Resources → TTP-06

AI + Human Collaboration: Who Actually Did the Work?

The credit problem

A polished report now contains traces of human judgement, AI-generated drafting, copied inputs, automated analysis, and manager review — folded together so cleanly that the work history disappears. The output looks simple. The contribution doesn't.

Consider four scenarios already playing out in 2026:

  • The AI-assisted analyst. A consultant uses an AI tool to draft sections of a £200K client deliverable. The AI draft is 60% of the page count, 30% of the substantive insight, 0% of the relationship management.
  • The AI-assisted vendor invoice. A research vendor delivers a 40-page report in three days that previously took three weeks. The invoice is unchanged. No record exists of which sections justify which fraction of the price.
  • The AI-assisted developer. Half the lines are AI-generated. All architectural decisions remain human. Code review caught two AI-introduced bugs and missed one.
  • The radiologist + AI scenario. An AI flags a scan as low-risk, the radiologist agrees, and the patient is later diagnosed with something the scan missed. Who is responsible?

In each case, contribution has become invisible, and downstream systems — performance reviews, bonus calculations, vendor billing, liability assignment — assume contribution is visible.

Why existing attribution methods fall short

Shapley values break down when contributors are asymmetric — a human reviewer's 30 minutes carries different weight than an AI's 30 seconds.

Actual causality approaches are closer to the right shape but treat the agent's confidence and the human's epistemic level as undifferentiated.

Disclosure-based attribution ("AI was used") satisfies a policy requirement and tells a reviewer almost nothing.

What the human-AI collaboration case actually needs is a structured record that captures contribution, authority, review, and outcome together.

The Attribution Stack — six layers

The Attribution Stack is TimeToPoint's canonical framework. Each layer answers a specific attribution question:

  1. Input — What entered the workflow
  2. Contribution — Who contributed which part
  3. Authority — Under whose mandate
  4. Review — Who checked the work
  5. Outcome — What happened after
  6. Credit / Pay — How value is distributed

Layers 1–4 are operational. Layer 5 becomes important for outcome-based AI execution. Layer 6 becomes important when AI-human work moves from cost centre to revenue instrument.

Layer 6 should usually remain advisory at first; compensation, liability, and settlement policies remain with the customer's HR, finance, legal, or procurement teams.

Why this is operational, not theoretical

Microsoft Work Trend Index 2025 found 82% of leaders expect digital labour to expand workforce in the next 12–18 months. 28% of managers are considering hiring AI workforce managers. 46% of leaders say their organisation uses agents to fully automate workstreams.

The attribution question is no longer hypothetical. It is the question payroll, procurement, performance, and risk are all asking simultaneously, in 2026, often without a shared framework.

"AI didn't make the work invisible. The lack of attribution did."

Deeper reading

See the Work Attribution Record SKU on the Product page →

Make the work visible.

See a sample Work Attribution Record Get my CFO AI Bill Evidence Brief

The Attribution Stack surfaces contribution evidence. Compensation, performance, liability, and settlement decisions remain with the customer's functions.