My AI versus The Company AI: How Knowledge Workers Conceptualize Forms of AI Assistance in the Workplace

Share Share Share Share Share

Individual in-depth interviews

The researchers conducted a dozen in-depth ethnographic interviews with knowledge workers in the United Kingdom and Switzerland working across industries such as financial services, marketing, design and manufacturing among others. The individual perspective pursued in the interviews allowed the researchers to explore personal narratives around worklife, past, present and future. It also allowed for deep dives into actual work-flows with each respondent, which were essential in developing our framework for assistance, which will be discussed later in this paper.

Organizational ethnographies

To complement the individual perspective from the in-depth interviews the research also consisted of participant observation in three companies in Switzerland, an apparel manufacturer and a manufacturing services company, and in France, a gas company. By attending meetings, speaking to employees and colleagues working together, the organizational part of the research complemented the individual interviews in providing the organizational perspective of work. The organizational perspective lays both in the collective and collaborative work process that most knowledge work happens within and is constituted by, but also shows the role of tools and formal structures in how work is conducted. Seeing the formal structures of work within an organization also avoided over-emphasising the role of individual agency in doing work. The tensions that individual vs collaborative working modes surface in relation to personal assistance are discussed below.

Focusing on ‘assistance’ as a way into exploring AI

As discussed above, AI is a topic regularly discussed in mass media, often communicating strong claims about its potential role in changing the future of work. Against the background of such claims, having a conversation with a respondent about their own job and the potential role of AI in it risks becoming about public narratives of AI rather than the respondent's own working reality.

To avoid this trap the research was framed around the concept of ‘assistance’ in the workplace. Assistance was consciously framed as tech-neutral and machine-human-neutral, i.e. assistance could be provided by a person or some form of technology, AI-enabled or not.

In essence, we explored instances of when people received some form help and support, and what kind of help and support they wanted or didn't want in the future. This enabled the researchers to discuss work with respondents and draw out nuances around work the respondent does themselves, work where they get assistance from other individuals and work where they get assistance from technology. Importantly, it also allowed for discussing when and where respondents would like more assistance, from either another person or technology.


Figure 1. participant places tasks on a spectrum of support preference and perceived complexity (© Google, used with permission.)

However, from an ethical perspective we did not want to obscure the nature of our enquiry. So at the end of each interview we made the idea of machine assistance more explicit and encouraged a full and frank discussion about it. These discussions were informed by the previous exploration of assistance, meaning they were rooted in the reality of the individual's work rather than existing media narratives.

Mapping workflows to reveal the reality of everyday work

Beside avoiding existing narratives overly influencing the research, there was also the difficulty of capturing the complexity of knowledge work with the limited time and methods at the disposal of the research team.

Most knowledge workers spend 40+ hours every week doing work. How is it possible capture anything tangible from such a mass of data? And how is it possible capture something beyond a superficial view of an individual's work? The solution was to dig into specific projects, processes and workflow with each respondent. By taking a significant, ongoing task the respondent was currently involved in, the researcher could explore the various workflows involved and furthermore the smaller constituent tasks making up the workflow. The result at the end of the research was that the research team could map a number of very detailed workflows across time and tools used.

For example Perry, a financial analyst based in Zurich, was responsible for a routine but multi-layered piece of work every week: updating a financial forecast for the C-Suite in his organisation. To do this he required sales data from multiple co-workers spread across Africa to be delivered on time and in the right format. Every week Perry needed to manage and fix the same inconsistencies before he could generate the forecast. To him this was a waste of time. In the framework we subsequently developed, this is ‘Peripheral’ work.


Figure 2. Workflow mapping of Perry a financial analyst. (© Google, used with permission.).


When observing and exploring everyday work a clear pattern emerged across all industries and roles. Workers days were split between a variety of activities, some of which they talked to as being core to their job, but the majority of which they talked about as peripheral.

Tina, a researcher and analyst for a finance firm, represents a typical story from the study. A typical day consisted of three hours spent on ‘real work’ and five hours on tasks she regarded as peripheral.


Figure 3. Tina spends more time on Peripheral work than Core work. (© Google, used with permission.)

Defining Core Work

“As a category manager I'm supposed to have a vision of where the category is going and what the trends are”

Louise, Category Manager, global CPG firm

‘Real work’ is the work that is core to one's job role, aspects of which are also core to one's professional identity. Core work was described as the tasks and activities that directly contribute to achieving the aims of one's job role. It is those activities that feel meaningful, that are part of your job description and that you get rewarded for. In other words, they are recognized by the employer as core to your role: it is what you are ostensibly employed to do.

Core work is often also core to your personal skill set and your professional identity, at least to the extent that you are in a job that matches your skills and experience. Thus, core work is not only core to the employer and job role, but it is also core to the individual worker as those tasks and activities that use your particular skills, where you get to use your skills and experience and where you can develop further within your professional field. As such core work is also central to the worker's professional identity and career trajectory. During interviews it was often the tasks that individual workers wanted to focus more on and do more of.

Defining Peripheral Work

“My role is about dealing with people… but every time I travel I have to waste 2 hours filling in my expenses”

Alan, Project Manager, Gas Company

A large proportion of work that is only indirectly contributing to achieving the goals of one's job. When asked respondents estimated the size of this more peripheral work to between 30% and 60% of their workday. While these tasks and activities only peripherally contribute to work goals, they are nevertheless important tasks that need to be done correctly. The risk of avoiding or delegating peripheral work can be high.

One recurring example of peripheral work was recording and reporting travel expenses. It does not contribute to the job goals of the person travelling, but is necessary for the accounting within the organization as a whole.

It also highlights a common characteristic of peripheral work, namely that what is peripheral to one person's job is central to someone else's job. In the case of travel expenses they are likely a core part of the job of someone in the accounting department of the organization.

Peripheral work, as tasks that do not directly contribute to your job goals, is also work you do not get rewarded for and rarely use your particular professional skills to do.

Davenport's model of knowledge work

Business professor Thomas Davenport is one of the leading theorists of Knowledge Work and his “classification structure for knowledge-intensive processes” (Davenport, 2005) maps broadly to our conception of core and peripheral work.

In broad terms core work maps to Davenport's concept of “interpretation / judgement” work, while peripheral work reflects “routine” work. However, Davenport's model is a more accurate mapping of knowledge workers aspirations than the reality of their core work. Often key responsibilities were routine and, in a technical sense, were therefore core. However, most workers we spoke to intended to increase the proportion of “interpretation / judgement” work that was core to their job. This became an important factor in defining how workers wanted to experience assistance at work.


Figure 4. Davenport's model of Knowledge Work (Davenport, 2005)

Using Davenport's model we developed a framework which helped us to categorise the different forms of work we were observing and how it is experienced by workers. This, in turn, mapped to our core-peripheral model, with routine work generally mapping to routine work and complex work mapping to core - with some important exceptions which related to job role.


Figure 5. Adaptation of Davenport's framework based on primary research. (© Google, used with permission.)

We then identified common pain-points using this model which helped Google teams to apply the model of assistance detailed in the next section.

Pages: 1 2 3

Leave a Reply