At 7:58 a worker receives the order of her tasks on her phone. The system has calculated the fastest route, shaved two minutes from each order and marked every pause as a deviation. It looks efficient, but the important question is different: what did we give up to gain those minutes? Perhaps professional knowledge, room to handle an emergency, privacy about movement, or the chance to explain why an automated decision is wrong. Speed is not bad in itself. The problem begins when it becomes the only measure and conceals who pays for the saving.
In many workplaces, a decision is no longer presented as a decision. It arrives as a notification, a score or an “optimised” route. A manager may say that the system assigns shifts; human resources may say that the supplier calculates risk; and the supplier may say that the client defines the rules. This chain dilutes responsibility. Yet every result contains human choices: which objective is maximised, which data count, how much error is tolerated and who receives an exception. Understanding those choices lets a team debate the tool without accepting everything as modern or rejecting all automation.
Speed always trades something away#
A system can shorten the time required to rank applications, distribute calls, prepare routes or identify incidents. The benefit is real when it removes pointless waiting and releases time for meaningful work. But an organisation must complete the sentence: “we gain speed in exchange for…”. The exchange may involve collecting more data, standardising unlike cases, reducing conversation, giving less notice of schedules or making workers correct the system’s mistakes. If the end of that sentence is never written down, the trade still exists; it simply remains outside the accounts.
A system can shorten the time required to rank applications, distribute calls, prepare routes or identify incidents.
The first common loss is context. A metric knows that a call lasted twelve minutes but not that the caller was frightened. It records that a courier departed from a route but not that a road was closed. It detects fewer keystrokes but not that an experienced employee was teaching a new colleague. When speed rewards only what can be counted, essential but difficult-to-measure activities look inefficient. The team quickly learns the lesson: stop helping, avoid complex cases or work for the score. The indicator improves while the real service becomes poorer.
Autonomy is also surrendered. A good professional does not apply a rule blindly: they interpret, prioritise and respond to exceptions. If every step is prescribed by an application and every departure requires justification, that judgement atrophies. The consequence is more than low morale. During a breakdown, emergency or novel case, the team may no longer be practised at deciding. Automating repetitive work should expand human capability, not turn responsible people into operators who cannot stop an obviously mistaken instruction.
The third currency is privacy. To refine predictions, some tools request constant location, keyboard activity, images, voice, connection times or behavioural patterns. Technical availability does not make data necessary. The questions must be asked before collection: which legitimate decision needs this item, is there a less intrusive alternative, how long will it be retained, who can see it, and could it later influence another purpose? The Spanish Data Protection Agency’s guide to data protection in employment relationships explains principles of information, necessity and proportionality relevant to this setting.
How to see the cost when the dashboard says everything improved#
Average productivity is not enough. Distribution matters. Saving one hundred hours for the company may create five hundred small interruptions across the workforce. Faster delivery may depend on late schedule changes for families arranging care. Automated classification may reduce administrative work while increasing the time an individual spends proving that a data item was wrong. Measuring only the central process makes displaced costs invisible. Evaluation should therefore include correction time, complaints, turnover, absence, incidents, shift changes and the experience of different groups.
A simple test can use one real working week. Record every significant automated decision: allocation, scoring, monitoring, access, scheduling or penalty. For each, identify the promised benefit, the person carrying the risk and the route to correction. Then speak to people at the edges: new staff, disabled workers, carers, night-shift teams, self-employed contractors and people working in another language. The aim is not a perfectly representative opinion survey. It is to discover exceptions the design failed to anticipate.
Hidden labour is another signal. Employees may clean data, correct orders, answer false alerts or explain decisions they do not understand to customers. Such tasks do not appear in a sales demonstration and often fall on jobs with the least bargaining power. They must be counted as a cost of the system. Asking “what new task appeared after we introduced the tool?” reveals more than comparing the duration of the automated operation alone. If a machine saves five minutes but creates eight minutes of checking, it has not accelerated the process; it has moved the clock.
Pressure can also move into the body. Countdown timers, targets recalculated in real time and continuous alerts can prevent breaks or intensify the feeling of surveillance. Occupational risk management should consider mental workload, predictability and a genuine ability to disconnect. An organisation need not wait for a collective diagnosis before acting. If errors rise at the end of shifts, breaks are postponed or people avoid drinking water so they do not deviate from a route, there is an operational warning that deserves immediate correction.
Rights and responsibilities do not disappear when software is purchased#
In Spain, Article 64.4(d) of the Workers’ Statute published by the BOE recognises legal worker representatives’ right to information about the parameters, rules and instructions underpinning algorithms or artificial-intelligence systems that affect decisions capable of influencing working conditions. This does not amount to handing over source code or settle every case, but it requires more than “the system decided”. Purpose, relevant factors and consequences must be explainable in language that permits meaningful intervention.
The European Union Artificial Intelligence Act classifies certain systems used in employment, recruitment and worker management as high-risk. Duties depend on the system and implementation timetable, but the practical principle is already valuable: document risk, data quality, human oversight, logs and operation. The European directive on platform work also addresses algorithmic management, transparency and human review. A company should establish which rules apply with specialist advice rather than copying a general conclusion from the internet.
Buying a tool does not transfer responsibility. The customer organisation defines its use, who receives the output and which consequences follow. It must require sufficient information from the supplier, test the system on its own cases and retain authority to suspend it. A contractual clause does not help an affected worker if nobody inside the organisation can interpret an alert or correct a record. The operational owner needs time, access and real power; if that person can only open a supplier ticket, human oversight is decorative.
A method for deciding whether speed is worth the price#
First, define the problem without mentioning technology. “Schedules take three days to complete” is better than “we need artificial intelligence”. Then set an objective with constraints: shorten planning time without reducing notice, increasing involuntary changes or collecting location outside working hours. Those constraints are requirements, not aspirations. If a supplier cannot measure or respect them, the solution is not ready for that use.
Second, establish a baseline. For several weeks, measure total time, quality, errors, administrative burden, complaints and differences between groups. Record who currently decides and how a mistake is corrected. Without this picture, any future improvement may be attributed to the system even if it came from additional staff, lower demand or team learning. The comparison must include the time of affected people, not only management time.
Third, run a small and reversible trial. Include ordinary and difficult cases: missing data, disability, connection loss, a family emergency, similar identities or atypical performance. Before the trial starts, agree who can stop a decision, how quickly they respond and what happens during review. No substantial penalty, exclusion or loss should be executed when the appeal route takes longer than the consequence itself.
Fourth, provide a useful explanation. It should state which decision was made, what information influenced it, which data may be wrong, who will review it and the applicable deadline. An explanation need not disclose trade secrets, but it must let a person provide context. “Below threshold” is inadequate when nobody identifies the threshold, its purpose or how to correct a false input. Access to someone with authority is part of the system, not an optional support service.
Fifth, set an expiry and review point. A tool that worked with one team, level of demand and set of rules may degrade when conditions change. Every deployment needs a review date, withdrawal indicators and an accountable owner. If complaints rise, group disparities widen or new uses of data appear, pause and reassess. Money already invested is a poor reason to preserve a system while harm continues.
What a worker or family can do#
When a decision appears automated, document facts without trying to guess the algorithm. Keep the communication, date, result and concrete effect. Ask in writing: “Which criterion was applied, which data were used and who can review this decision?” If there is a union representative or works council, share a short chronology. Important matters may require employment or data-protection advice. The strategy is to turn a diffuse concern into a verifiable request without disclosing more personal information than necessary.
Families can be particularly helpful to young people in a first job. Instead of resolving a dispute for them, they can rehearse questions, organise documents and distinguish urgency from disagreement. If a shift changes, record when notice arrived; if a score falls, ask which factor changed; if an application requires permissions, ask for the purpose. Boundaries also deserve discussion: do not share passwords, do not install workplace software on family devices without understanding its permissions and do not accept surveillance that invades private time as normal.
Collective conversation matters because many costs look like individual failures. One person thinks they work slowly; another believes they do not understand the application; a third conceals skipped breaks. Comparing experiences may reveal one badly designed common rule. The organisation should maintain a channel without retaliation for reporting patterns, not only isolated incidents. Reviewing complaints together can show that the problem is not an exception but the criterion itself.
Speed that is genuinely worthwhile#
Some automations are valuable: they avoid entering data twice, warn about incompatibilities, reduce waiting or help adapt a job. The difference lies in the design of power. A responsible tool makes its purpose visible, uses minimum data, preserves professional discretion, keeps a human exit and accepts suspension. Success is not producing more decisions per minute. It is improving work without treating the people who perform it as disposable.
Before celebrating a faster dashboard, ask a concrete final question: who gained time, who lost control and what evidence demonstrates both? If the saving coexists with predictable schedules, privacy, learning and effective appeal, there is probably progress. If it requires silence, surveillance and obedience to inexplicable results, the organisation has not bought efficiency; it has financed a human debt that will later appear as errors, turnover and harm. Writing down that exchange is the first step towards choosing a speed compatible with dignity.
