Determine whether the output should be used
Model outputs and calibrated uncertainty provide a stronger basis for assessing the individual case.
Research-based decision technology
Discern when AI merits trust
Grejna makes calibrated uncertainty actionable – helping organisations assess when an AI output is sufficiently well supported for use, when human review is needed and when a decision should be deferred.
The starting point
Not every AI output merits the same trust
A predictive model can be accurate on average without being equally reliable in every individual case.
For an organisation, knowing what the model predicts is not enough. It also needs to assess whether the output is sufficiently well supported to be used. Grejna helps determine which outputs can be used directly and which require human review, supported by calibrated uncertainty, customer-specific decision rules and traceable follow-up.
How does uncertainty affect the decision?
Illustrative example
01
Decision threshold
The interval stays clearly on one side of the threshold → Can be handled directly
The practical problem
Incorrect automation or unnecessary manual review
When all model outputs are treated the same, organisations face a difficult trade-off. Too much automation increases the risk of errors. Too much manual review, meanwhile, erodes much of the value that automation is intended to create.
When Grejna is relevant
When AI outputs affect recurring decisions with real-world consequences
This can include claims handling, fraud review, industrial quality control, predictive maintenance or other workflows where both errors and human review carry a clear cost.
What Grejna does
From calibrated uncertainty to controlled decisions
The name Grejna is inspired by the Old Norse greina – to discern and distinguish. For us, that means discerning when an AI output is sufficiently well supported for its intended use and when it requires review or further investigation.
Grejna sits on top of the organisation’s existing model. The scientific core comes from the open-source Calibrated Explanations method. Grejna provides the integrated commercial layer needed to make this information operational in real decisions and to monitor it over time.
Read the detailed explanationVerify that the basis for trust remains valid
Grejna is being developed to monitor whether the uncertainty signal remains useful for the intended decision context as data and operating conditions change.
Connect the evidence to the right action
The customer’s decision rules translate uncertainty, risk and review capacity into whether an output should be used, reviewed or escalated.
Preserve what actually happened
The model output, calibration status, decision rule, human handling and subsequent outcome are linked in a traceable decision chain.
Product boundary
Grejna complements the organisation’s systems. We do not try to replace them
The product has a focused role: preserving and verifying the relationship between uncertainty, trust, decision rules, actions and outcomes. This keeps the offering clear and close to the research expertise on which Grejna is built.
What Grejna is intended to handle
- Whether the uncertainty signal remains suitable for use.
- How uncertainty is translated into a controlled action.
- The link between decisions, human handling and outcomes.
- Traceability between calibration, decision rules and individual decisions.
- CE-specific verification, rollback and controlled changes.
What Grejna is intended to integrate with
- The customer’s model, data and model registry.
- Existing case-management and human-review systems.
- The organisation’s MLOps, security and logging environment.
- Governance and compliance systems.
- Identity, access-control and security solutions.
Research-based core
Calibrated Explanations provides the mathematical foundation. Grejna makes it usable in the organisation
Calibrated Explanations complements the model output with calibrated uncertainty information and local explanations. The calibration mechanism depends on the type of predictive task: calibrated probabilities for classification, conformal prediction and uncertainty intervals for regression. The method is open to scrutiny and continued research.
Grejna’s commercial value is being developed beyond the mathematical core: with the decision rules that govern action, the verification of whether the basis for trust remains valid, and the traceable record showing what the organisation and the system actually did.
Open core
Calibrated Explanations.
Commercial layer
Decision rules, trust monitoring and decision traceability.
Verification
The mathematical meaning of the CE method must be preserved throughout the decision chain.
Grejna Decision Assurance Pilot
Evaluate the value in a defined decision workflow
The pilot begins with a real decision and a model that the organisation already uses. The aim is to evaluate whether calibrated uncertainty can help the organisation discern which model outputs have sufficient support for their intended use, while improving the balance between automation and human review.
The pilot should give the organisation a clear basis for determining which cases can be automated and which should be reviewed by a person. It should also show how Grejna may affect automation rates, quality and review requirements compared with the current way of working.
The customer provides
A clearly defined decision and relevant evidence
An existing predictive model, a defined decision workflow and data that make it possible to compare the current process with a Grejna-supported approach.
Grejna provides
Uncertainty, decision rules and a traceable decision record
Calibrated Explanations, a customer-specific rule for whether an output should be used, reviewed or deferred, and a record that links the different parts of the decision together.
Together, we evaluate
Whether the solution creates measurable operational value
The pilot can be carried out by replaying historical cases or within a clearly defined operational workflow. The format is determined by the level of risk, data availability and the organisation’s objectives.
What the pilot should help answer
- Automation and quality
- What proportion of cases can be handled without manual review, and how often are those decisions correct?
- Review volume
- Can human review be focused on cases where the evidence is weaker or the risk is higher?
- Calibration and coverage
- Does the uncertainty signal meet the level of quality required for its intended use?
- Review time
- Does the time required to assess an individual case change?
- Decision reconstruction
- How quickly can the organisation establish what evidence was available, which rule was applied and what happened afterwards?
EU AI Act
Technical controls and traceable evidence
Grejna can provide technical support for areas including transparency, human oversight, documentation and monitoring. A pilot can also show which controls have been demonstrated, what evidence has been created and which responsibilities remain with the organisation.
Grejna does not replace legal assessment, risk management or the organisation’s overall compliance programme. The product’s role is to provide the decision controls and technical evidence that these processes require.
TEAM
Research, product development and commercialisation
Grejna AB is a Swedish research-based product company with roots in Småland. The team combines the work behind Calibrated Explanations with technical development, strategy and customer-focused commercialisation.
Contact person
Sam Löfström-Cavallin
Head of Commercialisation
Sam is responsible for company building, customer development, partnerships, pilot projects, financing and commercialisation.
Tuwe Löfström-Cavallin
Research and Technology
Tuwe is responsible for the research behind Calibrated Explanations and Grejna’s core technical expertise.
Helena Löfström-Cavallin
Research and Strategy
Helena initiated the Calibrated Explanations method and contributes research expertise, strategic guidance and the link between the scientific method and the commercial product.
Do you have a decision where uncertainty needs to become actionable?
Tell us briefly which model you use, which decision it affects and what you would like to improve. Sam will get back to assess whether a focused pilot could be relevant.
Contact person
Sam Löfström-Cavallin
Head of Commercialisation