Cognitive automation describes diverse ways of combining artificial intelligence (AI) and process automation capabilities to improve business outcomes.
It represents a spectrum of approaches that improve how automation can capture data, automate decision-making and scale automation. It also suggests a way of packaging AI and automation capabilities for capturing best practices, facilitating reuse or as part of an AI service app store.
What are examples of cognitive automation?
It is important to note that the term is a bit murky and not explicitly used by the major analyst firms such as Gartner, Forrester or IDC. They default to a range of other terms to describe various aspects of cognitive automation, including:
- intelligent process automation (IPA)
- intelligent business process automation
- digital process automation (DPA)
- cognitive services
Owing to this murkiness, it is essential to tease out which dimension of cognitive automation someone may be referring to in a given discussion. These can include:
- Combining intelligent data capture with process automation using things like optical character recognition (OCR), machine vision, speech recognition or natural language understanding. This is traditionally characterized as IPA or DPA.
- Automating process workflows and decisions using AI decision engines to complement or replace traditional business rules management systems or business process management systems. These autonomous enterprise capabilities, essentially, bring autonomous driving capabilities to business systems.
- Using process mining and AI tools to automate the process of identifying automation opportunities and then automatically provisioning them. Gartner calls this hyperautomation.
- Packaging up a set of services that combine AI and automation capabilities provisioned via a commercial or private app store. This is, essentially, the evolution of offerings such as Microsoft Cognitive Services.
- Differentiating how automation processes are kicked off as a more dynamic variant compared with unattended vs. attended vs. hybrid automation approaches.
How does cognitive automation work?
Cognitive automation typically refers to capabilities offered as part of a commercial software package or service customized for a particular use case. For example, an enterprise might buy an invoice-reading service for a specific industry, which would enhance the ability to consume invoices and then feed this data into common business processes in that industry.
Basic cognitive services are often customized, rather than designed from scratch. This makes it easier for business users to provision and customize cognitive automation that reflects their expertise and familiarity with the business. In practice, they may have to work with tool experts to ensure the services are resilient, are secure and address any privacy requirements.
More sophisticated autonomous cognitive services require a bit more engineering expertise. These use cases are likely to be managed by a dedicated team or automation center of excellence with an understanding of best practices for scaling autonomous systems. This can reduce the business risks such as unexpected cloud bills for systems that scaled larger than expected
What are the uses of cognitive automation?
Cognitive automation use cases include any process that could be improved using AI to capture data or automate more complex decisions. These include:
- Automatically categorizing product data from various sources into one global set of structured data.
- Copying information from differently formatted invoices into a standard format and then loading this into an accounting system.
- Automatically retrieving customer or support data in response to an ongoing service call using speech recognition and natural language understanding.
- Using AI recommendation engines to capture information about a customer's intent to streamline the customer experience.
What are the benefits of cognitive automation?
The main benefit of cognitive automation is that it helps weave unstructured data from documents, customer interactions, voice and machine vision into business workflows. Others benefits include:
- Streamlining information technology (IT) service management tasks around identifying the issue and automating incident response.
- Automating the value of existing automation by bridging the gaps between existing robotic process automation (RPA) bots, low-code applications and application programming interface integration tools.
- Automating decision-making to reduce manual decision-making, mitigate bias and speed business processes that may have stalled with human decision-makers.
- Improve the customer experience by combining RPA bots, conversational AI chatbots and virtual assistants.
What are the challenges of cognitive automation?
The biggest challenge is that cognitive automation requires customization and integration work specific to each enterprise. This is less of an issue when cognitive automation services are only used for straightforward tasks like using OCR and machine vision to automatically interpret an invoice's text and structure. More sophisticated cognitive automation that automates decision processes requires more planning, customization and ongoing iteration to see the best results.
Other specific challenges to consider include:
- Longer time to achieve positive return on investment (ROI)
- Harder to find expertise with autonomous business systems experience
- Need to vet AI algorithms for bias
- Potential for automatically scaling costs out of control
- New security issues from intelligent bots accessing a wider range of IT systems and workflows
- Potential for privacy or compliance breaches from feeding personally identifiable data into new workflows
What are the differences between RPA and cognitive automation?
There are several critical differences between cognitive automation and RPA involving how they complement human workers, the types of data they work with, project timelines and development approaches. These include:
- RPA automates repetitive actions, while cognitive automation can automate more types of processes.
- Traditional RPA only works with structured data, while cognitive automation can process unstructured data from emails, phone calls and video.
- RPA provides quick ROI, while cognitive automation requires more time to set up the infrastructure and workflows.
- RPA can provide quick tactical wins, while cognitive automation can provide a more strategic long-term advantage.
- RPA is simple to manage, while cognitive automation requires additional management overhead.
- RPA is brittle, which limits its use cases, while cognitive automation can adapt to change.
- RPA bots are explicitly programmed, while cognitive automation is better at learning the intent of a use case and adapting.