How I planned the strategic implementation of AI

Original article was published on Artificial Intelligence on Medium


How I planned the strategic implementation of AI

We strive to be a world class field service organization and central to that goal is customer satisfaction. With AI, I would like to automate business processes to improve productivity and quality. Most importantly, the tools should be available for everyone at their fingertips through a user-friendly interface.

My company is a leading Internet service provider, installing network equipment at customer promise and managing the service thereafter. Once the order (FSO) is received, it is relayed to an external contractor to do the installation job. The FSO is a bundle of products identified to serve a residential, small business, franchisee or a large enterprise customer.

Cost Leadership is the key focus for cost centers such as ours. Once received, these orders are required to move along the chain in a rapid manner to start generating revenue for the organization. Due to time-sensitive nature of the job, several sub-tasks are outsourced to staff augmentation agencies. These tasks are usually laborious, repetitive and boring in nature. That often leads to attrition and thereby adding to the cost of hiring and training the replacement. The objective here is to get better at what we do, keeping intact the golden triangle of cost, quality and time.

The vision is to identify the processes that not efficient and translate them into micro-tasks, that are don’t require any human judgement. These tasks are coded into bots using RPA software. Bot performance is timely, accurate and consistent. It simplifies the operational process, leaving out the ambiguity of ‘what’s next’.

Some processes can be maneuvered by a machine learning solution prior to being presented to the operational team for further review. The staff may then rely on the solution or over-ride it’s action. Iterative use of this process builds effective machine learning model that can eventually relieve the staff of that job. The staff is now able to focus on exceptional customer service, which is usually an expensive affair. The company is now well positioned to solicit feedback from internal and external stakeholders to improve its products and services. Again, NLP technology could be utilized to synthesize the incoming bulk of response to generate analytics and identify actionable items.

Current state

Like most other field service management organizations, we use technologies such as GPS, cutting-edge mobile technology, real-time data for all key business operations and synchronized customer touch points.

Due to the sophisticated nature of order-to-cash processes that is required to cater to a variety of customers, operational staff is usually burdened with managerial duties such as status checks, customer communication and third-party interactions. The high number of possible variations in the workflows makes it a complex job. The cost of operations has been increasing in the recent years.

All the data is captured in real-time and hosted in a data warehouse for analytics. A recent transformation of the business systems and processes has set-up the stage for adoption of AI. The analytics solution that is in place provides insights to the business processes and assists operations staff to propel the FSO to completion. However, it does not perform the workflow itself. Hence, the operational staff are burdened with pending jobs.

When the network device is installed by the contractor, they are required to upload pictures to confirm quality. Owing to the high volume of photos, the administrator can audit only a small percentage of the install activities. This poses a business problem with the quality of the outsourced job.

The strategy I propose is to use implement AI solutions that would automate the workflow and intelligently perform the action needed. The orders in the pipeline can now move faster and serviced by the field technician. All this in conjunction with the operational staff, who can trigger the micro-workflows ensuring that the job is done right. The augmentation of ‘machine hours’ to ‘man hours’ boosts productivity of the organization.

Proposed initiative

The strategy proposed is to assist the operational staff with multiple AI solutions that work hand-in-glove with humans using their judgment call at each step. In the end-to-end process of the order-to-cash workflow, each of the micro-processes could be investigated to find a potential candidate for RPA, Machine learning or NLP. The best-fit analysis is conducted based on the characteristics of the process. Listed below are the considerations:

1. RPA is best suited solution for repetitive, time-sensitive and laborious tasks

2. Machine learning is used where there is high volume and the ability to predict adds value to the solution

3. NLP is used for crunching large amount of text to provide meaningful insights

The AI Roadmap is illustrated in the Figure 1 below:

Proposed initiative

The strategy proposed is to assist the operational staff with multiple AI solutions that work hand-in-glove with humans using their judgment call at each step. In the end-to-end process of the order-to-cash workflow, each of the micro-processes could be investigated to find a potential candidate for RPA, Machine learning or NLP. The best-fit analysis is conducted based on the characteristics of the process. Listed below are the considerations:

1. RPA is best suited solution for repetitive, time-sensitive and laborious tasks

2. Machine learning is used where there is high volume and the ability to predict adds value to the solution

3. NLP is used for crunching large amount of text to provide meaningful insights

The AI Roadmap is illustrated in the Figure 1 below:

RPA Box:

The following process have been identified as bot tasks, that will allow the staff to move with speed and efficiency:

1. This simple task of web-scraping information from the third-party service providers is time-consuming and laborious. The bot would log into their web portals periodically and look for status changes. This is then recorded in the B2B customer portal against the corresponding account, allowing for further order progression.

2. Another area where RPA can be useful is to nudge various workforce involved in the order fulfilment chain, to get their part of the job done in a timely manner. This is accomplished by identifying the orders in jeopardy and sending follow-up emails to that person who is responsible for marching the order to the next milestone.

3. In some cases, the orders are stuck because there is pending input from customer, such as a revised quote approval. RPA can be used to keep track of the number of days of ‘no-response’ status and remind the customer periodically.

4. RPA is also used for onboarding a contractor. Contractors that are interested to work for the organization are required to send a signed copy of ‘background and drug testing’ consent form to an Inbox. The RPA agent reads the hand-written information, which is often sensitive (SSN, DOB), validates the data and marks it as acknowledged, in the system.

Machine learning box:

One of the initiatives is to harnesses the power of artificial intelligence to bring image recognition to field services. The strategy is to leverage pre-trained image classifiers — or train custom classifiers — to handle a vast array of specialized image-recognition use cases. Thousands of photos uploaded by contractors could be scanned for adherence to policy. The mobile app could proactively alert the technician for missing photos or blurred images. Once the set of photos are captured at the location, the system can review every photo and classify the image to a predefined set of categories. It would flag the instances where safety measures have been compromised or if the job quality is dissatisfactory. Also, implementing a machine learning solution handles the volume issue instantaneously.

NLP box:

A machine learning model is developed to classify the text from customer feedback, into different topics. A sentiment analysis is associated with each topic to further understand the gravity of the situation. Relevant topics are then stored and made available to multiple relevant internal organizations for further action. The synthesized data is more valuable to any department than unstructured data. NLP tools add value to the customer feedback information by providing insights that may be easily overlooked otherwise.

Benefits

When the proposed initiatives are rolled out to their full potential, it would result in freeing up the workforce from mundane jobs. The staff could then be upskilled/reskilled to explore innovative solutions. This allows the organizations to expand its line of businesses.

Technical requirements

RPA Software can be procured from Automation Anywhere, which provides the required infrastructure for bot creation and bot runner.

Google cloud platform provides machine learning models for image classification. A popular one is CNN (convolutional neural networks). It also provides NLP for use in specific industries.

Managerial or leadership requirements

With current situation in the economy, the greatest challenge is the team assembly. The main analytics and leadership role would be a “business translator,” this person should be capable of leading the initiative. The role requires the person to be an advocate of change with the domain expertise in field services. This person is responsible for aligning AI/ML with business needs.

As illustrated in Figure 1, the organization should foster cross-functional collaborations with IT, Data science and ML experts. The IT department is building in-house ML expertise by training its employees. A approach that would provide faster results is to outsource these projects to incubators.

As stated in the previous sections, the proposed initiative improves customer satisfaction. Automation usually reduces the chance and scope for errors. This results in improved relationship with suppliers and vendors.

Managerial or leadership requirements

With current situation in the economy, the greatest challenge is the team assembly. The main analytics and leadership role would be a “business translator,” this person should be capable of leading the initiative. The role requires the person to be an advocate of change with the domain expertise in field services. This person is responsible for aligning AI/ML with business needs.

As illustrated in Figure 1, the organization should foster cross-functional collaborations with IT, Data science and ML experts. The IT department is building in-house ML expertise by training its employees. A approach that would provide faster results is to outsource these projects to incubators.

As stated in the previous sections, the proposed initiative improves customer satisfaction. Automation usually reduces the chance and scope for errors. This results in improved relationship with suppliers and vendors.