How Organizations Can Harness AI In Their Workflows

Enterprises frequently contemplate integrating AI into their workflows, driven by the desire to increase their operational efficiency/productivity. Yet, many falter due to the absence of a well-defined framework prior to embarking on AI automation.

Following discussions with numerous C-level executives on this matter, there are some common steps that are generally followed during this process. This process was formulated into a comprehensive framework to facilitate the integration of AI into traditional workflows.

Observation: AI automation typically commences as an innovation arm within the company, dedicated to research and development (R&D). They also work to translate the innovation that has been developed through R&D for commercial viability. However, achieving success in the latter endeavor proves notably challenging.

We can now look at the framework that can be used for the incorporation of AI into an organization's workflow in detail.

STEP – 1: Job Decomposition

The first step according to the framework is to dissect each job role within the organization to create a comprehensive workflow. The workflow will encompass the individual tasks that are required to be completed in order to finish the job. The overall contextual workflow must be understood to ensure that the job decomposition process is accurate.

To illustrate this better, the example of diagnosis of brain tumor can be taken. The diagnosis of brain tumor is considered as a job in this case. In order to do the same, there are various tasks that needs to be undertaken. They are as given below:

The above workflow demonstrates the how the job i.e. diagnosis of brain tumor is dissected into various tasks to create a workflow.

STEP – 2: Evaluating Tasks for AI Compatibility

The second step according to the framework is to evaluate each task for AI compatibility. In this step, the tasks are separated into the tasks that can presently be done by AI and tasks that AI can manage only in the foreseeable future. This step can enable us to have a clearer picture of which tasks could potentially be AI automated currently.

This can be better explained using the same example of diagnosis of brain tumor. In the previous step, the workflow for this job had been created. In this step, each task that is a part of the workflow is classified as tasks that require a purely human touch, tasks that can purely AI assisted and tasks that require the collaboration of humans and AI assistance. This can be visualized as below:

STEP – 3: Workflow Analysis

The third step of the framework involves conducting a thorough workflow analysis. Here, the workflow encompasses both contextual insights (domain knowledge) and the task flow. Automating a process through AI necessitates the availability of both data and algorithms. The combination of both brings about intelligence. The data represents the contextual information indispensable for automation, which may entail leveraging existing datasets or employing innovative data collection methodologies. Subsequently, this data must undergo digitization and analysis before integration with algorithms. Unfortunately, many organizations rush into the final step of incorporating intelligence into the process without proper data collection, often resulting in failure.

To understand this step, we can look at a part of a case study that Tericsoft did for one of their clients during the pandemic.

During the COVID-19 pandemic, two primary testing methods were employed to diagnose infections: Rapid Antigen Testing and RT-PCR testing, with accuracies of approximately 72%-75% and 99%, respectively. Patient symptoms and location data were digitized through a health intake form to train an AI-ML model specifically developed for this purpose.

When the model was initially developed, it was able to achieve an accuracy of approximately 35 – 40%. To enhance accuracy, we balanced the dataset by equalizing the number of COVID-19 positive and negative cases, elevating the model's accuracy to around 60%. However, surpassing the accuracy of the Rapid Antigen Test (72%-75%) was essential in order to skip that particular step in the overall workflow to enable faster detection of COVID-19. To achieve this, we incorporated zone-based COVID-19 pattern data pointers and assigned varying weightages to each question in the health intake form, to produce science-backed biases. Additionally, we employed multi-model techniques, ultimately achieving an accuracy of 88%, which was then integrated with the algorithm. Given the project's healthcare context, we integrated explainable AI, empowering users to comprehend the rationale behind the model's outputs.

STEP – 4: Decision Making on AI implementation

The fourth step involves making decisions about the methods of implementation of AI. These decisions may entail choosing to develop the necessary AI capabilities in-house, purchasing them externally, or integrating with existing AI solutions. Such determinations are often influenced by factors such as the industry or domain within which the organization operates.

For example, the AI model for COVID-19 detection that Tericsoft built, which was quoted in the third step, required custom development from our side with constant support of subject matter experts from our client's side.

STEP – 5: Implementation of AI Framework

In the fifth step, the focus shifts to implementing the entire framework which can ensure smooth integration of AI into the existing workflows of the organization. The primary objective here is to automate repetitive and mundane tasks, all the while ensuring that humans remain in the loop to ensure optimal productivity.

Tericsoft's AI Automation framework provides an optimal and smooth transition for specific mundane job roles from human-dependent to AI-automated processes. Applying this framework across the 30+ AI projects that we have undertaken, we witnessed a realization of many AI concepts.