Economists have long talked about the “productivity dividend” from AI. We’re finally seeing it come to life.

AI will reshape our industries one process at a time. The tools are here. The technology is ready. The only question is—who will lead the transformation, and who will be left behind?

For those willing to experiment, adapt, and lead, the future is incredibly exciting.

Over the past three decades, the IT industry has observed significant transformations and breakthroughs that have revolutionized business operations. However, since the advent of the internet, it is argued that no development will be as impactful as the rise of Artificial Intelligence (AI). From the reduction of operational overheads to the enhancement of productivity, AI is considered the most significant productivity driver encountered.

Yet, according to the OECD, the global productivity growth rate has slowed. This raises the question of what is holding business back and how these challenges will be overcome?

Businesses have successfully navigated numerous IT obstacles to growth over the years, stemming from changes in architecture and processing capability.  While these changes may initially be met with resistance, the promise of improved productivity ultimately makes their adoption inevitable.

Many examples illustrate this, including the widespread adoption of mobile phones, Software as a Service (SaaS), and cloud computing. It would have been difficult to imagine 20 years ago that enterprise companies would be comfortable subscribing to applications shared with competitors, hosted in the cloud, and accessible from any device, anywhere in the world. 

AI Isn’t New—But It’s Entering a New Era

AI is not a novel concept. It has been employed in various forms, such as machine learning, for two decades to decrease administrative overheads in finance and document processing. This has not eliminated the need for human involvement entirely but has reduced their role to managing exceptions.

Applications utilizing AI measure the confidence level, or calculated accuracy, for each action. If this level falls below a desired target, for instance, 95% confidence, the process is routed to a human for review and correction.

The hyperscalers, namely Microsoft, AWS, and Google, have elevated the accuracy of document extraction from OCR, extractor, and machine learning techniques by introducing prebuilt AI services for extracting information from structured and semi-structured documents (such as invoices and fixed high-volume forms), and even handwritten documents.

Organizations have held differing views on using these hyperscalers for processing inbound documents due to concerns about cost, security, and flexibility.

The Game-Changers: Generative, Extractive and Agentic AI

A pivotal change has been the easy accessibility of generative AI following the launch of Chat GPT in late 2022, and the subsequent development of extractive and agentic AI. While generative AI is being utilized by many workers as a "co-pilot" for quickly generating plans, research, and content, extractive AI takes document processing to the next level, and it is agentic AI that offers substantial gains in the productivity of knowledge workers.

Extractive AI uses Natural Language Processing (NLP) to identify and extract specific information from an unstructured document accurately.  Unlike generative AI, it doesn’t create content, it only transforms complex documents into structured information to support efficient processing.

Agentic AI enables processes to operate with a degree of independence by making decisions based on incoming information. It can analyse situations and act on decisions automatically. If it is uncertain about a decision, based on established business rules, it will forward the information to a human for intervention.

So Why Isn’t Everyone Using Agentic AI?

While this may seem straightforward, the widespread adoption of agentic AI to reduce operating costs and improve customer response times faces challenges. These advanced agentic AI Large Language Models (LLMs) require significant computer processing power, data centres, and power sources.

Current power demands may make it difficult for many countries to power larger LLMs domestically, leading to data privacy and security concerns for users.

This raises questions about how to leverage the power of these large LLMs securely and economically. Will countries need to revise their privacy regulations concerning organizational use of private information in transactions crossing borders and accessible by foreign governments?

The prospect of waiting for such changes is not ideal. Alternatively, will technology evolve to reduce the size and computing power requirements of LLM models, allowing them to be hosted on secure localized servers and trained on specific use case data? Historically, technological evolution, driven by feedback from early adopters to early conservatives and broad market adoption, has often provided the answer.

China's launch of Deepseek AI demonstrates that there are methods to reduce the size and energy consumption of LLMs without compromising performance. A different trajectory is emerging, and it is believed that many small, secure, trainable, and specific LLMs will become available in the market in the coming years.

These could be highly specialized, such as estimating car repair claim costs based on images for cars in a particular state or city. The LLM would be integrated into a local business process, run on a local service, and be isolated from other systems for security.

An industry focused on maintaining and selling these specific LLMs as services to drive productivity and assist knowledge workers is expected to develop.

The Key: Orchestration of Agentic AI Services

The next challenge will be orchestrating these agentic AI services to streamline business processes. The technology and capabilities for this already exist. The primary difference will lie in the number of services being orchestrated to achieve the most effective outcome.

It is foreseen that business processes will require several AI services to achieve desired outcomes, including models for understanding incoming information, extracting information, summarizing information, reasoning and acting, generating responses, and finally, monitoring the process for potential exceptions and improvements.

Assembling all the capabilities into a single process whereby you can change out services as technology develops is the value an orchestration platform offers. 

Humans currently manage business processes using these skills daily and automating this human capability accurately and at scale within individual process activities, while becoming easier, requires experience and expertise. 

“The singularity” discussed by Ray Kurzweil in his book “Singularity is Nearer” refers to the theoretical point in the future where AI surpasses human intelligence.  Futurists are predicting singularity to occur around 2045, and until then, orchestration of AI services will be key to automating business processes.

The Time to Act is Now

Although many boards and management teams are still in the process of establishing governance and secure usage policies for AI and related services within their organizations, a significant portion of their employees are already using AI daily through various online applications, Office 365, and streaming services.

AI is becoming increasingly integrated into everyday life, and people have either willingly or unknowingly accepted the trade-off between convenience and the sharing of some personal information.

Boards and management teams should now be considering how to transform their key business processes to improve employee, customer, and operational performance.

While this may seem like an obvious necessity, many companies already have established strategies focused on expansion through acquisition, product expansion, or market reach, and may not prioritize the new opportunities presented by orchestrating AI services until compelled by competitive pressures. This delay can be detrimental, as properly managed change takes time.

Leaders need to recognize that the introduction of next-level technologies facilitates the entry of new "AI first" players into their markets, and these entrants will leverage AI. The opportune time to begin transforming legacy processes is now, before disruptive entrants and second-tier players fundamentally alter the market landscape. This undertaking is not overly complex; it requires leadership and process transformation.

The necessary tools are available and continue to improve. Partnering with vendors and consultants who can share their expertise and assist in orchestrating various AI services to continuously enhance capabilities and processes is advisable.

Where Should Organisations Start?

Organizations should prioritize transforming processes that represent the “soft underbelly” of the organization, such as those driving customer experience like claims and applications; processes that consume significant administrative resources, such as triage and invoice automation; and processes requiring expensive knowledge workers to summarize and analyse unstructured information before responding to customers. These are the processes within various industries that will be transformed by AI, altering the nature of competitive dynamics.

Public companies should evaluate the impact of these changes on their business models by considering scenarios such as a 70% reduction in customer response time and the cost of serving customers, and how this would reshape their business.

Government agencies, departments, and councils can enhance their contributions to their communities and vulnerable stakeholders by increasing productivity with existing resources. An accessible starting point is the automation of information and application ingestion into their teams.

It is not uncommon for local government councils, for instance, to have over 50 online forms, completed in multiple languages, for community service requests, with most of these forms typically emailed to the council for processing. Often, one or two administrative roles are responsible for triaging these requests to the appropriate department or individual.

With AI-driven process automation, the triage task, along with extracting necessary information from the forms, can be completed in seconds. By augmenting the process with business rules, applicants may receive a response within minutes with minimal or no human intervention. The resources currently dedicated to managing administrative tasks can be redeployed to higher-value community benefits, thereby driving productivity.

Frank Volckmar is Managing Director CANZ of TCG Process