Enterprise AI leaders must consider two ways to scale up the low-code AI platform leverage

The most recent HFS surveys (December 2019) on the adoption of AI, analytics, and automation revealed that the biggest challenge enterprise leaders are facing with these Triple-A Trifecta technologies is in achieving scale.

Exhibit 1: Most of the enterprise AI & automation leaders consider scaling as the biggest challenge to achieve targeted benefits 

Source: HFS Research, “State of Integrated Automation” 2019

Sample: Global 2000 Enterprise Leaders = 317

Low-code AI development platforms have become popular since the inception of Google AutoML and Azure, and they are supported at different levels of maturity by tools platforms like Appian. They ensure ease of training and use, with little knowledge of coding, by visual programming aids. Along with accelerators, function libraries, and pre-built domain ontologies, these platforms can ensure faster scale-up to augment business’ productivity, process accuracy, service quality, and efficiency.

Low-code platforms, pre-built ontologies, and accelerators make AI accessible to businesses for faster adoption and scale

The legal team of a global high-tech and services company was considering using AI to build certain use cases for some of their critical functions, such as verifying the presence of relevant clauses like indemnity, penalty, and dispute resolution in large legal contract documents.

The team didn’t have access to AI engineering teams or trained programmers, and they had not had enough exposure to AI algorithms. If the legal team had started from scratch and needed to first understand the semantic extraction algorithms like recurrent neural networks and long-short-term-memory networks and then build the use case, even a POC of this use case would have never seen the light of day.

The legal team started toying with autoML. Within a few weeks, the team built a simple but effective solution on the low-code AI platform, with text-processing AI modules trained on the legal terms, rules, and ontologies supplied by the domain experts. The success of the initiative gave the team the confidence that AI is the science of the real and not just the art of the possible only for the geeks, by the geeks.

Why low code is more apt for AI than no-code

The difference between low-code and no-code is in the amount of coding skill required. Many RPA platforms require near-zero coding skills—users don’t even have to understand the backend code. But such no-code platforms are mostly effective for rule-based, deterministic problems with predictable and certain specific outputs. They don’t work in complex, deeply contextual, domain-specific, uncertain, judgment-based, unstructured, nondeterministic problems. These scenarios require visibility and knowledge of the code. For example, each organization has historical and labeled data for training AI modules, ontologies and semantic nets, unique feature sets, data structures, and rules and prioritization logic, such as for pricing or negotiations. While some contract clauses, such as for taxes and legal liabilities, might be similar and drawn from the same regulatory frameworks and guidelines, the business logic layer can be different.

If any technology, tool, or service partner is telling you that you can build and implement AI use cases without any customization with a completely no-code, hands-off, plug-and-play method, they are bluffing, and it’s better to call their bluff on their face, because:

1. AI needs training, testing, and validation data specific to your organization.
2. While RPA automates standard rules-based processes like closures, it is designed for success in improving efficiency, accuracy, and consistency more in transaction-based deterministic systems, such as for scenarios involving many dimensions with weights and priorities varying by situation.
3. AI use cases can be generic, but implementations have to specifically reflect how decisions and actions happen in your company.

Two ways you can augment the scale of leverage using low-code AI platforms

Consider the following ways, along with the low-code platforms, to scale up AI adoption significantly within businesses:

• Use pre-built k-bases, accelerators, pattern-bases, domain lexicons, and ontologies: Patterns and knowledge models, such as pre-identified fraud patterns, can accelerate build and adoption, by facilitating relevant reusability and k-base richness for AI modules built on low-code platforms. “Train once and infer many times” can become the norm for many AI use cases, like write-one-read-many times (WORM) is a common norm in database systems. Business teams can then keep updating the fraud pattern base, retraining the models additively or incrementally, with more data or more dimensions or by changing the feature engineering, parameters, and hyperparameter selections of models.
• Use pre-built function libraries: Libraries ensure consistency, so using standard general or domain rules and lexicons ensures the consistent use of terms and logic. This makes the inference consumption easier because the interpretation is uniform. It offers the best of both worlds—you can touch code when you feel you must for the sake of your own business logic, but you don’t have to write every line from scratch.

The Bottom Line: Leaders must consider using low-code platforms, pre-built accelerators, and domain lexicons to scale up the adoption of AI across the enterprise.

While businesses must be aware of the risks of too much generalization due to over-reusability, which can disrupt organizational uniqueness and strategic differentiation and priorities, low-code platforms and accelerators help businesses adopt AI faster. Use the low-code AI platforms as AI ice breakers to get over the rocket science fear and enable AI for the business masses while not over-generalizing and ignoring risks.