Understanding AI-Generated Models
When you generate the same Calculator more than once using the same business requirements and assumptions, the resulting models may not be identical.
This is expected.
Ekphrastic uses AI to reason about the business scenario and construct an appropriate model. It is not retrieving a fixed Calculator template or reproducing a previously generated model.
As a result, two generations can represent the same underlying business case in different ways while both providing a valid foundation for further refinement.
Why AI-Generated Models Can Differ
When translating your business requirements into a Calculator, AI makes a number of modelling decisions.
These can include:
which assumptions should be independently editable;
which intermediate calculations are useful to expose;
how a benefit should be decomposed into calculation steps;
how operational improvements are translated into financial value;
which outputs are most useful as key metrics;
which metrics are useful for comparison;
how supporting outputs are logically grouped;
the wording of labels, descriptions and guidance; and
reasonable starting values for assumptions that were not explicitly supplied.
For example, one generation might represent an AI operating cost as a fixed annual amount, while another might calculate the same cost as:
AI interactions × cost per interaction
Both approaches can represent the same underlying economic concept.
This flexibility is intentional.
What Should Remain Consistent?
Variation in model design should not mean that the fundamental business logic becomes arbitrary.
For the same business scenario, independently generated Calculators should normally reflect the same core economic relationships.
For example:
Operational activity → Improvement → Measurable impact → Financial benefit → Solution cost → Net benefit → ROI
The exact number of intermediate calculations, assumptions or outputs may differ, but the relationship between the assumptions and the resulting financial outcomes should remain coherent.
Ekphrastic's generation and processing pipeline applies modelling and structural controls designed to promote principles such as:
inputs feeding meaningful calculations;
transparent calculation chains;
explicit performance assumptions;
separation of operational and financial measures;
appropriate cost and benefit boundaries;
avoidance of obvious benefit double counting;
traceable ROI and payback calculations; and
editable assumptions rather than hidden performance constants.
This means two models can be different without being inconsistent.
An Example
Consider a business case where automation releases employee capacity.
One generation might model the value chain as:
Hours saved → FTE capacity → Labour value
Another might use:
Baseline workload → Future workload → Workload reduction → FTE capacity → Realisable labour saving
The second Calculator contains more intermediate calculations, but both can describe the same underlying value mechanism.
Starting assumptions can also vary.
For example, one generation might propose that 30% of released capacity can realistically be converted into financial savings, while another might propose 35%.
Neither assumption should automatically be treated as an established fact.
They are starting assumptions for you to review and refine based on the organisation, customer or opportunity being modelled.
Why Doesn't Ekphrastic Force Every Generation to Be Identical?
Doing so would fundamentally change what the AI is being asked to do.
Ekphrastic is designed to support different industries, technologies, operating models and business cases. Using a rigid set of predefined calculations could make generation more deterministic, but it would also reduce the AI's ability to reason about the specific business problem being modelled.
Instead, Ekphrastic provides a controlled structure within which AI can determine an appropriate modelling approach.
This allows the platform to remain business-case agnostic while applying consistent modelling and processing rules.
Expected Variation vs Modelling Integrity
It is useful to distinguish between expected variation and modelling integrity.
Expected Variation
The following differences between independently generated Calculators can be normal:
different but reasonable starting assumptions;
different intermediate calculation steps;
different wording and descriptions;
different output ordering;
different key metric selection;
different logical output groups; and
different but economically equivalent calculation structures.
Modelling Integrity
Regardless of how the model is structured, areas that should remain sound include:
calculation direction;
mathematical relationships;
cost and benefit boundaries;
units and percentages;
dependency chains;
treatment of recurring and one-time costs;
avoidance of inappropriate double counting; and
ROI and payback methodology.
Ekphrastic allows the first category to vary while applying controls intended to strengthen the integrity and transparency of the generated model.
AI generation should nevertheless be treated as the creation of a model for review and refinement, rather than a guarantee that every proposed assumption or economic interpretation is appropriate for your specific organisation or customer.
AI-Generated Foundations, Human Refinement
An AI-generated Calculator should not be treated as an unquestionable financial forecast.
It is a structured first draft of the business case.
Ekphrastic is designed around the principle:
AI-generated foundations. Human refinement.
AI removes the blank page by constructing the initial assumptions, calculations, outputs and model structure.
You then apply the knowledge that AI cannot reliably possess about your particular organisation, customer or opportunity.
This includes validating:
baseline volumes and costs;
expected performance improvements;
adoption assumptions;
benefit realisation;
commercial pricing;
implementation costs;
financial treatment; and
customer-specific constraints.
This combination allows AI to accelerate business-case creation without assuming that an automatically generated model contains knowledge that only you, your organisation or your customer can provide.
Regenerating a Calculator
Generating the same request again should be understood as creating another interpretation of the business case, not reproducing an exact copy of the previous Calculator.
If you already have a strong model and only need to change assumptions, calculations or presentation, it will usually be better to refine the existing Calculator in ROI Studio rather than repeatedly regenerating it.
Regeneration is more useful when you want AI to reconsider the overall modelling approach and provide a new starting point.
The Key Principle
Consistency does not require identical models.
Two generated Calculators can use different assumptions, intermediate calculations or structures while representing the same underlying business economics.
The important question is not:
"Did AI generate exactly the same model?"
It is:
"Does the model provide a coherent, transparent and financially defensible representation of the business case that I can validate and refine?"
That is the standard Ekphrastic is designed around.
Last updated