Reinventing Jobs

1. 1Take the deconstructed task list (Step 1), the ROIP-to-strategic-value map (Step 2), and the feasible automation types and roles (Step 3), and lay them side by side before combining anything — the optimized combination is a synthesis of all four, not a fresh idea.
2. 2For each task, decide automation where it makes sense and avoid it where it doesn't; explicitly reject the temptation to insert technology into every transaction to cut costs, because that ignores where humans add organization-level value.
3. 3Reassign the automatable tasks to their automation type (RPA for data gathering, cognitive/AI for image and literature analysis, social robotics for physical procedures) and reroute the human's remaining work toward reviewing output, handling exceptions, and concierge-level judgment.
4. 4Rewrite the actual job description around the new balance: state the old job task, the automation-optimized task, and who owns each — as in the oncology table where the oncologist now decides what to delegate and handles only nonroutine cases.
5. 5Check organization-level implications: ask whether removing humans from a task destroys a value stream (e.g., assembly-line workers spotting improvements) before finalizing, and slow or abandon the change if the organization isn't ready.
6. 6Verify the combination produces the intended strategic outcome — faster cycle time, greater precision, new data for pricing — and confirm it required reinventing jobs, not merely substituting algorithms for analysts.
7. 7Trace second-order strategic shifts the new data enables (new pricing basis, new risk focus) so the reinvented process is built to capture them, not just to run the old process faster.

In the auto insurance claims process, the optimized combination starts by pairing the customer with their own cellphone camera — no insurance job involved — to photograph the damage and upload it, with a drone dispatched by a company operator only when the vehicle is remote, dangerous, or the customer is hurt. Cognitive automation, perpetually learning from thousands of daily uploads and millions of stored images, links the photo to make/model records and identifies damage location such as the bumper and right rear fender. The appraiser job is reinvented: instead of executing routine individual transactions, the appraiser reviews automation output, oversees multiple stakeholder interactions, and provides concierge-level service. The cycle shrinks from days or weeks to hours, and the quote to the repair shop is more complete and precise because it draws on thousands of similar cases, reducing overlooked damage. Critically, these gains came from reinventing several jobs and processes, not from swapping algorithms for analysts. The larger strategic payoff is that the improved claims data transforms pricing — now based on regional accident history, precise parts and labor cost for the specific vehicle, and driving style — a game-changing benefit only available if the work optimization is done correctly.

• If a task is routine, high-volume, and generates little intelligence when done manually (hundreds of low-memory transactions daily), then automate it and redirect the human to oversight and exceptions.
• If automating a task destroys organization-level value — like assembly workers who spot and communicate line improvements — then keep the human even though the machine would be faster, safer, and cheaper.
• If the automation handles typical cases well but fails on unfamiliar ones (as WFO does with atypical oncology cases), then route routine work to the machine and reserve nonroutine judgment for the human.
• If the reinvented process generates new data (damage histories, driving styles), then design the combination to capture that data for downstream strategic use, not just to run the old process faster.
• If your plan simply substitutes algorithms for analysts without redefining the surrounding jobs, then you have not optimized — reinvent the jobs and processes or the strategic outcome won't materialize.

• Inserting technology into every transaction to cut costs and errors — the tell is a design that never asks where automation makes sense versus where it doesn't.
• Treating optimization as 'rip and replace' or straight substitution of algorithms for humans — the tell is that no job descriptions changed, only headcount.
• Asking 'which jobs will automation replace?' instead of 'what is the optimal human-automation combination?' — the tell is a dead-end replacement debate, like predicting ATMs would end tellers when teller numbers rose.
• Optimizing at the job level while ignoring organization-level consequences — the tell is a faster line or process that quietly kills a source of innovation or stakeholder value.
• Stopping at efficiency and precision and missing the larger strategic shift the new data enables — the tell is a reinvented process that runs the old business model faster instead of changing pricing or risk focus.

• ▸Task (name one deconstructed task from Step 1)
• ▸ROIP / strategic value (does better performance return-on-improved-performance matter here — high/low)
• ▸Feasible automation type (RPA, cognitive/AI, social robotics, or none)
• ▸Automation role (substitute, augment, or none — what the machine does)
• ▸Human role in new combination (review output, handle exceptions, concierge service, nonroutine judgment)
• ▸Make-sense test (write why automating this task does or does not make sense)
• ▸Organization-level implication (what value is gained or lost beyond the job; go/slow/abandon)
• ▸New data or strategic byproduct (what intelligence this combination now generates for downstream use)
• ▸Reinvented job statement (one line: old task → automation-optimized task → owner)

1. 1Complete the four-step job framework first (deconstruct, assess ROIP, identify options, optimize), then treat every reinvented job as an input to organization-level design—do not stop at the single-job payoff.
2. 2Choose your direction: run outside-in (start with how automation transforms industry, strategy, and organization, then derive the work) when a top-officer big idea drives change; run inside-out when you are aggregating many job-level automation decisions upward.
3. 3For each reinvented job, map its ripples across five elements—structure, decision rights/power, accountability, information sharing, rewards, culture—asking whether each amplifies or collapses the effect (the Jenga test).
4. 4Redistribute decision rights where AI now assists front-line workers: eliminate escalation paths that existed only because supervisors held information workers lacked, and reassign accountability to match the new decision point.
5. 5Rewrite cultural norms that automation invalidates—shift 'customers come to us because only we have the information' to 'customers arrive with more information than we have and come for a trusted, collaborative relationship.'
6. 6Rebuild rewards and learning as one system: link a talent platform (e.g., Upwork) to online learning (e.g., Lynda) so workers see the price and pathway of skill upgrades, and reward optimal-close-to-qualified over fully-prepared.
7. 7Flag jobs where organization-level implications are net-negative and decide NOT to reinvent them even when the single-job case looks logical—capture this decision explicitly so cost/productivity gains don't seduce you into a tower collapse.

Pioneer Natural Resources reinvented the oil driller's job so drastically it cut costs by 25 percent in completed wells and added nearly 240 wells to the Permian Basin in 2015 without adding one new employee. That result did not come from swapping a machine for a driller; it came from deconstructing the work, letting automation replace and augment tasks, and generating new work activities—the future-state driller's activities expanded rather than disappeared. For this to hold at the organizational level, front-line workers gained access to information and AI-assisted decisions that previously required escalation up to supervisors and managers, which collapses layers of the old structure and reassigns accountability downward. The same pattern then propagated to related jobs across the natural resources value chain, showing that automation reinvents individual jobs but also systematically reinvents the relationships among them. The organizational payoff (fewer added employees, faster wells) only materialized because structure, decision rights, and information sharing were realigned to the reinvented work, not just the tasks automated.

• If a big idea comes from top officers about transforming the industry or strategy, then run outside-in and derive the required work changes from it; otherwise aggregate job-level decisions inside-out.
• If automation gives front-line workers the information a supervisor formerly monopolized, then remove the escalation layer and move the decision right and accountability to the front line.
• If a job's single-job automation case is positive but its organization-level ripple threatens culture, alignment, authority, or accountability (the tornado), then decline to reinvent that job.
• If a skill pathway's price and alignment are transparent (e.g., Java to Android via four courses, fourteen hours), then reward and route workers toward it—even if the optimal path leads them out of your organization temporarily.
• If a norm depends on information asymmetry that automation just erased, then rewrite it toward trusted collaboration before the mismatch erodes trust.

• Optimizing single jobs for cost, risk, and productivity while ignoring the organizational implications—the tell is a stack of individually 'logical' job reinventions that together destabilize authority and accountability.
• Treating automation as 'technology replaces human jobs' or 'lift and shift' outsourcing—the tell is a strategy that never even considers the work or the supporting work systems.
• Leaving structure, decision rights, and escalation paths intact after front-line workers gain AI-assisted information—the tell is workers who can now decide but still route everything upward for approval.
• Keeping obsolete cultural norms ('only we have the information customers need')—the tell is customers arriving better-informed than staff while the organization still positions itself as the sole information source.
• Rewarding only fully-certified, formally-qualified candidates while enabling skills stay locked in proprietary competency systems—the tell is unfilled roles beside workers who are optimally close but invisible because their skills aren't transparent or transportable.

• ▸Reinvented job / task cluster: name the deconstructed job being automated
• ▸Direction used: mark outside-in or inside-out and the trigger
• ▸Structure impact: which layers or units this adds, removes, or decouples
• ▸Decision rights & power shift: which decision moves, to whom, and what escalation is eliminated
• ▸Accountability reassignment: who now owns the outcome after the decision moves
• ▸Information sharing change: what information front-line workers now access that they didn't
• ▸Culture norm to rewrite: old norm invalidated + the replacement norm
• ▸Rewards & learning link: the skill pathway, its price signal, and the platform connecting learning to reward
• ▸Amplify or collapse verdict: net organizational effect and go / no-reinvent decision

1. 1Name the legacy obstacle first. Before you orchestrate anything, identify which mindset or infrastructure is now the primary obstacle to reinvention, because in the passages legacy is what blocks sustainable automation.
2. 2Build the trust precondition. Workers only share ideas about how to automate their own work when they trust leaders and the organization; explicitly commit that surfacing an automatable task will not trigger arbitrary termination.
3. 3Redistribute information and decision rights. Give front-line workers access to information and AI-assisted decisions that previously required escalation to a supervisor or manager, and rewrite the norm from 'only we have the information' to 'we offer a trusted, collaborative relationship.'
4. 4Deploy work through architecture, not job slots. Use the four-step framework (deconstruct jobs into tasks, assess job-performance-to-strategic-value, identify recombination options, optimize the human-automation combination) so reinvented work is placed against structure, processes, rewards, and people practices rather than forced into legacy job descriptions.
5. 5Build enabling skills, not just technical ones. Make continuous learning and adaptive flexibility explicit selection factors, and develop the psychological capability to accept changes in power and accountability alongside new technical hybrid capabilities.
6. 6Provide continuity without assuming constant work. Devise pathways that give workers continuity while preparing them to shift between employment, contracting, and freelancing; treat their careers as perpetual deconstruction and reinvention.
7. 7Rewire rewards to follow the reinvented value. Reward the whole team including automation designers and supporters, raise the premium on uniquely human tasks, and lower rewards on formerly high-paid but now routine tasks.
8. 8Orchestrate the ecosystem. Shift your own role from managing a self-contained organization to guiding and orchestrating an ecosystem populated with robots and AI, collaborating with workers as never before.

Take GE's reinvention of the maintenance technician job. Leaders deconstructed the technician's role into data-gathering, analysis, and diagnosis tasks plus physical execution, then applied cognitive automation via the digital twin — a data replica fed by sensors from thousands of machines (about 750,000 twins in 2017). AI now diagnoses faults, predicts maintenance needs, builds schedules, and sends recommended best practices to human technicians in the field, cutting the downtime that came from scheduling delays and learning time. For leadership, this means field technicians now receive centrally-generated, constantly-updated best practices, so leaders must build trust that technicians will still feed their field discoveries into the common learning platform rather than hoard them. It also reassigns value: the analysis-and-diagnosis task moves to automation while technicians shift to higher-value execution, which forces reward and people-practice changes. The parallel oncology case shows the strain — surgeons request more discretion because reinvented jobs require them to follow cognitive-automation best practices even when they believe their own approach is better, which is exactly the power-and-accountability shift leaders must manage.

• If workers won't share ideas about automating their own tasks, then treat trust and continuity pathways as the blocking constraint before any technology decision.
• If a reinvented task no longer fits a traditional job description or employment relationship, then source it through the work architecture (gigs, contracts, projects, tours of duty) rather than reshaping it to fit legacy structure.
• If a task becomes routine after automation, then lower its reward and shift the premium to uniquely human tasks and to the software developers and data scientists who become more pivotal.
• If the recommendation is simply substituting automation for humans to cut costs (the RPA vendor's pitch), then reject it and run the full four-step framework, because simplistic substitution produces unforeseen problems.
• If experts believe their judgment beats the automation's proven best practice, then decide the power/accountability boundary explicitly rather than leaving it to on-the-spot friction.

• Chasing technology from the tech side alone. The tell: no one has deconstructed jobs into tasks or mapped reskilling pathways — the two areas companies rated themselves most unprepared on in the Global Future of Work Survey.
• Substituting automation for humans to cut costs. The tell: the business case is framed as headcount reduction, and 'unforeseen problems' surface downstream that the framework would have anticipated.
• Reinventing the job but leaving the organization untouched. The tell: reinvented work is jammed back into old job descriptions, structures, decision rights, and reward systems, so gains stall.
• Ignoring the psychological side of reskilling. The tell: you train new technical skills but experts (like the oncology surgeons who 'bristle') resist because you never built the capability to accept changed power and accountability.
• Assuming work will be constant. The tell: continuity is promised as a stable job rather than as a pathway, and workers are unprepared to shift between employment, contracting, and freelancing.

• ▸Legacy obstacle: name the mindset or infrastructure now blocking reinvention.
• ▸Deconstructed tasks: list the component tasks and mark each substituted, augmented, or human-retained.
• ▸Trust commitment: state what you guarantee workers who surface automatable tasks.
• ▸Redistributed decision rights: name the escalation that AI now removes and the norm you are rewriting.
• ▸Star-model changes: note the required shift in structure, processes, rewards, and people practices.
• ▸Enabling-skill plan: list continuous-learning, adaptive-flexibility, and psychological-capability needs.
• ▸Continuity pathway: define the continuity offered without assuming constant work (employment/contract/freelance shifts).
• ▸Reward re-alignment: name tasks rising in value and tasks falling in value.
• ▸Ecosystem role: describe how your leadership shifts from managing to orchestrating robots, AI, and workers.

1. 1Frame the transparency ask around work evolution, not job elimination. Tell workers you want them to help identify which of their tasks could be automated or augmented, and that the goal is reinvented work, not headcount cuts. Use the ATM parable to make the point concrete: automating tasks historically expanded and evolved teller work rather than ending it.
2. 2Provide continuity commitments before asking for candor. Leaders must devise pathways that give workers continuity without assuming the work stays constant — name the retraining, reassignment, or perpetual-learning systems that exist so workers see a future beyond the task they might automate away.
3. 3Build the psychological capability to accept shifts in power and accountability. When automation gives front-line workers access to information formerly held only by supervisors, and gives nonsurgical staff data once held only by surgeons, openly acknowledge that authority is being equalized so no group feels quietly demoted.
4. 4Establish lateral trust between treatment/delivery workers and the information/technology people. Convene the collaboration explicitly — operating-room staff with database and equipment technologists, customer service with product designers — because trust between these groups is the pivot point for sharing how work should be recombined.
5. 5Import credibility into the room before automating high-authority work. Where a group traditionally held all power, recruit programmers and AI experts who also carry that group's professional qualifications so discussions about reinventing their work are trusted rather than resisted.
6. 6Reinvent metrics and rewards to reflect the new shared work, then show workers the change. Reward the whole team including automation designers, raise the premium on uniquely human tasks, and let formerly high-paid routine tasks fall in value — visible reward changes signal that candor about automating your own tasks won't be punished.
7. 7Pace the automation to the level of trust. If organization-level drawbacks (territorial conflict, loss of standing) are too serious, slow the rollout to let teams understand and acclimate rather than forcing candor before it can exist.
8. 8Model the courage yourself. Leaders share candidly what they know about how work is changing first; expecting workers to volunteer where automation could replace their tasks while leaders stay opaque destroys the reciprocity trust depends on.

At Haier, adding customer-coaching work to the call-center representative job means those reps start discovering ways to improve the product interface with Amazon and other cloud services. That raises a trust-loaded question: is generating new product-design ideas the work of customer service or of Haier's talented refrigerator designers? If leaders rush to embed customer coaching in the call-center job, they risk creating territorial battles between customer service and product design, each believing it owns the AI-with-customer work. The transparency fix is to surface the overlap openly — perhaps moving the coaching task into the product-design team, or connecting the teams laterally around 'zero distance to the customer' — so neither group feels its work was quietly transferred. Because designers must openly react to the prospect that their tasks will be automated or reassigned, leaders have to make it safe for them to say so rather than defend turf silently. The trust here is horizontal as much as vertical: workers sharing how their evolving work bleeds into another team's domain.

• If workers hear 'which jobs get replaced' instead of 'how does our work evolve,' then reframe first — candor collapses the moment the conversation reads as a layoff screen.
• If automation will equalize power between a high-authority group and others, then bring in credible, qualified insiders (e.g., programmers with surgical experience) before asking the powerful group to share how to automate their own tasks.
• If two teams both believe they own the same AI-enabled work, then slow down and design lateral accountability explicitly rather than assigning the task and triggering territorial conflict.
• If you cannot offer workers a continuity pathway, then don't yet ask them to volunteer where their tasks could be automated — you're asking for candor without cover.
• If leaders won't share candidly what they know about changing work, then don't expect workers to; transparency is reciprocal and starts at the top.

• Leading with job-replacement language. The tell: workers go quiet, hedge, or defend the necessity of every task rather than proposing what could be automated.
• Asking for candor while withholding continuity. The tell: workers describe their work as fixed and irreplaceable, protecting themselves because no pathway forward has been named.
• Automating a high-authority group's work without importing credibility. The tell: surgeons or senior experts dismiss the automation discussion as coming from people who 'don't understand the real work.'
• Rushing task reassignment across teams. The tell: territorial battles emerge — two units each claim they were meant to do the AI-enabled work, and information stops flowing between them.
• Ignoring the shift in power and accountability. The tell: workers who suddenly share authority (or lose it) resist automation not on capability grounds but because the change to their standing was never acknowledged.

• ▸Task under review: name the specific work element, not the whole job
• ▸Evolution framing: one sentence on how this work changes, not whether the person is replaced
• ▸Continuity pathway offered: the retraining/reassignment/learning path the worker can see
• ▸Power/accountability shift: who gains or loses information and authority as this automates
• ▸Lateral partners: which other teams' work this task touches, and the trust risk between them
• ▸Credibility bridge: who in the room carries the affected group's professional qualifications
• ▸Reward change: how metrics/rewards shift for human vs. routine tasks here
• ▸Pace decision: automate now, or slow to let the team acclimate — and why
• ▸Leader's own disclosure: what the leader candidly shares about their changing work first

1. 1Name the strategic outcome you are actually chasing before touching technology: cost reduction, risk reduction, quality improvement, speed, or innovation. Each implies a different automation approach, so write it down explicitly.
2. 2Map the outcome to the ROIP logic (return on improved performance) of each task. Ask whether the goal is to prevent mistakes, incrementally improve existing performance, or augment humans toward exponential/inventive value — these are three different payoff types.
3. 3Deconstruct the relevant jobs into tasks and classify each on the three dimensions (repetitive–variable, independent–interactive, physical–mental) so you can match tasks to the outcome you want.
4. 4Choose automation type by outcome: RPA for high-volume, low-complexity, repetitive tasks (speed, cost, variance reduction); cognitive automation for pattern recognition and prediction (risk reduction, quality); social robotics for physical human-collaborative work.
5. 5Trace second-order strategic outcomes, not just the single-job efficiency. Ask what new capability the improved data or process creates (e.g., new pricing models, new services) — that is where the larger strategic benefit lives.
6. 6Test organizational implications before committing. Run the outside-in check: does this work-level automation amplify strategy, or does it collapse the surrounding system like a Jenga tower?
7. 7Reinvent the human job around the outcome. Redeploy freed human time toward the tasks machines can't do (empathy, judgment, invention) so the outcome is achieved by the human-automation combination, not substitution alone.

An auto insurer targets speed, cost, and risk outcomes on claims. RPA and cognitive automation now perform damage analysis and draw on thousands of similar cases, cutting the process from days to hours and producing a more precise, complete quote that reduces the risk of overlooked damage or omitted parts and labor. But the attractive outcomes were not achieved by simply substituting algorithms for analysts — the appraiser job was reinvented to review automated output, oversee stakeholder interactions, and provide concierge-level service. The larger strategic outcome emerged one level up: instead of pricing by credit score as a risk proxy, the richer claims data lets the insurer price on the historical riskiness of driving routes, the precise cost of parts for a specific vehicle, and even driving style. That is a game-changing innovation outcome, but it only appears if the work optimization is done correctly. Combined with driverless vehicles, the same data reshapes the nature of insurance itself.

• If the goal is to prevent mistakes, then automate to reduce error/variance; if the goal is incremental improvement, then augment existing performance; if the goal is exponential value, then reconfigure jobs so humans do the inventive work.
• If a task is repetitive, independent, and physical or transactional, then RPA delivers cost/speed; if it is variable and mental, then cognitive automation supports quality and risk outcomes; if it is interactive and physical, then social robotics applies.
• If a single-job automation shows strong cost, risk, or productivity gains but collapses the surrounding organization, then do not automate that job — the organizational implication overrides the local payoff.
• If automation only speeds up an existing broken process, then stop and reengineer — don't pave the cow paths.
• If the improved process generates new data, then ask what new pricing, service, or product it enables before declaring the outcome complete.

• Chasing efficiency on single jobs while ignoring organizational implications. The tell: impressive per-job ROI decks with no view of how surrounding roles or the strategy shift.
• Jumping straight to 'what automation is possible' before deconstructing jobs and analyzing payoff. The tell: leaders imagining automation options with no task-level ROIP mapping underneath.
• Using technology to mechanize old ways of working — embedding outdated processes in software. The tell: disappointing IT results and processes merely sped up, not redesigned.
• Treating automation as substitution (algorithms replacing analysts) rather than reinvention. The tell: humans removed, but the promised quality/service outcome never materializes.
• Applying one automation approach to every outcome. The tell: RPA bolted onto tasks whose real value is empathy, collaboration, or invention, where it adds no value or destroys it.

• ▸Target strategic outcome — name one: cost / risk / quality / speed / innovation
• ▸Task deconstructed — the specific task, not the whole job
• ▸Task profile — score on repetitive–variable, independent–interactive, physical–mental
• ▸ROIP payoff type — prevent mistakes / incremental improvement / exponential value
• ▸Automation type fit — RPA / cognitive automation / social robotics, with why
• ▸Reinvented human role — what the person now does that the machine cannot
• ▸Second-order outcome — new capability, data, or product the change unlocks
• ▸Organizational implication — amplifies strategy or collapses the system (Jenga check)
• ▸Go/No-go decision — proceed, redesign first, or avoid despite local payoff

1. 1List every discrete activity the job performs, not the job title. For the oil driller, that means separate tasks like 'decide when to perform maintenance,' 'operate rig controls,' 'inspect equipment,' rather than one line called 'drilling.' Use a task inventory like O*Net as a starting reference if you lack internal data.
2. 2Attach the daily minutes each task consumes based on a real shift. The driller example uses a 720-minute (12-hour) shift as the denominator so every task's payoff and cost is measured in the same unit.
3. 3Classify each task on the three automation-compatibility dimensions: repetitive vs. variable, physical vs. mental, and independent vs. interactive. Credit-analyst data gathering is repetitive/mental; a manufacturing assembler's welding is repetitive/physical; an HR consultant's diagnosis is variable/mental.
4. 4Record where each task must physically happen — on-site versus a remote/centralized location — because that determines whether a task can shift to a control center. Driller monitoring moved to a central control room watching multiple rigs at once.
5. 5Score each task's automation compatibility separately. Never ask 'is this job automatable' — ask how compatible each deconstructed task is, since one job holds many tasks with different scores.
6. 6Map each task to one of four automation outcomes: shift to another role, augment with automation, eliminate, or create as new work. Keep the minutes balanced so the future-state shift still totals the original shift length.
7. 7Recombine the surviving and new tasks into a reinvented role, then reassign shifted/augmented tasks to the roles or crews best positioned to perform them (e.g., specialized maintenance crews, centralized analysts).

A driller's 720-minute shift was deconstructed into individual activities, each classified as physical/mental, repetitive/variable, and on-site/remote, with minutes attached. Traditionally the work was repetitive, independent, and physical, and the driller made maintenance decisions by feel — producing wide performance variation rig to rig. After deconstruction, sensors and AI moved the monitoring and maintenance-timing decisions to a centralized control center overseeing multiple rigs, and specialized crews received AI-optimized maintenance schedules. The minutes accounting showed 62 shifted to other roles, 82 augmented, 65 eliminated, and 45 new activities created — netting back to 720 in the future state. The reinvented driller role became more mental, interactive, and variable, since automation absorbed the repetitive parts and left the human for unique situations. At Pioneer Natural Resources this pattern let the company add nearly 240 Permian Basin wells in 2015 without adding one new employee and cut costs 25 percent in wells completed.

• If a task is repetitive, predictable, and rule-driven (like a credit analyst's red-flag checks), then route it to RPA — it runs up to fifteen times faster with almost no errors.
• If a task is repetitive/interactive/mental and requires constant re-optimization against changing data (like GE's maintenance diagnosis), then augment with cognitive automation and a machine-learning model rather than a fixed formula.
• If a task is physical, manual, and repetitive (like an assembler gathering, welding, moving parts), then evaluate collaborative robotics combining AI, sensors, and mobile equipment.
• If a task is variable and every situation is unique (like HR-consultant diagnosis or unusual cancer cases), then keep it human and let automation handle the typical cases so humans concentrate on exceptions.
• If a task can produce the same result from a remote location, then shift it to a centralized center that supervises many units at once instead of stationing a person on each site.

• Treating the whole job as the unit of analysis — the tell is asking 'is this job automatable,' which the passages call meaningless because each task has different compatibility.
• Mechanizing the old process instead of reinventing it — Hammer's warning: using computers to speed up existing workflows leaves fundamental performance deficiencies intact ('don't automate, obliterate').
• Skipping the minutes accounting, so you can't see that shifted-plus-augmented-plus-eliminated-plus-new must reconcile back to the original shift and can't quantify payoff.
• Deconstructing but never reconstructing — the passages note deconstruction is only the critical first step; leaving tasks scattered without recombining them into a reinvented role and reassigning shifted work leaves the transformation unfinished.
• Ignoring the organizational layer — forgetting that reinvented tasks change structure, decision rights, sourcing, and leadership; the tell is a redesigned role that still 'fits' a traditional job description and reporting line.

• ▸Task name: one discrete activity performed within the job
• ▸Minutes per shift: time this task consumes on a standard shift (state shift length)
• ▸Repetitive vs. variable: is it routine/rule-driven or unpredictable/adaptive?
• ▸Physical vs. mental: manual/dexterity-based or cognitive?
• ▸Independent vs. interactive: solo execution or requires collaboration?
• ▸Location: must be on-site, or can run remote/centralized?
• ▸Automation type fit: RPA, cognitive automation, or collaborative robotics — or none
• ▸Outcome: shift to another role / augment / eliminate / create new
• ▸Reassigned to: which role, crew, or center now owns this task

1. 1Deconstruct the job first into discrete work elements (tasks) before scoring anything—asking whether a whole job is automation-compatible is meaningless because each task scores differently.
2. 2For each task, score dimension one: place it on the repetitive–variable continuum. Ask whether it runs on predictable routines and predefined success criteria (repetitive) or demands adaptive criteria, decision rules for new situations, and innovation (variable).
3. 3Score dimension two: independent–interactive. Ask whether one person performs it alone with little communication (independent) or whether it requires collaboration, communication, empathy, and emotional intelligence with others (interactive).
4. 4Score dimension three: physical–mental. Ask whether it relies on strength and manual dexterity (physical) or on cognition, creativity, and judgment (mental).
5. 5Record the three-part profile as a combined tag (e.g., repetitive-independent-physical) and log supporting facts like frequency and time-per-day so the profile is defensible, not impressionistic.
6. 6Read the combined profile against automation type: repetitive-independent-physical points to RPA/process automation; variable-interactive-mental points to human retention or cognitive/social automation done very differently.
7. 7Only after all three dimensions are profiled, move to the payoff (ROIP) question and automation-type selection—the profile is the foundation, not the decision itself.

Deconstruct the bank teller. 'Providing requested cash' and 'verifying sufficient funds' score repetitive; 'giving cash to customers' scores physical; 'calculating cash balances' scores independent—so the cluster reads repetitive-independent-physical and is highly compatible with replacement by an ATM. By contrast, 'conversing with customers' and 'counseling those with insufficient funds' score interactive because they require empathy and emotional intelligence, 'identifying appropriate additional bank services' scores mental, and 'collaborating with product designers to improve products' scores variable—so this cluster reads variable-interactive-mental and stays with humans or is automated very differently. Desiree Dixon at Navy Federal Credit Union illustrates the interactive-mental end: understanding a military spouse handling a power of attorney while a partner is deployed is not a routine cash transaction. The lesson is that no single answer (ATM or not) fits the whole teller job—the reinvented job splits into ATM-substituted tasks, human-retained tasks, and human-plus-automation tasks.

• If a task profiles repetitive-independent-physical, then it is a prime candidate for RPA/process automation substitution (e.g., ATM cash dispensing).
• If a task profiles variable-interactive-mental, then keep it human or apply cognitive/social automation to augment rather than replace.
• If a task is repetitive and mental (pattern-heavy, high-volume analysis), then route it to RPA or cognitive automation—credit analysts' red-flag scanning runs up to fifteen times faster with almost no errors.
• If a task is variable but analytical (HR consultant diagnosing unique client problems), then default to human, but test whether cognitive automation can learn from prior engagements to handle sub-tasks.
• If a task requires empathy or emotional intelligence, then treat interactivity as decisive and preserve the human even when other dimensions look automatable.

• Scoring the whole job instead of each task—tell: you're debating 'should we replace tellers, yes or no,' a question that illuminates very little and never resolves.
• Collapsing the three dimensions into one—tell: you call a task 'automatable' without saying whether it's repetitive, independent, or physical, so you can't pick the right automation type.
• Treating repetitive as automatically physical—tell: you miss repetitive-mental tasks (credit analysis) that suit cognitive/RPA rather than mechanical automation.
• Ignoring the interactive dimension because a task looks routine—tell: you automate a customer conversation and lose the empathy and trusted relationship customers actually came for.
• Jumping straight to imagining automation options before profiling and payoff—tell: you have a jumble of tech anecdotes but no basis to say whether automation should replace, augment, or create work.

• ▸Task name: one deconstructed work element, not a whole job
• ▸Repetitive–Variable score: mark where it sits and cite the routine/criteria evidence
• ▸Independent–Interactive score: note whether collaboration, communication, or empathy is required
• ▸Physical–Mental score: note whether strength/dexterity or cognition/judgment dominates
• ▸Combined profile tag: e.g., repetitive-independent-physical
• ▸Time per day: minutes spent on this task (basis, e.g., 12-hour shift)
• ▸Supporting evidence: concrete facts justifying each score
• ▸Indicated automation type: RPA, cognitive, social, or human-retained based on the profile

1. 1Deconstruct the job into discrete work tasks before assessing any payoff — ROIP applies to tasks, not whole jobs, because a job mixes elements with different payoff patterns (e.g., a science director's 'research' vs. 'team leading').
2. 2Name the pivotal strategic goal driving the decision (agility, customer responsiveness, cost control, innovation). The ROIP of a task shifts with the goal, so fix the goal first.
3. 3For each task, plot value of performance (vertical) against level of performance (horizontal) and identify which of the four prototype curves it fits: reduce mistakes (negative value below a minimum standard), reduce variance (constant value across a range), incremental value (each improvement adds value), or exponential value (little downside, high payoff only at the top).
4. 4Test each curve with the diagnostic questions: Does poor performance damage value/brand below a standard? Does better-than-standard add nothing? Does each increment add value? Does only exemplary performance pay off big?
5. 5Map the ROIP curve to an automation intent: mistake-reduction tasks → automate to reduce risk; incremental tasks → automate to raise quality; constant-value tasks → automate to cut cost; exponential tasks → keep or augment humans where inventiveness pays.
6. 6Re-run the analysis against a different strategic goal to confirm your task priorities are goal-specific, not generic (avoid 'get the best person for every job' thinking).

Take the 1970s ATM decision. Instead of asking 'how many teller jobs can ATMs replace,' deconstruct the teller job and assign each task an ROIP curve. Receiving a withdrawal request, verifying sufficient funds, debiting the account, and counting cash all follow 'avoid mistakes' — once you clear the standard, extra precision adds nothing, so automation reduces risk. Greeting customers and referring them to other employees follow 'many different ways, same value,' so they can be automated to cut cost without harm. Detecting receptivity to and recommending additional services follows 'incremental value' — each gain in accuracy and enthusiasm adds value, so automation should raise quality here. Collaborating with product designers has 'little danger from poor performance, moderate value at medium, very high value for exemplary inventiveness' (exponential), so keep that human. The result: automating some tasks reduces risk, others increase quality, others cut cost — far more nuanced than 'replace tellers.'

• If a task damages value or brand below a minimum standard but adds nothing once cleared (verifying funds), then treat it as reduce-mistakes and automate to eliminate risk.
• If different performance levels produce the same value (greeting customers, filing before the due date), then treat it as reduce-variance and automate to cut cost, not to raise quality.
• If each performance increment adds value (recommending services, accuracy that changes outcomes), then treat it as incremental and invest automation in lifting quality.
• If poor performance rarely damages but only exemplary performance pays off big (collaborating on breakthroughs), then treat it as exponential and keep or augment humans rather than replace.
• If your strategic goal changes (innovation via research vs. via team processes), then re-score the tasks — the same task can carry a different ROIP under a different goal.

• Assessing ROIP at the job level. Tell: you can't answer 'what is better performance worth?' because the job holds too many elements with conflicting curves.
• Asking 'how many workers will automation replace?' Tell: you're treating the job as one indivisible box of tacks and missing the task-level pattern.
• Automating whichever tasks consume the most time. Tell: you cut labor cost but reduce productivity or raise risk because a time-heavy task needed a human touch.
• Assuming automation itself produces the strategic result. Tell: you skip execution and how work is organized, so the payoff never materializes.
• Applying generic performance investment ('best person for every job,' 'everyone strive for their highest'). Tell: improvements aren't targeted to any strategic goal and value gains are diffuse.

• ▸Strategic goal in focus: state the one pivotal goal driving this decision.
• ▸Deconstructed task: name a single work element pulled out of the job.
• ▸Below-standard risk: does poor performance damage value/brand? (yes/no).
• ▸Value above standard: does better-than-minimum add value, add nothing, or add only at the top?
• ▸ROIP curve: label as reduce mistakes / reduce variance / incremental / exponential.
• ▸Evidence/example: one concrete instance justifying the curve choice.
• ▸Automation intent: reduce risk / cut cost / raise quality / keep-or-augment human.
• ▸Goal-sensitivity check: would this curve change under a different strategic goal? (note it).

1. 1Deconstruct the target job into component tasks before naming any automation type; you cannot select a type against a whole job, only against a task.
2. 2Classify each task on the three compatibility dimensions: repetitive vs. variable, independent vs. interactive, physical vs. mental. Record the reading for all three per task.
3. 3Map each task's ROIP (return on improved performance): does performance pay off by avoiding mistakes (constant/negative-value), by each increment adding value (exponential), or does variation create no differential value?
4. 4Match type to the task-plus-payoff signature: RPA for repetitive, independent, rule-bound tasks; cognitive automation for variable, mental tasks where data-gathering and decision rules raise human output; social/collaborative robotics for interactive or physical tasks where a machine works alongside people.
5. 5For high-payoff variable tasks, default to augmentation (human keeps curation/improvisation/relating; automation supplies data, analysis, and decision starting points) rather than replacement.
6. 6Resist the RPA-vendor pitch of 'rip and replace to cut costs' whenever a task's ROIP is exponential or the work is interactive—substitution there destroys the value you compete on.
7. 7Reconfigure the surrounding job: redesign metrics, rewards, and selection to reflect the new human-automation combination, because types converge and the reinvented task rarely sits inside the old job boundary.

Stitch Fix deconstructs the stylist job into tasks and finds two distinct signatures. Gathering and analyzing customer data—questionnaire answers on fit, color, and jeans cut, plus web and Pinterest data—is repetitive, mental, and rule-producible, so cognitive automation handles it: the AI styling algorithm pre-selects items and even matches a compatible stylist to the customer. Curation, seeing the box as a cohesive set, improvising, and relating to humans are variable, interactive tasks with exponential payoff, so those stay human. The chief algorithms officer Eric Colson notes 'there are things humans can do much better,' so the design augments rather than replaces the stylist. The cognitive automation's output becomes the stylist's starting point, making every stylist's performance incrementally better. The result is an integrated collaboration, not a substitution—the type was selected task-by-task, not job-by-job.

• If a task is repetitive, independent, and its ROIP only rewards avoiding mistakes, then use RPA—but stop there and don't extend it to adjacent variable tasks.
• If a task is variable and mental with exponential ROIP, then use cognitive automation as augmentation: automation produces decision rules and data, the human keeps judgment and relating.
• If a task is interactive or physical and performance variation delights (or annoys) customers, then use social/collaborative robotics beside the human, not in place of them.
• If a vendor pitches automation purely as cost-cutting substitution, then re-check the task's ROIP; proceed only where payoff is constant/negative-value and the work is independent.
• If two task types touch the same person (e.g., surgeon closing incisions vs. following cognitive best-practice), then reconfigure metrics and rewards so the human-automation split is accepted, not resisted.

• Asking 'which jobs will automation replace?' instead of 'which tasks?'—the tell is a project scoped around headcount removal rather than a task inventory.
• Applying one automation type across a whole job—the tell is you never produced per-task compatibility and ROIP readings.
• Using RPA on exponential-payoff or interactive work—the tell is customer complaints like Starbucks baristas too busy with online orders to chat, meaning efficiency crowded out the value that competes.
• Treating cognitive automation as replacement rather than augmentation—the tell is humans stripped of curation/improvisation tasks they do better, and declining output quality.
• Automating the task but leaving old metrics and rewards—the tell is teams that 'bristle,' like the surgeons forced onto cognitive best-practices while pay and accountability structures haven't been reinvented.

• ▸Task name: one deconstructed task, not the whole job
• ▸Repetitive/Variable: circle which and note why
• ▸Independent/Interactive: circle which and note the human-contact element
• ▸Physical/Mental: circle which and note the dominant demand
• ▸ROIP pattern: avoid-mistakes / same-value / exponential
• ▸Candidate type: RPA / cognitive / social-collaborative robotics
• ▸Augment or replace: state which and the human tasks retained
• ▸Metrics & rewards change: what must be reinvented for the new combination

1. 1Deconstruct the job into discrete work tasks before assigning any automation role — a job is never uniformly substitutable. For a bank teller, separate cash-handling from sales and customer service; for an oil driller, separate the repetitive/physical rig operation from judgment-heavy monitoring.
2. 2Rate each task on the three characteristics: repetitive vs. variable, independent vs. interactive, physical vs. mental. These ratings pre-filter the viable automation role and type.
3. 3Assign each task an ROIP (return on improved performance) payoff pattern: does value come from reducing mistakes, reducing unhelpful variation, incrementally adding productivity, or exponentially creating new value? The payoff pattern determines whether substitute, augment, or create is worth pursuing.
4. 4Match role to task+payoff: substitute where the task is repetitive and the payoff is reducing mistakes or eliminating value-less variation; augment where a human's judgment plus machine intelligence raises existing performance; create where combining human and machine produces work that neither could do alone.
5. 5Match automation type to role and characteristic: RPA for high-volume, low-complexity, repetitive tasks; cognitive automation for complex mental tasks needing pattern recognition or language; social/collaborative robotics for physical, interactive tasks in shared space.
6. 6Reassemble the reinvented tasks into new work — expect that the human-automation combination will not fit the old job description, so redefine structure, decision rights, and sourcing (employment, contract, freelance) around it.
7. 7Re-run the framework continuously: automation shifts over time from transactional 'known unknowns' toward 'unknown unknowns,' so today's augment case can become tomorrow's create or substitute case.

Take the ATM. Asking 'how many tellers can we replace?' is naive; deconstruct teller work instead. Cash dispensing is repetitive, independent, physical with a payoff from reducing mistakes and value-less variation — so the ATM substitutes there. But sales and customer service tasks are interactive and mental, where the payoff is exponential value creation, so automation augments or frees humans rather than replacing them. The result: US tellers per branch fell from twenty in 1988 to fewer than five, yet cheaper branches let banks open 43% more urban branches, so total teller employment actually rose. The work mix moved away from repetitive transactions toward sales and relationship work machines cannot do. Looking forward, AI moves from 'known unknown' portfolio questions to 'unknown unknown' anticipation — automating even highly cognitive trading and analysis, which shifts more tasks into substitute or augment roles over time.

• If a task is repetitive, independent, and its payoff is reducing mistakes or value-less variation, then automation substitutes (use RPA for mental, social robotics for physical).
• If a task is mental and interactive and human judgment still adds value, then augment — feed the human machine intelligence (e.g., call-center sensors detecting caller emotion to serve with more empathy).
• If combining human and machine would produce work neither could do alone (e.g., AI anticipating an 'unknown unknown' portfolio reallocation), then the role is create, not substitute.
• If your only question is 'which humans get replaced,' then stop — you have skipped deconstruction and ROIP and will reach a dead-end yes/no debate.
• If cost savings tempt you to jump straight to automation type, then return to step one; you can only optimize options after deconstructing tasks and identifying payoffs.

• Treating automation as 'rip and replace' of whole jobs — the tell: your analysis outputs a headcount-reduction number instead of a task-by-task work mix.
• Jumping directly to imagining automation options (RPA, robots, AI) before deconstructing and scoring ROIP — the tell: leaders debating tools with no map of which tasks pay off.
• Assuming the work stays constant after automation — the tell: you build continuity for workers by freezing job descriptions rather than planning for perpetually upgraded work.
• Applying one automation role uniformly across a job — the tell: you conclude 'yes' or 'no' to automating tellers/drillers instead of finding different roles for cash-handling vs. service, or rig operation vs. monitoring.
• Leaving reinvented work inside legacy structures and sourcing — the tell: new human-automation work is forced into an old job title, decision rights, and full-time employment when contracting or an ecosystem would fit better.

• ▸Task: name one deconstructed work element (not the whole job).
• ▸Repetitive/Variable: rate whether it applies fixed rules or handles new circumstances.
• ▸Independent/Interactive: rate whether done solo or requires communication/empathy.
• ▸Physical/Mental: rate whether it needs dexterity/strength or cognition/judgment.
• ▸ROIP payoff: pick reduce-mistakes, reduce-variation, incremental-productivity, or exponential-create.
• ▸Automation role: choose substitute, augment, or create based on characteristics + payoff.
• ▸Automation type: choose RPA, cognitive, or social robotics matching the role.
• ▸Sourcing/structure change: note how the reinvented task should be organized or sourced.
• ▸Re-review trigger: state the event (new AI capability, cost shift) that forces re-running this row.