Sustainable scaling - taking Financial Services AI to the next level
In the high stakes game of scaling AI, who gets the first roll of the dice?
AI is poised to change the world of financial services in the 2020’s in the same way that the internet changed the world in the 2000’s. New combinations of human ingenuity and machine speed and precision are changing how financial services operate and how people work and live. The stakes have never been higher when it comes to setting up to take advantage of these imminent sweeping changes. And the time is right to do this now. Three years ago, organisations were just finding their feet with AI. In three years’ time, it will be too late and you will have been left behind. But right now, we’re in the ‘Goldilocks zone’ where AI adoption is gathering momentum; proofs of concept are becoming ubiquitous and there is a huge benefit if you define then adopt industry best practice as you introduce and scale AI into your business.  The question is - who will make the first big moves to take advantage of AI at scale?

Relationship status?
It’s complicated. Many financial services organisations have commitment issues when it comes to scaling AI. Many organisations are adopting Artificial Intelligence (AI) tools and practices that speed up the automation of basic tasks in existing workflows. For example, virtual Customer Assistants are using the latest AI tools to answer the most common questions quickly and then maximise their ability to use critical customer data to offer personalised recommendations on products, policies and pricing. Another example within Financial Services is that Machine Learning methodologies (ML) are being implemented to fight against fraud and financial crime where organisations are moving away from rule based typologies to detect suspicious activities. There are several examples where Machine Learning and AI have been successfully deployed in specific circumstances, but while developing basic applications is fairly straightforward, it gets complicated when you start to scale these solutions and deploy them across an enterprise rapidly while keeping the models current and relevant.

This last point is vital. Machine learning models rapidly become critical to the business they support after they evolve past the proof of concept and become real-world solutions. However, datasets grow, and grow fast; models need to be retrained and business needs change, all of which drive an ever increasing frequency of the model’s release cycle. This is not going to stop or get any simpler when AI applications scale. Combine this with the integration of new or external data such as those from sophisticated sensors, for example telematics in-car sensors, arriving in real time stream (e.g. 15Hz) and the complexity ramps up even more.

It is not enough to live at the surface when it comes to AI. Businesses that are able to understand the depth of requirements, the ones who are prepared to commit to a relationship with AI embedded within their business structures are the ones that successfully scale.
“AI is poised to change the world of financial services in the 2020’s in the same way that the internet changed the world in the 2000’s”
No going back

Organisations that are successfully scaling AI have accepted that traditional approaches do not deliver the results that the new world wants. These successful organisations have defined and aligned a portfolio of AI initiatives to their business strategy; they have processes, platforms and tooling in place to manage these AI initiatives from proof of concept, through to productionisation and operation with a supporting operating model and a high level of data literacy throughout the organisational structure. In short, they have rebuilt their entire ecosystem to allow for success. As highlighted before, this is the only way to success. And, there are no shortcuts but there is a successful pathway that can be followed. The way this is approached is through three key stages:
Stage 1: Proof of Value
Build your prototype and make sure that it can deliver business benefit
Stage 2: Productionise the Value
Stabilize your model; integrate it into your production ecosystem and streamline the management of it
Stage 3: Operationalise/Increase the Value
Automate everything from deployment to monitoring to integrity assessment. Build an ability to test, retrain and re-deploy your models based on real-time information These three stages are the discipline of DATA AND MACHINE LEARNING OPERATIONS.  It is similar to, and shares many traits with the DevOps and DevSecOps approaches to system development and integration, but also accommodates the non-deterministic characteristics of AI and ML applications.
Expectations are changing
The potential of AI is growing daily and with this a corporate and consumer expectation of a more personalised and expedited experience.  AI can support this demand but in the rush to implement, organisations must not forget the importance of the wider business change.  It is completely legitimate for a consumer to think “if I can get a smooth buyer journey via my online supermarket why can I not get this from my bank?” and it is up to organisations to implement the processes beyond AI that will support this.  AI on its own is not a silver bullet. Instead, AI must be seen as a vast enabler and embedded across an ecosystem - planned and incorporated in wider business strategy to have impact. 
Organisations that lag behind, stay behind
Organisations that invest in innovation are the ones that outperform those that don’t. This is compounded during times of crisis. Scaling AI is a good example of the innovation that you need to bounce back from crises quicker and stronger.
Accenture research on ‘Innovation Governance’ with companies that show a commitment to innovation governance having the greater revenue trajectories.

If you do not have scaling as a clear goal on your entire AI and ML journey, you are going to be left behind.
Shaping the vision

Seeing into the crystal ball showing the perfect future of FS and the way in which it will be changed by AI is an unrealistic, and unnecessary aspiration. What matters is that you understand the opportunities that the technology can offer you. As an industry leader you will employ specialists to make YOUR VISION of the future happen. Everything else will follow that vision, empowered by your ability to get the right people in the right place at the right time to make it happen. And when you get it right, the results speak for themselves.  We have seen that organisations that move from the proof of concept stage, to productionised AI/ML and then onto scaled AI achieve nearly 3 x the ROI than those organisations that stall at the Proof of Concept stage (Accenture, ‘Ready, Set, Scale’ 2020).
Gaining commitment from executive leadership

Scaling AI is not a data & technology only problem.  Regardless of the business problem that AI is being deployed to solve, your teams also need to be prepared to spend as much of their budget and invested effort on change and adoption—workflow redesign, behavioural change design, communication, training—as they would on the technology itself. And all of this, of course, demands the buy-in of the CEO.  But successful scalers will know that the AI buck does not stop there. Across the C-suite, executives are well aware of the urgent need to shift from one-off AI experimentation to gaining a robust organisation-wide capability that acts as a source of competitive agility and growth and they know too that they will benefit from this. We are already seeing how the AI augmented CEO is an emerging trend in the organisations with which we work and in a report from 2019, Accenture found that 84% of C-suite executives believe they must scale AI to achieve their growth objectives. Three quarters believe the failure to scale AI could put them out of business in five years! (Accenture, AI:Built to Scale, 2019). This is in line with our specific Financial Services insight - we are experiencing the “climate of now” while also seeing the need for long-term infrastructure planning so that  the uptake of AI and advanced analytics is not hobbled by technological deficiencies.
Adopting AI as the norm

This final piece in the scaling puzzle, i.e. adopting AI into your BAU, cannot be understated. However there are a number catalysts to consider. Organisations are already shifting towards new ways of working that enable faster adoption of AI, thereby faster speed to market - such as the emergence of citizen data scientists, commoditised AI, data-driven culture change, explainability by design, shift from project delivery to product management mindset etc.
“Citizen data scientists” is a term that has come to be used for employees who are not software engineers or data scientists, but have been given access to a toolkit of AI and machine learning capabilities that allows them to solve problems without the need for technical resources. However, it is not enough to have a few AI ‘enabled’ practitioners spread thinly within your organisation meaning that a wider understanding of AI and improved data literacy is required across the business in order to take greater steps further towards being an AI native organisation.  Explainability by design is a concept that has been gaining traction in the field of Machine Learning. It refers to the idea that when designing a model, a trade-off should be made between accuracy and interpretability. The concept of “commoditised AI” is that the cost of advanced AI tools is falling to the point where it is available as a service to businesses, in a similar way to how you can buy and use AWS services. Culture change is an important issue when it comes to scaling up AI adoption. A lot of the technology and concepts around AI can seem quite daunting, and it is important to remember that AI is just another tool in the business toolbox. If your executive team had, perhaps, looked into scaling AI into your business structure 18 months or even one year ago - the cost of development and time to launch may have felt not worth the investment.  The barriers to entry, cost and time to launch are not rapidly diminished. Even previously, labour and time intensive processes such as data warehouse migration can take weeks and months rather than years to implement. Now what if we were to tell you that these last few paragraphs had been written by an AI? (They weren’t but they could have been). Recent advancements in AI such as GPT3 (Generative Pre-Trained Transformer, created by OpenAI) - a deep learning model that emulates human-like text have opened up the possibility of AI functionality.  Such advancements take us from narrow Artificial Intelligence to a more general Artificial Intelligence. They take us from an imagining of AI’s potential to a realisation of its application at scale. But more importantly, the intention is to make such powerful models available for organisations (such as you and your competition).  AI has moved into a new stage in its maturity as an industry which cannot now be ignored. 
“Machine learning models rapidly become critical to the business they support after they evolve past the proof of concept and become real-world solutions”
Still hesitant to make the commitment to scaling AI?
Ultimately, AI will address a vast array of business cases, each with its own nuances, and you need to learn how to scale and move fast to do so. Only with this approach will AI impact your business in a positive way.
Industry best-practice is emerging and will significantly reduce the risk of scaling AI - meaning that stakeholders can concentrate on the bigger vision of scaling AI and how it will directly impact their business and industry rather than be concerned with the intricacies of how to build it. It is said that the best time to plant a tree was a hundred years ago. The second best time is now. The same is true for investing in scaled AI. Get ready to roll the dice. Scaling AI is no longer a gamble.  Let us show you how.