The question organisations should be asking isn't "will AI replace our people?" It's "which functions in our operation genuinely require a person — and which don't?" The answer changes everything about how you approach automation investment.
There is a version of the AI and workforce conversation that is not useful. It is the version that asks "how many jobs will AI eliminate?" and responds with either a number or a reassurance. Neither is actionable. Neither helps an organisation — or a society — make better decisions about how to deploy AI and how to support the people affected by those decisions.
This paper argues for a different framing: functions, not jobs. Not "will AI eliminate the Customer Service Coordinator role?" but "which functions that a Customer Service Coordinator currently performs are genuinely automatable, which require human expertise, and which require a combination of both?" That question has a specific answer for every function in your organisation. And working through those answers — systematically, honestly, with real operational data — is the foundation of a workforce and AI strategy that is both practically useful and genuinely fair to the people involved.
The problem with the job-level framing
When organisations think about AI and workforce at the job level, they make two predictable errors. The first is overestimation — assuming that because AI can perform some tasks a person performs, the whole role is replaceable. The second is underestimation — dismissing AI relevance because the role clearly requires human judgment in its most complex moments, while ignoring the significant volume of structured administrative work that surrounds those moments.
Both errors are expensive. The first creates unnecessary workforce anxiety and — where it drives actual headcount decisions — eliminates human capability the organisation actually needs. The second leaves significant administrative burden unremedied, consuming the capacity of people who could be doing higher-value work.
"Most roles in regulated service organisations contain a mix of functions — some that require human expertise, some that are genuinely automatable, and some that work best as a collaboration between a person and a system. The starting point is separating them clearly."
The function-level analysis produces more useful answers because it maps what actually happens in an organisation. A Customer Service Coordinator in an aged care organisation might perform eight distinct functions across a typical week. Two of them — handling complex welfare concerns and managing difficult family conversations — require empathy, judgment, and relational context that no AI system can replicate. Three of them — after-hours call routing, shift cancellation capture, and co-contribution enquiry handling — are structured, rules-based, and high-volume enough that they are clear automation candidates. The remaining three sit somewhere in between, benefiting from AI assistance but requiring human oversight and occasional escalation.
The job doesn't change. The mix of functions within it does — and so does the time available for the work that actually requires a person.
How to map functions in a business unit
The function mapping process is the starting point for any serious workforce and AI strategy. It is also the most frequently skipped step — organisations jump from "we should use AI" to "here is a platform" without the analysis that would tell them whether the platform fits the work.
A structured function mapping engagement for a single business unit typically produces three things:
- A complete function inventory — every distinct function performed in the unit, described at a level of specificity that allows automation assessment. Not "customer service" but "responding to after-hours welfare calls," "processing shift cancellations," "generating co-contribution statements," "managing escalated complaints."
- A time allocation estimate — how much of the team's collective capacity is consumed by each function. This is the most important and most frequently absent input to automation prioritisation. Without it, you are selecting automation targets by intuition rather than evidence.
- An automation disposition — for each function, an honest assessment of whether it is a full automation candidate, a partial automation candidate (AI-assisted with human oversight), or a human-only function. The criteria: volume and frequency, rules-based suitability, data availability, and the consequences of an error in automated handling.
A representative function map: aged care customer service
The table below is a representative function map for a customer service function in a mid-sized aged care provider. The time allocations and automation dispositions are illustrative; the actual figures for any specific organisation will differ. The point is the structure of the analysis, not the specific numbers.
| Function | % of team time | Disposition | Notes |
|---|---|---|---|
| After-hours call handling | 18% | Full | High volume, rules-based, structured capture. Sophie handles this end-to-end with SIRS-aware escalation. |
| Shift cancellation capture + routing | 12% | Full | Structured workflow, predictable inputs, clear routing rules. Automation with human review for complex cases. |
| Co-contribution enquiry handling | 10% | Full | Rules-based calculation and explanation. High volume, low variation. Clear automation candidate. |
| Appointment confirmation + rescheduling | 15% | Partial | Routine confirmations automatable; complex rescheduling or preference changes require human judgment. |
| Inbound service enquiries (daytime) | 20% | Partial | Straightforward enquiries (status, schedule, billing) automatable; complex or emotionally sensitive calls require a person. |
| Complaint handling and escalation | 8% | Human | Requires empathy, judgment, and relational accountability. AI can assist with documentation; human must lead. |
| Welfare concern management | 10% | Human | Clinical and welfare judgment required. AI can capture and alert; human must assess and act. |
| Family liaison and relationship management | 7% | Human | Relational context and continuity essential. Not an automation candidate. |
This analysis shows that approximately 40% of the team's time is consumed by functions that are full automation candidates — high volume, rules-based, structured. A further 35% is in partial automation territory. Only 25% is genuinely human-only work. The implication is not that 75% of the team should be replaced — it is that the team currently has only 25% of its capacity available for the work that actually requires a person.
The cost picture changes when you work at the function level
The financial case for automation looks very different at the function level than at the job level. At the job level, the comparison is typically "cost to hire versus cost to automate a whole role." At the function level, the comparison is "cost of the time currently consumed by this function versus cost to automate it" — which is a much more tractable calculation and produces clearer, more defensible numbers.
The maths — a representative example
Illustrative figures only. Actual costs vary significantly by organisation size, call volume, and specific function mix. These numbers are intended to show the structure of the calculation, not to provide a precise estimate.
The net position is not "eliminate headcount and pocket the difference." It is "redirect the capacity that automation creates to the work that genuinely requires a person." In a regulated service environment under workforce pressure, that capacity reallocation is valuable — it means the complaint handling, the welfare concern management, and the family liaison work gets more attention, not less. Staff spend more of their time on the functions they were hired for and less on functions that a well-built automation handles as well as or better than a person.
The organisational transformation dimension
The function-level analysis is not just a cost reduction exercise. At an organisational level, it is the foundation of a genuine workforce transformation strategy — one that takes seriously both the efficiency case for automation and the human dimension of redesigning how work gets done.
The organisations that navigate this transition most successfully are the ones that address it in three dimensions simultaneously: technology (what can be automated), process (how should the work be redesigned around automation), and people (how do we support staff through the change, what new capabilities do they need, and what does the new shape of their work look like).
At a broader level — as a society and as an economy — this framing also offers a more constructive path through the AI and employment conversation than the displacement narrative. The honest reality is that AI will change what work looks like in most sectors. The functions that are genuinely automatable will, over time, be automated. But the skills, judgment, and relational capacity that make a person valuable in a care role, a clinical role, or a service role are not being automated. The question is whether the transition is managed deliberately — with thought for what people will do when the structured administrative work is no longer theirs to carry — or whether it is allowed to happen through attrition without a plan.
Regulated service organisations are not the cause of this transition. They are the ones navigating it in the most demanding context — with governance obligations, workforce constraints, and client vulnerability that make both getting automation right and protecting their workforce important. The function-level framework is, in our experience, the most practical starting point for managing both.
The practical framework: from analysis to action
The six-stage framework below describes how a function-level analysis translates into operational change. It is the approach Taidotech uses in UpliftX engagements, and it is designed to be workable for a single business unit over four to eight weeks — not a multi-year transformation programme.
The pilot stage is where most organisations build the confidence they need to scale. Starting with one function — after-hours call handling, shift cancellation capture, key personnel notifications — and proving the model against real operational data is more persuasive than any business case. The subsequent functions become configuration exercises rather than new builds, each at lower cost than the one before.
The honest version of what this means for people
The function-level framework is deliberately honest about the employment implications of automation. Some functions that people currently perform will, over time, be performed by AI systems. The honest answer to "what does this mean for my team?" is not a reassurance — it is a plan.
The organisations that manage this most responsibly are the ones that engage their workforce in the process. Not as passive subjects of a management decision, but as active participants in mapping how their own work flows, identifying which functions they find most and least fulfilling, and shaping what the new version of their role looks like. The functions that automation takes over are rarely the ones people most value doing. The functions that remain — and the new ones created by the capacity automation frees — are typically more meaningful, more judgment-intensive, and more aligned with why people chose a care or service role in the first place.
That is not universally true, and it would be dishonest to claim otherwise. There are genuine workforce transitions ahead in regulated service organisations, as there are in every sector. But the organisations that approach those transitions with a clear, evidence-based picture of their function landscape — and a genuine plan for their people — are in a better position than those that avoid the conversation until the pressure forces it.