Generative AI in action: Corporate Case Studies

Generative AI is taking the world by storm with its human-like responses and uncanny learning potential. Because it is so exciting, there’s a lot of understandable hype surrounding these new tools and we’re devoting a lot of energy to fit them in our customized analytical solutions.

How should companies address Generative AI?

First and foremost, Generative AI is a new exciting tool that enables companies to interact in new ways with data that was previously difficult to process or extract insights from. The immediate example is text data such as Reports, Customer feedback, internal communications, or product descriptions. While it requires investment in internal or external expertise in this field, there are many use cases with potential gains that far outweigh that investment.

We can now gather, process, and use such data to extract new features to empower the analytical journey of organizations and understand qualitative correlations that, previously, were very difficult to map.

It is relevant to point out that this new technology is not a substitute for the AI models that have been developed thus far. It should be seen as an enabler to leverage qualitative data and to have users interact with data and models in a more “natural” way. There is great value to be derived by combining Generative AI with other Machine Learning Models in truly customized solutions.

Keeping this in mind, there are several use cases for these new technologies that will help shape leading companies’ priorities in the short to medium term.

Use case 1: Improve customer knowledge

Leading B2C companies have thousands of interactions and transactions with their customers that are difficult to structure and process, leading to an immense volume of untapped customer feedback. This customer feedback is invaluable to these companies and, up until now, there were limited tools to process them in a structured way to extract meaningful and actionable insights.

Generative AI enables this. First by providing the framework to read and generate new summaries of customer feedback. Then, these summaries can themselves serve as inputs to a LLM to further group them in categories by content, sentiment, tone, or relevancy.

By being able to process and categorize customer feedback, we can use this as inputs to AI models to understand what the main drivers for poor feedback are, what feedback categories lead to the most significant loss in future sales and what countermeasures have the most positive impact (e.g. refunds, free product trade-in, offer in future purchase)

Use case 2: Making customer interactions more efficient

Contact centers are key for maintaining an open line of communication with customers. However, they are also very intensive in Human Resources and have limited scalability when they rely on human-to-human interaction.

Generative AI introduces two very interesting use cases that are bound to be implemented in most contact centers around the world.

First, we can design human-like machines that will behave close to a real human and assign them to specific user stories. These machines can be created to support human training through low-cost role-playing exercises where no two interactions are the same and the model’s behaviours are very close to what trainees will encounter in real life challenges.

Then, we can introduce robots to execute or complement day to day operations. These robots are geared towards answering the most common and frequent questions posed by consumers. They lack flexibility and the human touch that, sometimes, may reduce the quality of the interaction.

With Generative AI, we can fine tune a language model for these first interactions until a human agent is, indeed, required. These models can be customized to specific interactions so that the answers produced are accurate and as close to human agents as possible.

In an intermediate use case, employees can validate and edit answers provided by an LLM. This way, we can ensure that the responses are checked by humans who are responsible for the interaction while speeding their response times.

Hence, we can easily envision three different scenarios related to AI integration.

1. There can be human centered processes where Generative AI plays little to no role and all actions are taken by humans.
2. We can augment with generative AI to train and assist human agents in their tasks while keeping humans as the main owner.
3. With full integration, most tasks would be done by AI assisted robots while human agents would only take over in very specific and fringe situations.

These improvements will have an enormous impact on reducing the cost of operation of contact centers, as well as massively improving the quality of both human and robot interactions.

Use case 3: Enrich your HR data

Human Resource departments are often dependent on human effort for going through prospective applicants, generating feedback for current employees, and managing other internal and external communications. As such, there’s obvious value in having standard processes that can be streamlined and reduce inefficiencies.

Now, it will be possible to enrich those processes by employing Language Models in two ways:

• Reading: There can be a big volume of unstructured information that reaches an HR department via emails and other channels. These models can be leveraged to have a first level of data structuring, summarization and categorization so that the team can focus on the documents that have higher priority or a higher fit probability.
  When recruiting, CVs can be processed into structured databases that are easy to filter for the best candidates. Alternatively, senior feedback can be processed into key takeaways that are easier to pass on to more junior team members
• Writing: On the other hand, HR teams also have a big volume of information that they need to relay, internally and externally. Language models can assist by generating first drafts or other suggestions.
  For recruitment purposes, rejected candidates can be contacted automatically while maintaining a degree of personalized feedback. Internally, several communications can be automatized with varying degrees of personalization. The user can provide key inputs and the remainder of the text will be automatically generated following previously defined guidelines.

Use case 4: Bring customer perspective to product organization

Another use case for big B2C companies is to be able to add features to organize and segment products by. These companies’ vast assortment means it is very time consuming to manually map categories, segments and hierarchies in their product line. As a result, the exercise of revising the product structure is left for very specific and rare moments. Additionally, adding SKUs to the product line will usually entail fitting the new products to the existing structure, which may not be ideal. As a result, product segmentations are usually very linked to the commercial structure and ignore the client perception.

Now, we can build robust frameworks to revise the product structure automatically, setting any desired frequency. Furthermore, since this new framework is based on an automated process, we can add additional information and features for an even more robust segmentation.

Thus, we can gather product descriptions, characteristics, customer reviews and even other quantitative, commercial variables to structure the product line in a much more meaningful way while being able to revise it at any moment.

Use case 5: ERP

The last use case we’re covering regards the use of Generative AI within companies that rely on a ERP for most of their processes. In these instances, the ERPs have a lot of data regarding the processes being done, who’s interacting with the ERP and when all the operations are being done.

By tapping into the ERP logs, we can now process them in a streamlined fashion so we can get two very relevant outputs:

• Diagnostics – How is the ERP being used? Are there recurring issues or delays in specific processes that should be addressed?
  Having a tool to analyze the current processes will help identifying pain-points and inefficiencies that would otherwise go unnoticed.
• Sanity checks – Do the logs match what has been mapped and designed?
  It is common for actual processes to stray from their design. Currently, we can only identify these deviations and lack of adoption through human interaction and monitoring. With Generative AI, we can build robust models to monitor real life processes and compare them with their ideal design.

Conclusion

In sum, Generative AI reshapes how companies leverage their text data and how they interact with their customers. It brings forth the prospect of improving the quality of customer interactions while reducing costs, as well as empowering and enhancing decision making processes across several teams within a company.

It is a pivotal moment for innovation and the number of use cases will multiply in the short-medium term as more research is done. Having the necessary expertise is essential to identify and implement the relevant, company-specific use cases that will provide a competitive edge.