A 118-person company, task by task

The company

The model company is a business-to-business firm with its own assembly plant. It designs, assembles, configures and installs modular systems that monitor and control its customers’ technical infrastructure, and then services them. It is a manufacturer, a technology integrator and a service company in one.

• 118people
• 31distinct roles
• 375task families analysed

• Customers: about 70 active customers, such as factories, logistics centres and food processors.
• Workload: 12 to 18 installations under way at once, and about 260 installations under service.
• Where the money comes from: products and kits 43% of revenue, design and installation services 27%, maintenance contracts 19%, repairs and parts 11%.
• How it is organized: by function, with project managers who coordinate people from several departments. Production is the largest function, with 34 people. Service has 17.

Where work gets stuck

The company needs better productivity and a better flow of information. It cannot weaken technical responsibility, quality or the security of customers’ data to get them.

The question

The management question was not “how many jobs can AI replace?”. It was four questions:

1. Which tasks can realistically be automated?
2. Where should AI strengthen a person instead?
3. What stays firmly human or physical?
4. What new duties appear once AI is in use?

Asking about whole jobs hides the answer. A service technician diagnoses faults, talks to customers, writes reports and repairs equipment. AI relates to each of those differently. See map tasks before deciding what to automate.

The task map

The 31 roles were broken into 375 task families. Each one was scored from 0 to 10 on four separate scales:

• T, technical feasibility. Could AI do the task, given clear inputs and a clear result?
• P, practical deployability. Could this company put that in place and keep it running?
• A, augmentation. How much could AI help the person, without taking over the responsibility?
• H, dependence on a human. How much does the task still need judgement, relationships, responsibility or physical work?

The four scores are never merged into one. A high P does not mean a person can be let go. It means the task has good conditions for deployment. A high A together with a high H points to an assistant for an expert, not to a handover.

The scores come from an AI model that evaluated each task against a description of the company and the role. The owners of each process then check them. The method is described on the TPAH assessment page.

Most of the 375 task families sit high on augmentation and in the middle on deployability

A: how much could AI help the person?

9–10 616245 8–9 314224244 7–8 1522254219 6–7 2143343 5–6 23631 4–5 11 3–4 221 2–3 41 1–2 1 0–1 1 0–11–22–33–44–55–66–77–88–99–10

P: could this company deploy it?

Show the numbers as a table

A ↓ / P →	0–1	1–2	2–3	3–4	4–5	5–6	6–7	7–8	8–9	9–10
9–10	0	0	0	0	0	6	16	24	5	0
8–9	0	0	0	3	14	22	42	44	0	0
7–8	0	1	5	22	25	42	19	0	0	0
6–7	0	2	14	33	4	3	0	0	0	0
5–6	2	3	6	3	1	0	0	0	0	0
4–5	1	0	0	0	1	0	0	0	0	0
3–4	2	2	1	0	0	0	0	0	0	0
2–3	0	4	1	0	0	0	0	0	0	0
1–2	0	1	0	0	0	0	0	0	0	0
0–1	0	0	1	0	0	0	0	0	0	0

The largest cluster is high on augmentation and medium to fairly high on deployability. That is the first result: the main way to get value here is a person supported by AI, not an AI agent running a whole process unchecked.

What the map shows

The company as a whole

Weighted by the headcount in each role, the company scores:

• 4.83P, practical deployability
• 7.39A, augmentation
• 4.46H, dependence on a human

The company has real potential to use AI, mostly by supporting people. Dependence on a human sits in the middle of the scale: a large part of the work still needs knowledge, responsibility, relationships and hands-on work in a real technical setting.

The work portfolio

17% of work exposure has high deployability; 25% has high augmentation

• 17% High deployability Tasks with good conditions for real automation: digital, repeatable, easy to check.
• 25% High augmentation AI can help the person a great deal, but should not take over the responsibility.
• 26% Physically anchored Work tied to a site, a place, safety or doing something by hand.
• 31% Mixed Tasks with a mixed profile.

Show the numbers as a table

Part of the work	Share	Full-time equivalents
High deployability	17.1%	20.2
High augmentation	25.3%	29.9
Physically anchored	26.2%	30.9
Mixed	31.4%	37.0

• 17% of work exposure: high deployability. The place for fast automation of tasks that are digital, repeatable and easy to check.
• 25%: high augmentation. AI prepares an analysis, a draft or a recommendation. The person keeps the decision.
• 26%: physically anchored. Work tied to a site, to assembly, to safety or to the customer’s premises. Automate the information around it.
• 31%: mixed. Improve these process by process, without promising that AI will replace the work.

The result does not support a simple headcount-reduction scenario. It supports a programme to improve productivity, quality and the company’s ability to cope when things go wrong.

By function

In every function, AI could help people more than the company could automate

• P: could this company deploy it?
• A: how much could AI help the person?

0246810

Production 34 people

4.1 · 6.0

Service 17 people

4.9 · 7.6

Sales and marketing 13 people

4.9 · 8.0

IT and product 13 people

5.1 · 8.2

Logistics 10 people

5.0 · 7.2

Quality 8 people

5.1 · 7.7

Operations 7 people

4.9 · 8.1

Finance 6 people

6.0 · 8.7

Purchasing 5 people

5.8 · 8.5

HR 3 people

6.1 · 8.6

Management 2 people

3.7 · 7.9

Show the numbers as a table

Function	P	A
Production (34 people)	4.10	6.00
Service (17 people)	4.90	7.60
Sales and marketing (13 people)	4.90	8.00
IT and product (13 people)	5.10	8.20
Logistics (10 people)	5.00	7.20
Quality (8 people)	5.10	7.70
Operations (7 people)	4.90	8.10
Finance (6 people)	6.00	8.70
Purchasing (5 people)	5.80	8.50
HR (3 people)	6.10	8.60
Management (2 people)	3.70	7.90

Where to automate

The highest deployability score was 8.53. No task scored 10. Even the best candidates need integration, quality control, a process owner and a way to handle exceptions.

The ten tasks that would be easiest to deploy are digital, with a clear input and output

0246810

Reporting leads in the CRM Marketing

8.53

Recording what was agreed with customers Key accounts

8.47

Drafting internal newsletters Marketing

8.47

Reconciling bank statements Accounting

8.42

Tracking health checks and safety training HR and payroll

8.17

Monitoring market trends Sales engineering

7.51

Checking who is available Operations planning

7.50

Warehouse and shipping documents Warehouse

7.46

Effect of the schedule on stock Materials planning

7.46

Logging and sorting service requests Service desk

7.45

The best candidates share a pattern: structured data, a clear result and a simple way to check it. AI prepares or checks a record. A person approves the exception, or any decision with a financial or operational effect.

Tasks where the scores and the business effect line up

These tasks sit in different departments. The value comes from a shorter cycle, fewer gaps, better data or a faster reaction to an exception. A high score helps to choose a pilot. It does not replace a look at cost, data and integration.

Where to augment

In the technical core of the business, the highest scores were for augmentation. These are tasks where AI can quickly work through logs, diagrams, code, risks and options. It does not have the full picture of the installation, the safety consequences or the customer relationship.

For expert tasks, the help AI could give is far ahead of what could be automated

• P: could this company deploy it?
• A: how much could AI help the person?

0246810

Analysing electrical diagrams Electrical fitter

6.4 · 9.0

Analysing fault logs and alarms Service technician

6.0 · 9.0

Writing the report after a service visit Service technician

6.5 · 9.0

Analysing software logs Industrial programmer

6.0 · 9.0

Forecasting project deadline and margin Project manager

6.0 · 9.0

Keeping the risk register Project manager

6.4 · 8.9

Designing diagrams and control systems Automation engineer

6.1 · 9.0

Implementing control logic Industrial programmer

6.3 · 8.9

Choosing measurement and control equipment Automation engineer

5.1 · 9.0

Reviewing code and checking changes Industrial programmer

6.3 · 9.0

Show the numbers as a table

Task	P	A
Analysing electrical diagrams (Electrical fitter)	6.38	8.99
Analysing fault logs and alarms (Service technician)	6.02	8.99
Writing the report after a service visit (Service technician)	6.47	8.96
Analysing software logs (Industrial programmer)	6.01	9.00
Forecasting project deadline and margin (Project manager)	6.00	9.00
Keeping the risk register (Project manager)	6.40	8.93
Designing diagrams and control systems (Automation engineer)	6.09	9.00
Implementing control logic (Industrial programmer)	6.33	8.95
Choosing measurement and control equipment (Automation engineer)	5.14	8.97
Reviewing code and checking changes (Industrial programmer)	6.33	8.99

The working model is plain. AI analyses, sorts and prepares options. The expert understands the situation on site, judges the risk, talks to the customer and answers for the technical decision. This route has more value than trying to replace automation engineers, industrial programmers and project managers, who are hard to hire.

The work AI creates

AI does not only remove or shorten tasks. It creates work: checking outputs, logging for compliance, fitting AI into the process, passing exceptions to a person, maintaining prompts and tools, and supervising quality.

The analysis found 187 pieces of new work: 132 inside roles and 55 that depend on someone else

• 132 New tasks inside roles Doing the work with AI, checking its output and logging it become part of the daily job. Owner: Function managers and staff If nobody owns it: AI seems to save time while duties grow unnoticed.
• 26 Dependencies on process owners Someone must set where AI sits in the process, the rules, the exceptions and the supervision. Owner: Process owners If nobody owns it: Each team does it differently and exceptions multiply.
• 29 Dependencies on a platform owner Someone must maintain the tools, prompts, settings and integrations. Owner: AI product owner or IT If nobody owns it: IT becomes a bottleneck and nobody answers for quality.

The minimum set of controls

If this work has no owner, part of the saving is eaten by manual corrections, unofficial supervision and queues for IT. Name an AI product owner and the process owners before scaling, and write the new duties into people’s real workload. See budget for the work AI creates.

A 90-day plan

The recommended plan runs for 90 days in three stages, each with more risk than the one before. Every stage has its own measures, a person in control, and a decision at the end: continue, fix, or stop. Scaling comes only when the quality of the output, the number of exceptions and the real balance of time are known.

Three stages: simple digital work first, transactions second, the technical core last

1. Days 1–30 1. Quick proof of value Show a time saving with no loss of quality. Pilots
   • Meeting notes and CRM entries
   • Sorting service requests
   • Alerts for health checks and safety training
   • A baseline for time and quality
   Measures Time per entry; completeness of the CRM; time to log a request. Go on only if The work is faster, quality has not dropped, and exceptions are visible.
2. Days 31–60 2. Transaction processes Automate repeatable operations, with a person in control. Pilots
   • Bank statements and e-invoicing exceptions
   • Comparing supplier offers
   • Material shortages and surpluses
   Measures Cycle time; number of exceptions; rush orders; number of corrections. Go on only if A person checks before anything is written to the system, and the exception process works.
3. Days 61–90 3. Core operations Help experts with diagnosis, analysis and documentation. Pilots
   • Diagram assistant for electrical fitters
   • Log analysis
   • Service report drafts
   • Quality document checks
   Measures Rework; share passing final test first time; mean time to repair; completeness of reports. Go on only if The expert confirms it is useful and safe, and that responsibility is clear.

In parallel, from day one: appoint an AI product owner, classify the data, set rules for using customers’ information, keep a register of pilots, and agree one standard for checking outputs.

Recommendations

1. Start from tasks, not jobs. Ranking whole roles hides the differences between documentation, diagnosis, decisions, relationships and physical work. The list of pilots should be made of specific tasks and flows.
2. Run two tracks. An automation track shortens repeatable digital processes. An augmentation track strengthens service, automation engineering, projects and technical analysis. See run two tracks.
3. Manage AI as a product. One person or a small team answers for the platform, the standards, the integrations and the list of work to do. Process owners answer for business results and exceptions.
4. Measure time, quality and the cost of exceptions. The amount of text generated measures nothing. The minimum set is cycle time, completeness, number of corrections, share of exceptions and the time people spend checking.
5. Protect customers’ data and technical responsibility. Installation data, logs, configurations and customers’ documents should not go into public tools without an approved set-up. Decisions on safety, quality and changes to a system stay with an authorised person.
6. Join the uses into a chain. The most value appears when CRM notes improve planning, service data feeds product development, and document checks cut rework and hold-ups at acceptance.

For the model company, AI is first of all a layer that organizes the flow of information and strengthens experts. Automation makes sense where the result is easy to check. In the technical core, the gain is a faster and better human decision.

The analysis shown here is offered by the Human for Future foundation. The provider’s page on TPAH is in Polish.

Practices used

The example applies these practices from the library.

• Map tasks before deciding what to automate Break jobs into tasks and look at each one. Published scores rate how much of a job AI could do, its “exposure”. Treat a high score as a place to look, not as a decision to automate. Do first · Effort: low (days, one person) Evidence: Moderate Contested Process redesign · Explore
• Start with one costly process, not a company-wide rollout Pick one process that visibly costs time or money. Apply AI there, with a person checking the output, and expand only when the numbers move. Do first · Effort: medium (weeks, a few people) Evidence: Common practice Strategy · Explore
• Run two tracks: hand some tasks to AI, give experts AI help on others Treat AI work as two programmes. One automates repeatable digital tasks whose results are easy to check. The other gives experts AI help on work that cannot be handed over. Judge each by its own measure. Do next · Effort: medium (weeks, a few people) Evidence: Common practice Strategy · Adopt
• Budget for the work AI creates AI removes some work and adds other work: checking its output, keeping records, handling exceptions and looking after the tools. Count that work and give it an owner before you count the saving. Do first · Effort: low (days, one person) Evidence: Common practice Process redesign · Adopt
• Name one accountable AI owner Give one named senior person the responsibility for AI: priorities, budget, risk and results. Shared responsibility tends to mean nobody decides. Do first · Effort: low (days, one person) Evidence: Common practice Governance and risk · Adopt
• Measure time saved per task against a baseline Record how long a task takes and how good the result is before you add AI. Without that starting figure you cannot tell a real gain from a hopeful one. Do first · Effort: low (days, one person) Evidence: Common practice Measurement · Adopt