ABN AMRO

In the second case study, the focus is on the application of AI within ABN AMRO. To gather information about the application AI within the bank, ABN AMRO was willing to make two members and the managers of the AI development team available for interviews. Next to developing AI tools, this team guides and monitor the end-users of AI closely. For instance, the agents of the customer contact center. The team members share their vision on the value of AI and the related challenges. The manager of the AI development department illustrates the application of AI with an example of a current project. This project is about an AI tool deployed in the customer contact center to assist the support agents. The AI tool supports the agents by providing multiple solutions/answers, which the agents can use as a response to the customer.

The purpose of this AI tool is to shorten the response time of agents. To refer to the statements

and experiences of these interviewees, the references [Manager 1], [Employee 4], and [Employee 5] are used.

Based on their own experiences and the shared experiences of the agents of the customer contact center, all the interviewees consider the introduction of AI in the bank as a big step forward. [Employee 4] and [Employee 5] both think that AI’s most significant benefit is cost reduction due to the automation of several processes (all answers of the written interview are present in appendix D). In the specific example of the customer contact center, the main benefit is the reduction in response time as told by [Manager 1]:

“Nou dan hebben ze toch al gauw 5/10 seconden tijd bespaard en dat dat loopt nu heel erg op.”

In this quote, [Manager 1] shares the measured results of a 5/10 seconds reduction in response time, and this reduction is still increasing. [Manager 1] also shares the following quote of a customer support agent who is experiencing unavailability of the AI tool:

“My timesaver doesn't work, can you please fix it” (personal communication, April 22, 2021).

[Manager 1] used this quote to illustrate how important the AI tool already is, just a few months after the implementation. According to [Manager 1], another potential benefit of applying AI in customer support is the ability to answer some questions autonomously. In the spoken interview with [Manager 1], he expects the AI tool to be able to answer some questions autonomously at the end of 2021.

Regarding the influence of training, [Employee 4] indicates that he has not received any training at all before starting to work with AI. Instead of being trained, [Employee 4] had to find out himself. In contrast, [Employee 5] received multiple training to develop AI tools and, more specifically, ML models. The support agents were trained and monitored in the pilot phase of the implementation. In the interview, [Manager 1] tells about the low usage threshold to use the AI tool:

“Die gebruiksdrempel is heel laag … alleen de drempel om het op een goede manier in je scherm te krijgen. Die is best wel hoog. In ons geval moest die interface ook handmatig door developers worden aangepast”.

In this quote, [Manager 1] confirms the low usage threshold and explains how developers resolved the issue regarding the display of the answer options. In other words, developers have an important role in keeping the AI tool's threshold low. Next to the agents being trained to work with the AI tool, the tool is also being trained by the agents as proudly told by [Manager 1]:

“Wij gaan daar nog een stap verder in, wij gaan de mensen ook de downvote mogelijkheid geven, dus als ze zeg maar twee of drie collega's een duimpje naar beneden geven, ja, dan gaan we de antwoord optie meteen negeren.”.

In this Dutch quote, [Manager 1] tells about the development of an upvote system, enabling the agents to upvote and downvote the suggested answers. High-rated answers are displayed as the best answer option. Low-rated answer options are automatically removed from the suggestions.

As mentioned in the first paragraph of this case study, the willingness to collaborate with AI tools is present. There are no signs of the inhibitory factor resistance as told by [Manager 1]:

“Niet perse weerstand, wel gewenning. Het is een andere manier van werken.”.

This quote describes the absence of resistance but the presence of habituation. [Manager 1]

describes in his spoken interview the first weeks of the customer support agents working together with the AI tool. After the employee getting used to the collaboration with the tool, the collaboration is going well. According to more general observations of [Employee 5] on the experience with AI, there is no resistance against the use/collaboration of AI tools if the additional value of the tool is evident. This statement is derived from the following quote of [Employee 5]:

“Ik merk weinig weerstand, zodra een AI model kan bewijzen dat het goed presteert en het ons leven makkelijker kan maken is er vaak enthousiasme.”.

[Employee 5] tells in this quote that she does not experience any resistance. Moreover, as soon as an AI model proves its value and can make the employee's work easier, the employee often shows enthusiasm. The quote of [Employee 5] is again an example of the benefit of AI to the Job experience of employees.[Employee 4] describes the general experience of working with AI tools with the following quote:

“everyone thinks it's cool.”.

[Manager 1] also mentions in his interview the coaching role managers play in preventing resistance and safeguarding the correct way of using new technology.

“Managers zullen meer moeten coachen op het accepteren en goed gebruik maken van nieuwe technologie.”

One of the other challenging factors influencing the collaboration between humans and AI is the level of reliability. The reliability of the AI tool used by the support agents in the customer contact center is not yet at the desired level. [Manager 1] emphasizes the learning process of the AI tool. The tool needs to learn from the feedback given by the agents. Sometimes a completely unrelated answer option appears. In the spoken interview, [Manager 1] shares the numbers related to the reliability:

“En nu zitten we tussen de 20 en 40 en we zien dat het steeds meer richting de 40 procent consistent beweegt”

[Manager 1] shows with this quote a percentage of correct answer options between 20 and 40 percent. In the opinion of [Manager 1], a percentage above 50 is desired. Regarding the reliability of AI tools, [Employee 4] consider the stability of the tools as a concern. In his view, the stability issues are the consequences of ABN AMRO being new in the AI field. This observation is shared by [Manager 1], [Employee 5] stresses the importance of safeguarding the quality and stability of AI in processes heavily depending on it already.

The perceived team productivity of the support agents is expected to increase based on the observations of [Manager 1]. The collaboration with the AI tool enables the agents to respond faster. This observation is illustrated by [Manager 1]. with the following quote:

“De medewerker is minder tijd kwijt aan het bijeenzoeken van informatie om het antwoord te complementeren (tel. nrs, links, mails adressen, processtappen).”

This Dutch quote describes the reduction in time an employee has to spend on collecting the correct information, like phone number, links, e-mail address, and process steps.

Next to a quicker response, the introduction of the tool should also lead to an increase in the quality of the provided answers. In the long run, the AI tool should focus on answering the more straightforward questions, and the agent should focus on the more complex issues as explained by [Manager 1]:

“Wat wij eigenlijk willen is dat de agents minder tijd kwijt zijn aan gewoon hele simpele antwoorden … dan kan het model gewoon prima even antwoorden en dat ze meer tijd vrij krijgen om echt complexe problemen te helpen oplossen.”

In this Dutch quote, [Manager 1] expresses the wish that employees spend more time on complex issues and leave the easier problem for the AI tool.

The division of tasks leads to a higher overall quality, according to [Manager 1]. In a year, [Manager 1] hopes to be able to share the results. For example, a reduced waiting time in the queue. The interview of [Manager 1] concludes with the following outlook on the contribution of AI: Eventually, the collaboration with AI leads to ABN servicing more clients with the same number of support or ABN AMRO servicing the same number of clients with fewer support agents. Regarding perceived team productivity, [Employee 4] observes a change of focus of employees to more complex processes after the deployment of AI. [Employee 5] has not made an explicit claim about productivity but stresses the benefits of AI. In her opinion, AI reduces the time of the different processes and enables processes to be executed on a large scale.

The future deployment of AI within ABN AMRO is increasing. All the interviewees expect more presence of AI in the coming years. [Employee 4] thinks the development team will develop more AI-tools/ML models, provided that the development environment is improved. He is advising the bank to allow faster deployment and testing of the different models. Regarding the AI tools deployed in customer support, [Manager 1] expects more autonomous working AI tools within five years. Finally, [Employee 5] foresee an increased deployment of AI if the quality of the AI models improves.