AI-first company

Assessment · Task and job maps

TPAH task analysis

Which of our own tasks could AI do or help with?

By
Last reviewed
Corrections
Report an error
Who answers Nobody is surveyed. Process owners check the results
How long An analysis from a description of the company and its roles
Can you use it? Paid service
Testing Emerging

Early testing only: a pilot, an expert review, or tests that found problems still to be fixed.

In short

A method that scores each task in a company, not each job, on four separate scales: can AI do it, can this company deploy it, can AI help the person, and how much the task depends on a human.

For whoever runs the assessment: an HR or operations lead, an internal project team, or an outside consultant.

What it measures

TPAH looks at a role in a particular company and breaks it into tasks. Each task gets four scores from 0 to 10:

  • T, technical feasibility. Could AI do the task, given clear inputs and a clear expected result?
  • P, practical deployability. Could this organization put that in place and keep it running? This depends on its data, processes, quality checks and the cost of a mistake.
  • A, augmentation. How much could AI help the person doing the task, without taking it over?
  • H, dependence on a human. How much does the task rest on judgement, relationships, responsibility, physical presence and knowledge that is hard to write down?

The four scores are never merged into one number. Each answers a different question, and the useful readings come from the combinations:

  • High T and low P: the technology could do it, but the organization is not ready.
  • High A and high H: AI is a strong assistant for an expert, not a replacement.
  • High T and low H: a candidate for automation, if deployment allows.

The method also lists the new work that AI brings with it, such as checking outputs and keeping records, and says who would have to own it.

How to use it

  1. Treat the output as a shortlist of pilots, not as a decision.
  2. Have the owner of each process check the scores for the tasks you plan to change.
  3. Start with a few tasks that score high on P and have a result that is easy to check.
  4. Build assistants for experts where A and H are both high.
  5. Name who will own the new checking and platform work before you scale.
  6. Measure time and quality against a baseline. See measure time saved per task against a baseline.

The worked example shows these steps on a fictional company.

What it does not measure

  • Results. A score is not a saving, a return or a share of staff who can be replaced.
  • What staff think or do. Nobody is surveyed or observed. For attitudes, trust or psychological safety, use the assessments built for those.

How it was tested

It has been through one expert review, published in the project report. The authors call it a prototype that needs further testing.

  • Expert review of 50 roles. At a workshop, people from business, the public employment service and education read the role reports and said whether each part matched their experience. Answers of “yes” or “rather yes”:
What was reviewedPositive answers
The overall diagnosis24 of 28
The split between automation and help for the person23 of 23
The human-dependence score19 of 24
The gaps between what is possible and what can be deployed19 of 29
  • What the reviewers corrected. Tasks listed under the wrong role, deployment gaps that underplayed data protection and security, and the weight given to some human tasks. The report says the direction of the results is solid, and the exact numbers for single tasks are less certain.
  • A documented method. The steps, the scales and the formulas are written down, and each score can be traced back to the material it was based on.
  • Scores from an AI model. In current runs an AI model produces the main scores, following fixed instructions and the documents assigned to each task. One model and one configuration are used.
  • A human-dependence score built from twelve questions. The method’s documentation calls this a working way to turn the idea into questions, not a fully tested scale.
  • No outcome data. The scores have not been compared with what happened in real companies that deployed AI.

Licence and access

The provider’s page states no licence and no price. The analysis is sold as a service by the Human for Future foundation, and its materials are in Polish. If you want to reuse the scales or the method in your own work, ask the provider.

Alternatives

Other assessments with a similar job, and when to prefer each.

See it applied

Practices that use it

The details

Testing
Emerging Experts from business, the public employment service and education reviewed role reports for 50 roles at a workshop. Most answers confirmed the picture: 24 of 28 on the overall diagnosis, 23 of 23 on the split between automation and help for the person. No study has yet compared the scores with outcomes.
Looks at
Job and its tasks
How the answers are collected
An AI model scores each task from documents about the role and the company, following fixed instructions. People have to check the scores.
Length
No staff survey. The analysis starts from a description of the company and its roles. Process owners then need time to check the results.
Cost
Paid service. The price is not published.
Licence
Not statedThe provider states no licence. The analysis is offered as a service. Ask the provider what you may reuse.
Languages
Polish. No English materials are available.
By
Research team at Rzeszów University of Technology and the Human for Future foundation
Provider
Human for Future foundation: TPAH The provider’s page is in Polish.

Sources

  1. Rzeszów University of Technology and Human for Future foundation (2026). Research report, part I: the state of AI and robotics, with the task analysis of 50 roles and its expert review (in Polish). aits.kia.prz.edu.pl
  2. Human for Future foundation. TPAH: description of the method (in Polish). human4future.org

Related studies on this site