How AI Can Help Businesses Improve Efficiencies

Total
10
Shares

Artificial intelligence and especially machine learning have been major buzzwords in the past half decade or so, and with recent advancements in AI technology, many businesses are looking for ways to implement AI to enhance their business processes and find solutions for their existing issues.

However, implementing cognitive AI initiatives can be easier said than done. On the one hand, the knowledge of many businesses surrounding AI field is still fairly low. On the other hand, plenty of technology vendors and suppliers are actively approaching businesses—especially big ones—with various implementation opportunities and viable solutions. 

Here, we will discuss all you need to know about implementing AI in businesses, and we will begin by discussing the three possible implementation types.

AI Implementations In Business: Three Different Types

When thinking about AI implementations, it’s more useful to classify AI through different business capabilities, rather than using technical classifications and technological lens. 

At least in its current state of Artificial Narrow Intelligence (ANI) or weak AI in 2020, we can differentiate AI implementation in businesses into three major categories: advanced data analysis to gain insights, smart engagement with internal (employees) and external (customers, business partners, etc.) parties, and automating business process.

1. Advanced Analytics

One of the key implementations of AI in recent years—and the one driving its rise to popularity—, is the advancements in machine learning and deep learning allowing AI algorithms to detect patterns in big data, interpret their meaning, find relationships between variables, and find actionable insights into future predictions.

In business, here are some of the common applications of using advanced machine-learning analytics:

  • Analyze the performance of advertising campaigns, improve personalizations and ad placements
  • Identify frauds in real-time (i.e. insurance claims fraud and credit fraud)
  • Predict the purchase behavior of specific customers (what products/services they are likely to buy, when, where, and how).
  • Analyze competitor’s tactics, for example how they place their ads and when they post on social media
  • More detailed modeling for businesses that require statistical models and calculations

This list is certainly non-exhaustive, as in recent years there have been countless AI implementations for analytics in business

However, it’s important to understand that cognitive, advanced analytics implementations made capable by AI machine learning are significantly different than traditional analytics, mainly in three different aspects:

  1. The detail and intensity of the data/information
  2. The analytics model is typically focused on just one specific part (or multiple small parts) of the data set
  3. The model can learn and ‘evolve’ —with its machine learning capabilities— and so its ability to analyze the data set (and future data) to gain insights and/or make future predictions gets better with time

Machine learning, and especially deep-learning implementations to analytics can, for example, recognize voice and images, and can actively and continuously curate new data to allow better analytics. 

Data curation, before AI, has always been a time-consuming and labor-intensive process. Machine learning can better—and faster— identify data in an unstructured data set across many different databases. For example, AI can identify two different data that actually belong to the same organization but appears in different, unstructured formats. 

GE, for example, implemented this analytics technology in 2017 and has saved $80 million by integrating supplier data, eliminating excess and re-negotiating contracts with the newly available (and understood) data. 

Advanced, cognitive AI analytics mainly enable (significantly) faster data processing—that will also enable various automation with the availability of real-time information, as we will discuss further below—-. 

2. Smarter Engagement

Most likely, each one of us already has access to this implementation in our smartphone—via Siri or Google Assitant—, or at our homes via Alexa or Google Home.

AI implementations in recent years have allowed voice recognition, among other implementations, that when combined with the cognitive analytics functions discussed above, can provide smarter cognitive engagement for both employees/executives (internal engagement) and customers/business partners (external engagement). 

Here are some of the common implementations in this type:

  • 24/7 chatbot offering continuous customer service that can address various user questions from simple FAQ answers to more advanced technical troubleshooting. This allows businesses to provide a 24/7 customer service that can handle several different users in real-time, which can only happen traditionally by having a lot of customer service employees.
  • Internal website(s) and smart assistants that can answer employee questions in real-time for issues like IT troubleshooting, HR policy, and so on.
  • Retailers can implement this technology for a more cognitive customer engagement service, for example by including personalization and engagement to recommend products and services with rich images and languages
  • In the health industry, various implementations have been developed to assist doctors and also to replace doctors in patient engagement and diagnostics, among other tasks. 

In recent years, there are various implementations of this cognitive engagement technology to interact with customers and employees. 53% of service companies expect to use chatbots in 2020, and 58% of customers mentioned that chatbots and voice assistants have changed their perceptions. 

For internal implementation, advanced AI avatar like Amelia has helped Becton Dickinson, one of the market leaders in the US medical technology industry, to provide internal employee support and mainly to solve internal IT-related issues. 

3. Business Automation

The most common AI implementation in businesses—and arguably the most important—, is to automate business processes. Some might consider this as a specific sub-field of AI implementation called robotic process automation (RPA), which is mostly implemented to automate financial/accounting tasks and administrative processes, which are often the most labor-intensive and time-consuming traditionally. 

It’s worth noting that RPA is much more advanced than traditional business automation implementations—for example, ‘traditional’ email scheduler—. This is mainly because, as a machine learning implementation, RPA can process data from multiple sources, ‘teach’ itself to better process future data, and become more efficient with time, just like a human executive.

Automation tasks that are recently enabled by RPA implementations include: 

  • In banking, automating the process to replace lost ATM and credit cards, which traditionally involve tedious tasks like updating customer records with new card numbers, notifying customers when the new card is ready, and so on.
  • Data migration, for example, transcribing audio data into text documents, updating customer database to include address/phone number changes, etc.
  • Data extraction—extracting information from various documents, often involving many different types of documents and files— to find useful information
  • Natural language processing and computer vision implementations to recognize specific information in big data, for example finding a specific person in hundreds of hours of CCTV footage. 

It is also worth noting that compared to the two other types of AI implementations we have discussed above, RPA is typically the most accessible. That is, easier to implement and is typically more affordable than analytics and cognitive engagement. 

On the other hand, RPA generally provides a more tangible and quicker ROI—since it can improve the efficiency of a more visible business process—. In NASA’s HR department, for example, 86% of business processes are completed without any human intervention at all.

How Companies Can Integrate AI Technology

Above, we have discussed the three main areas where we can implement AI technologies to achieve better efficiency. 

However, many organizations simply don’t know where they should start to integrate AI technologies in their current system, and some others might make wrong (and fatal) decisions.

Here are four key steps we should consider in integrating AI-based technologies to our organization:

1. Identifying the business’s objectives, capabilities, and needs

First, the business must systematically assess several factors: objectives that can be achieved through AI implementation(s), current needs, and the business’s current capabilities. We can do this by examining these key areas:

  • Assess current issues

Here, we mainly focus on figuring out which aspects/elements of the business could benefit the most from AI implementations. In most cases, this is an area of the organizations where insights and information are very valuable, but is not currently available for the business for one reason or another, such as:

  1. Difficulty to scale: the data/information does exist, but it’s difficult for the business to use it, or it might be too expensive to process—and scale—- the data. This is especially true for insights gained via third-party consultancy or adviser, where getting more data will cost a lot. 
  2. Data congestion: It’s very often that the data/information is unavailable due to the non-optimal data distribution within the organization. For example, when information is siloed within departments and not shared throughout the organization, creating a data bottleneck. 
  3. Insufficient resources: In this case, the data/information is available, but the existing human resource and technology resources are not capable to analyze it. A very common issue today with the increasing amount of data coming from digital sources—especially social media—
  • Assessing potential benefits

There might also be other problems and needs outside data analytics that might be addressed via AI technology, for example, inefficient business process might benefit from implementing AI automation.
Weigh the benefits provided by the AI implementation against investment value by asking questions like:

  • How critical the AI technology adoption is to your overall strategy?
  • Is addressing the target problem tackled by the AI technology essential?
  • How difficult, expensive, and long it will be to implement the AI solution? (We have to assess both the organizational and technical obstacles)
  • What is the potential ROI of the implementation?
  • How long we can achieve BEP?
  • Will the investment be justified?

Prioritize possible implementations based on the most value (both long-term and short-term). Also, assess whether the implementation can be a part of a bigger investment in the future (i.e., a bigger integrated AI system).

  • Finding a capable technology

Here, we assess whether the AI solutions that are considered can really tackle the existing issue. For example, which type or brand or chatbot is the best choice in fixing the existing customer service inefficiency.  We will discuss this in the next part below.

2.Understand the available options

Above, we have discussed the three different types of AI implementations in business. 

However, in practice, it can be more complicated, since, within each of the three ‘types’, there can be various different subtypes ranging from different theoretical applications different technological concepts thoroughly. Each of these subtypes offers different advantages and limitations.

For example, in RPA there is a ‘subtype’ called rule-based or rule-driven RPA, which is fairly simple and transparent but is incapable of machine learning. On the other hand, more recently there are RPAs featuring deep-learning capabilities, which is very capable of ‘learning’ from large data sets, but we can’t predict what kind of model it will create or how it builds the model. This can be an issue in businesses that value transparency—for example, in insurance or financial industry—.

Since AI as a field is still fairly young—and not too many businesses are familiar with the field—, many have made the wrong investments, wasting both time and money in the process.

So, it’s very important for companies to understand all the available technologies that might be useful for the business. This is useful so the business can determine which implementations might help in achieving the business’s objectives and address the business’s needs. Also, the business can better understand which vendor/supplier to work with and how long it will take to implement the AI technology.

On the other hand, truly understanding the implemented AI technology can help us in hiring and leveraging talents that will be responsible for it. AI-related technology talents like data scientists, data analysts, and robotics experts are still extremely rare and thus, in high-demand

Not directly related, but leveraging these talents—or outsourcing to the right external service provider(s) —- will also be a key factor in determining your business’s AI implementation.

3. Trial and error

An organization’s current capabilities to handle AI technologies aren’t always obvious, so an ‘experiment’ in the form of test or pilot project is necessary. 

These pilot projects will allow the business to test different AI technologies—and possibly several technologies at once—-. It’s important to avoid implementing pilot projects that are ‘injected’ by certain technology brands or vendors, unless their solution is really a good fit for your organization’s current situation based on the previous discussions.

It’s possible to launch several pilot projects at once. However, it can be difficult to manage them together while evaluating their performance and your team’s growth in capabilities. Consider building a dedicated AI/cognitive team to manage these several pilots if that’s going to be the case.

The implementations of AI technologies in your organization can—and should—impact existing workflows and decision-making processes. In some applications, AI might take over a huge portion of decision making from human employees. In others, it can be the other way around, where human supervision is necessary for the AI operation.

All in all, execute these pilot projects while maintaining several important principles:

  • Always consider how these AI implementations can help understand and fulfill the needs of your customers/end-users
  • Involve employees that are going to experience workflow redesign(s) due to the cognitive technology implementation. 
  • Since this is a trial and error experiment, always have a plan B, plan C, and so on.
  • Consider the capabilities of these AI implementations in the design process

End Words

While ideally, every company should explore AI cognitive implementation opportunities in 2020 and onwards, it’s important to first understand the organization’s needs, problems, and technical capabilities—both in the sense of human resource capabilities and technological capabilities—, to ensure a seamless implementation of the AI technology, and especially to make the most of the (expensive) investment.

Leave a Reply

Your email address will not be published. Required fields are marked *

You May Also Like

Non-Technical Guide to AI

The Two Ideas of AI AI in recent years has brought a lot of hype, and at the same time also brought unnecessary fears. Both of these overhype and AI-related…
View Post