MAIO

• Model is trained on each new datapoint/batch as it comes in
• Primary use cases:–Very large datasets that make batch retraining difficult–Applications that require real-time responsiveness to quickly changing environments (e.g. social media)

We deploy a model and put in production If retrained model performance exceeds the production model

image

Algorithmic management describes certain labor management practices in the contemporary digital economy
• “delegation of managerial functions to algorithmic and automated systems” (Jarrahi et al., 2021, p.1)
• “a diverse set of technological tools and techniques that structure the conditions of work and remotely manage workforces”
• Term coined in 2015 to describe the managerial role played by algorithms on ride-hailing platforms
-> assigning work to drivers, optimizing pricing for services, evaluating drivers’ performance

Algorithmic management is a form of management where algorithms oversee, evaluate, and direct work. It was originally developed by organizations in the "gig" economy, such as Uber or Deliveroo.

No, it is becoming increasingly common in traditional organizations as well.

It relies on large-scale data collection and the technological surveillance of workers.

It supports automated or semi-automated decision-making processes, often with real-time responsiveness.

It refers to decisions made with a mix of algorithmic recommendations and limited human intervention.

Through metric systems that track performance indicators such as speed, efficiency, or customer ratings.

It uses rewards (e.g. bonuses, preferred shifts) and penalties (e.g. reduced visibility, deactivation) to incentivize employees.

A worker might receive fewer tasks or be deactivated from the platform if performance metrics drop.

• Removing daily commuting time
• Overhead reduction, productivity boost, and better profitability
• Societal benefits: Improving climate, liberating parents’ time
• Attracting top talent
• Better-than-expected hybrid work experiences

Hybrid work is likely to increase, driven by technological progress, high adoption of start-up companies, and investments in physical and human capital

Possible applications:
• Optimizing delivery workers’ daily routes
• Automating scheduling
• Evaluating employees through consumer-sourced rating systems
• Recommendations systems

• Fast analysis of large amount of information
• Data-driven decision making
• Smooth coordination of activities of large, disaggregated workforces
• New employment opportunities (gig economy)
• Better and cheaper consumer services
• Increased transparency and fairness in organizations
• Accurate labor forecasting and refractive surveillance

• Potential biases and discriminatory practices
• Power imbalance (transparency, information asymmetries)
• Lack of company accountability in business decisions
• Loss of human touch in leadership (dehumanizing)
• Loss of employee discretion in decision making
• Employee blamed for algorithmic decisions beyond their control
• Increased instability of work scheduling
• Imperfect measures of employee performance
• Workplace privacy issues

• “Computational techniques that leverage digital data from multiple organizational areas to reflect different facets of members' behavior” 
• With respect to people analytics, algorithmic management can support selecting, recruiting, developing, monitoring, and planning human resources

Efficient hiring: algorithms can sift through large volumes of applications and select a subset of candidates that matches predetermined criteria and will be flagged to hiring managers

Improving wellbeing: algorithms can monitor the labor force by extracting real-time information from social media such as Twitter
-> managers can better understand dynamics in the labor force and preemptively take HR measures to increase employee wellbeing

Business leaders can monitor collective sentiment in social media.
• Sentiment analysis and topic modeling can help them understand the dynamics in the labor force.
• Can facilitate decisions on the optimal way to make hybrid work arrangements
->Ex.; working in the office from Thursdays onward

Performance management: Amazon has started to use algorithms to track employee productivity and automate warnings and even trigger job terminations without managerial intervention

Enhancing employee learning and development:
• IBM uses algorithms to advise employees on suitable training to take, based on the experience of similar employees
• Vendor Quine uses previous employees’ career progression to make recommendations to employees about their optimal next career moves

Algorithmic systems as instruments of control contested between employers and employees

Direct workers by restricting (access to information or behavior) and recommending (prompting employees to make decision preferred by the choice architect and suggesting specific courses of action)

Evaluate workers by recording (finely grained behavior and statistics) and rating (using online rating and ranking, and predictive analytics)

Discipline workers by replacing (automatically and immediately firing underperforming employees and replacing them with substitute employees) and rewarding (interactively and dynamically rewarding high performing employees and gamifying rewards)

Restricting
• Can continuously and covertly restrict information available to workers
• Can interactively restrict the behavior of platform workers
• Ex: Upwork used algorithmically powered chatbot warnings reminding workers of their agreement to not work outside of the platform when workers shared email addresses or phone numbers with clients

• Can augment workers’ decisions by automatically finding patterns in the data and prescribing actions based on this
• Can bypass the heuristics workers typically use to make decisions
• Ex: Uber using personalized data (e.g., braking and acceleration speed) to analyze whether drivers drive erratically and algorithmically recommend when they might need to rest

Can track a wide range of behaviors
Can enable real-time adjustments of worker performance

Ex: Klick Health, a large Canadian healthcare consulting firm, used a ML tool to calculate the average time it took workers to complete tasks and to alert managers when projects appeared to be going offtrack

Can aggregate quantitative and qualitative data to measure work productivity and evaluate workers based on external and internal sources
Can predict future worker performance—achievement, skills, retention, etc.
Ex: one consulting firm used algorithmic rating to predict turnover intention, identifying "high-flight risk" individuals who were likely to leave the company

• Can be used to fire underperforming workers and replace them with others who may better follow managerial directives
• Can recruit on a greater scale and at the fraction of the time because workers are more interchangeable and labor is mainly digital
• Ex: Upwork workers who were regularly submitting proposals but not winning projects had their free- lance accounts closed

• Can provide rewards in real time for behaviors that comply with predefined correct behaviors
• Can use gamification principles to make the affective experience of work more positive and “fun”
• Ex. Google’s embedded the methods of game design in their day-to-day business processes : smartphone-based apps, scoreboards, and video/app game elements (badges) to promote the structure, look, and feel of a designed game with the intent of advancing employer goals

Behavior-tracking tools: products and services that uses computer-based algorithms to continuously track information about users and provide feedback based on that information

• Wearable products (e.g., smart watches, smart wristbands) and digital applications on computers and mobile devices
• Microsoft’s ‘Productivity score’ feature in its Office products: track employee behaviors across 73 metrics (e.g., contribution to shared documents/group chats, frequency of using Office tools)
• Tracking non-traditional employee metrics like their emails, social media activity, biometric data, and with whom they met and how they used their workspaces

Employees’ unwillingness to accept tracking: less intrinsically motivated, counterproductive work behaviors, false impressions of engaging in tracked activities, reducing organizational commitment

• when it is conducted solely by technology (i.e., computer algorithms) rather than by humans
• when it is experienced as informational (i.e., feedback) rather than controlling (i.e., surveillance), as intrinsic motivation improves
-> Behavior tracking by technology feels less judgmental and allows for a greater subjective sense of autonomy

Nudge: "any aspect of choice architecture that alters people's behavior in a predictable way without forbidding any options or significantly changing their economic incentives"

Individual decision-making occurs subconsciously, passively and unreflectively rather than through active, conscious deliberation 
Decisional choice context can be intentionally designed to systematically influence human decision-making in particular directions

• Algorithmic decision-guidance techniques: personalize the informational choice context through algorithmic analysis of data streams from multiple sources 
• Goal: channeling attention and decision-making in directions preferred by the ‘choice architect’ (selection optimization)
• Individuals make the decisions (as opposed to automated decision-making)
• ‘Soft’ form of design-based control, difficult for individuals to resist as they operate through subtle persuasion rather than blunt coercion

 Amazon’s warehouse workers’ wristbands: vibrate to point them in the direction of a product

Rewards and punishments on ride-hailing platforms:
• Rewards: Uber rewarding badges to incentivize their independent, autonomous drivers to work longer hours without forcing them to do so
• Punishments: “Time out” or deactivation in case drivers refuse or cancel trips (incite drivers to take on trips that are not economically advantageous)