with complex legacy platforms and siloed infrastructure.
- Reuse: Successful AI implementation, and an AIOps model, require reuse of data across applications and even the ability to utilize new data created by your AI and ML applications.
Successfully deploying AI requires a data infrastructure that is up to the task of addressing these challenges. Pure created FlashBlade® as a unified fast file and object (UFFO) platform designed to meet the demands of modern data. As throughput-hungry applications demand more, the massively parallel architecture of FlashBlade is perfectly suited to address those needs. And as data reuse grows, FlashBlade can serve as a central repository rather than copying data across applications.
All-in-one solutions such as AIRI//S™ and FlashStack® for AI offer high-performance, architecturally optimized solutions that can seamlessly run within existing data centers to manage any workload on any node, any time.
5. Harness the feedback loop.
If you cannot measure it, you cannot improve it. ~ Lord Kelvin
This one may often be overlooked, but AI works best when it learns from itself, so measuring and reporting is crucial to continuous improvement. And knowledge gained through effective AI is also multiplicative. While initial gains may be small, if done correctly, they can be expected to grow exponentially over time.
Think about how you will quantify the ROI. The “whys” you identified in the beginning will determine your KPIs. Is it time saved in customer onboarding? Is it mistakes avoided that reduce compliance risk? Or is it identifying new trading opportunities that provide additional revenue?
Lastly, continuously auditing data science processes and outputs including the quality and relevance of the input data can help identify “model drift” and ensure reproducibility. Both of these are also important factors when looking to implement explainable AI, which is becoming increasingly important to address questions of fairness and bias in areas such as lending.
6. “Patience you must have.” –Yoda
This one may be a bit obvious, but nothing will kill AI efforts faster than unrealistic expectations.
AI success is a journey, not a destination, so patience is needed. And while all the hype may lead business leaders to expect near-instant results, most organizations will take many months to get from prototype to production—with no guarantees they’ll even get there. Set realistic timelines, establish objectives up front, keep stakeholders in the loop, and remember, there are no shortcuts in real success with AI.
Putting It All Together
The playing field is poised to become a lot more competitive, and businesses that don’t deploy AI and data to help them innovate in everything they do will be at a disadvantage. Paul Daugherty, chief technology and innovation officer, Accenture
It’s not an overstatement to say that AI is fast becoming table stakes: Enterprises need to adopt now or fail in the future.
At the same time, it’s important to take the time to do things right. This means realizing not only the business challenges, but also the best technology for each task and how to implement it.
And finally, it’s important to remember that AI is not an IT or data science exercise. Your AI initiative won’t bear fruit without active and sustained executive support.
We know that legacy IT solutions are too brittle to support many AI initiatives. They simply can’t handle the speed and scale required. Massive amounts of unstructured data will contribute to the success of practical AI projects but will need to be underpinned by high-performance storage solutions that can simplify and consolidate all this data.
At NVIDIA’s GTC 2022 online event, watch Pure’s on-demand session, “Unlock the Value of Data and Accelerate Your AI/ML Initiatives,” presented by Miroslav Klivansky, Principal Data Architect for AI and Analytics at Pure.