Meijerink & Bondarouk (2023) → The Duality of Algorithmic
Management
Core argument
- Most research treats algorithmic management (AM) as having fixed, negative effects
on workers, mainly that it restricts autonomy and destroys value.
- This paper proposes a "duality" lens instead
● AM restrains and enables autonomy and value at the same time, and it is also
shaped by workers just as much as it shapes them.
Key concepts
Algorithmic management → a system relying on machine-readable data as input,
automated processing, and decision-making or execution as output, used to
support or automate managerial decisions
Algorithm types
- Descriptive (record and relay information)
- Predictive (forecast likelihood of future events)
- Prescriptive (select and often autonomously implement a solution)
Job autonomy → the freedom workers have to control aspects of their own work
Value to workers → exchange value (monetary, e.g. pay) and use value (non-
monetary, e.g. growth, identity, enjoyment)
Algoactivism → worker resistance tactics against algorithmic control, either non-
cooperation (ignoring algorithmic suggestions) or data obfuscation (feeding the
algorithm misleading data)
The traditional, deterministic view
Algorithms restrict workers through:
- Direction (withholding information, e.g. Uber not showing drivers all ride requests)
- Evaluation (surveillance, ratings, sentiment analysis)
- Disciplining (automated rewards or punishments, e.g. deactivation).
Proposition 1
Algorithms simultaneously enable and restrain autonomy and value
Four mechanisms
1) Proxy data is incomplete, which limits workers but leaves gaps to exploit (e.g. "data
doubles", delivery riders using shortcuts not on the digital map)
, 2) Processes are opaque, which limits sensemaking but lets workers construct their
own interpretations
3) Algorithmic output can be leveraged unintentionally by workers (e.g. Uber routing
final rides towards home; workers exploiting a bug across two delivery platforms)
4) Algorithms enable value co-creation with customers, which cuts both ways (bad
reviews cost workers, but strong reputations let some workers disintermediate the
platform, e.g. Fiverr freelancers taking clients off-platform)
Proposition 2
Autonomy and value recursively shape the algorithms back
- Deliberate redesign by humans, where managers or engineers adjust the system in
response to worker behaviour (e.g. Upwork building detection for off-platform deals)
- Automated, AI-driven adjustment, where self-learning systems shift their own
weightings based on worker behaviour (e.g. inflating the importance of "gigs
completed" once workers game reputation scores)
Theoretical grounding
- Draws on Orlikowski's "duality of technology": technology is the product of human
action while simultaneously enabling and restraining that action
- This is the response to purely deterministic thinking about technology's effects.
Future research directions flagged
- Whether the enabling/restraining balance depends on algorithm type
(descriptive/predictive versus fully automated prescriptive) or task complexity
- How legal context, such as GDPR, shapes this balance
- Whether outcomes create "winners and losers" among workers
- How the balance shifts as self-learning algorithms take over more of the design work
from human engineers
Self-check questions
1) Define the duality of algorithmic management in your own words, with one concrete
example.
2) Name the four mechanisms behind Proposition 1 and give an example of each.
3) What is algoactivism, and how does it connect to Proposition 2?
4) Why do the authors argue a purely deterministic view of algorithmic management is
incomplete?
5) How does Orlikowski's duality of technology underpin this framework?
Management
Core argument
- Most research treats algorithmic management (AM) as having fixed, negative effects
on workers, mainly that it restricts autonomy and destroys value.
- This paper proposes a "duality" lens instead
● AM restrains and enables autonomy and value at the same time, and it is also
shaped by workers just as much as it shapes them.
Key concepts
Algorithmic management → a system relying on machine-readable data as input,
automated processing, and decision-making or execution as output, used to
support or automate managerial decisions
Algorithm types
- Descriptive (record and relay information)
- Predictive (forecast likelihood of future events)
- Prescriptive (select and often autonomously implement a solution)
Job autonomy → the freedom workers have to control aspects of their own work
Value to workers → exchange value (monetary, e.g. pay) and use value (non-
monetary, e.g. growth, identity, enjoyment)
Algoactivism → worker resistance tactics against algorithmic control, either non-
cooperation (ignoring algorithmic suggestions) or data obfuscation (feeding the
algorithm misleading data)
The traditional, deterministic view
Algorithms restrict workers through:
- Direction (withholding information, e.g. Uber not showing drivers all ride requests)
- Evaluation (surveillance, ratings, sentiment analysis)
- Disciplining (automated rewards or punishments, e.g. deactivation).
Proposition 1
Algorithms simultaneously enable and restrain autonomy and value
Four mechanisms
1) Proxy data is incomplete, which limits workers but leaves gaps to exploit (e.g. "data
doubles", delivery riders using shortcuts not on the digital map)
, 2) Processes are opaque, which limits sensemaking but lets workers construct their
own interpretations
3) Algorithmic output can be leveraged unintentionally by workers (e.g. Uber routing
final rides towards home; workers exploiting a bug across two delivery platforms)
4) Algorithms enable value co-creation with customers, which cuts both ways (bad
reviews cost workers, but strong reputations let some workers disintermediate the
platform, e.g. Fiverr freelancers taking clients off-platform)
Proposition 2
Autonomy and value recursively shape the algorithms back
- Deliberate redesign by humans, where managers or engineers adjust the system in
response to worker behaviour (e.g. Upwork building detection for off-platform deals)
- Automated, AI-driven adjustment, where self-learning systems shift their own
weightings based on worker behaviour (e.g. inflating the importance of "gigs
completed" once workers game reputation scores)
Theoretical grounding
- Draws on Orlikowski's "duality of technology": technology is the product of human
action while simultaneously enabling and restraining that action
- This is the response to purely deterministic thinking about technology's effects.
Future research directions flagged
- Whether the enabling/restraining balance depends on algorithm type
(descriptive/predictive versus fully automated prescriptive) or task complexity
- How legal context, such as GDPR, shapes this balance
- Whether outcomes create "winners and losers" among workers
- How the balance shifts as self-learning algorithms take over more of the design work
from human engineers
Self-check questions
1) Define the duality of algorithmic management in your own words, with one concrete
example.
2) Name the four mechanisms behind Proposition 1 and give an example of each.
3) What is algoactivism, and how does it connect to Proposition 2?
4) Why do the authors argue a purely deterministic view of algorithmic management is
incomplete?
5) How does Orlikowski's duality of technology underpin this framework?
