We are getting more familiar by the day with Artificial intelligence and Natural language Processing applications. eCommerce Recommendation systems, Vision-based algorithms or AI-powered chatbots that can handle multiple queries at once: are all based on AI and are all examples of General AI.
In all truth, while the idea of Artificial intelligence invariably evokes images of intelligent robots, smart-assistants or even self-driving cars, these applications just scrape the surface of what where AI can be useful for.
A much less talked to -and one of the areas where AI has been proving the most powerful- is at the intersection of Financial procurement, Legal Audit and Business Operations. Here, domain-specific AI unfolds its true potential.
Automating Knowledge and Insight generation
The pace of change, and evolution of technology is forcing organization’s to engage in increased and novel business activity to maintain competitive edge. With increased business activity comes diversified contract activity — proposals to be formed, agreements to be negotiated and terms to be reviewed. Much of the work is supported by departmental stakeholders, internal analysts or with the help of external legal advisors or law firm partners.
Stripped to the barebones, this sort of operational work invariably involves 3 phases:
- Sourcing, identifying and analysing relevant information. For instance, during due diligence, legal and audit teams need to select the parts of the business that need to be audited, sourcing the contracts and financial documents that needed to be looked at;
- Organizing information into understandable schemas: once collected, information needs to be sorted and much time is spent manually fitting data into existing schemas or systems. In the insurance industry for instance, customers fill claims in a variety of formats both on paper, online. Data needs to be organized in a standardised way so it’s easier to read, digest, store into DB for later analysis.
- Communicating the information to the relevant department and coordinating the work. Once the material has been processed and the information extracted, it needs to be sent to the relevant department for evaluation.
To this date, much of the focus has been on enabling more of 3: organizational efficiency has been taking the center stage. But the greatest challenges – yet most anticipated value – is expected to come from how businesses do 1) and 2).
Businesses need to be handling documents and paperwork – only at greater volumes; parsing through the text and selecting important information – only faster, and structuring into machine-enabled, digital formats for further processing – only more automatically, and effectively.
And all of this whilst upkeeping or increasing accuracy and being able to compete at a lower cost.
One of the areas where AI has been proving the most powerful- is at the intersection of Financial procurement, Legal Audit and Business Operations.
In these areas, AI proves to be a real game changer, offering endless capabilities to improve team and business operations when it comes to speed, accuracy and responsiveness, unlocking new efficiencies and enabling new ways of working.
Think of -widely used- Document Management, Enterprise Content Management, Collaborative file storage systems that are the backbone of every organization. These systems are the infrastructure for knowledge management. Starting with files and documents, and enhancing them with AI is becoming a focus for many providers as they try to serve more needs for their client base. Actually, leveraging NLP capabilities could provide multi-directional ways of clustering and sorting data which would enable novel information flows within the knowledge management strategy.
Further, Contract Lifecycle Management technology provides lawyers with a single platform for negotiating, drafting and signing contracts – leading to substantial efficiency and cost savings. Integrating those capabilities with AI – for instance for contract review, redlining or clustering is a much needed feature and a path that some companies are starting to explore.
Similarly ERPs, Analytics and Big Data platforms, and Workflow/Automation platforms are all prone to leverage the potential of AI.
The harsh truth about AI
But how easy is it for companies to build their own AI platform or add new natural language processing capabilities to a product? The short answer is: it’s hard.
Truth is, many AI projects fail. A study by Capgemini found that only 27% of enterprise data projects are successful, while Gartner predicted that 85 percent of AI projects won’t deliver. Although open resources are widely available, building steady, scalable Artificial Intelligence solutions is by no means easy.
Lack of data scientists is one of the reasons: the competition for qualified data talent is especially fierce – QuantHub rated “data science/analytics” as the second most difficult skill set to find (after “cybersecurity”).
Secondly, the research and development phase is costly, long and requires multiple iterations. The annotation and training process, essential to any NLP application, is painstakingly time-consuming. In a survey by data labelling company Alexion, 72% of the data scientists interviewed reported that Production-Level Model confidence will require more than 100,000 labeled data items. According to another survey by DotScience, 65% of companies said it’s taking them from 7 to 18 months to move their ML models from idea to production.
Finally, even when you get it right, scaling up proves to be difficult and the transition from software engineering into production environments that scale is rarely seamless. Having steady MLOps in place is critical and extremely difficult to achieve: a ML model requires constant monitoring and fine-tuning to make sure it performs optimally and stays updated. MLOps involves creating a machine learning pipeline to automate the retraining of existing models, optimizing the process reaching full efficiency.
This again requires time, costs and a fully dedicated team.
Before you commit to developing your own AI, you need to take into account a few factors: the time for building reliable, scalable AI; the cost for platforms and human resources; the opportunity cost of not being able to focus on your core business. Is it really worth it?
Read Part Two of this blog post