Overview of the Upright net impact model
This page introduces key algorithms used to quantify net impact.
The Upright net impact model consists of two main parts: the macromodel, and the company model. The primary output of the Upright net impact model is net impact. In addition, it also produces UN SDG, EU taxonomy, EU SFDR Principal Adverse Impact (PAI), and CSRD DMA metrics.
Macromodel
The macromodel integrates information from a variety of sources to produce estimates of the impact of all products and services.
Input data
The main data source used by the macromodel is a database of 200M+ scientific articles. Other data sources include databases from the World Bank, IMF, WHO, OECD, Eurostat, IPCC, CDC, USDA, IHME, and others.
When little quantitative data is available, the Upright net impact model relies mostly on the results of Upright's proprietary NLP deep learning algorithm that reads causal statements from hundreds of millions of scientific articles. When reliable, readily available quantitative data is available, the Upright net impact model puts a greater weight on this data.
Algorithms
The macromodel consists of three major algorithms:
pageExtraction of causal links from scientific literaturepageGeneralization of scientific knowledgepageAllocation of impact across value chainsCompany model
The company model combines the information produced by the macromodel with information on specific companies, most importantly information on what those companies are doing (i.e. what their products and services are) to produce an estimate of the impact of each company.
pageEstimation of company product mixesGranularity of products and services
Upright's models the impact of companies primarily through their products and services. For this to work well, the products and services must be modelled with sufficiently high granularity; the design principle is that any two products that have different impacts must be represented as two different products.
Below are examples of product labels used to model some well-known companies (click to expand)
Quantification of impacts that are lacking natural units
Some impact categories have natural, absolute, generally-accepted units, while some lack such scales. For example GHG emissions, taxes, and jobs have natural, absolute, generally accepted units, such as carbon dioxide equivalent emissions, dollars, or FTEs. Other impact categories, such as Meaning & Joy, do not have similar units.
Internally, the Upright net impact model understands impacts in relative terms, answering questions like "what is this company's share of all GHG emissions created by the private sector", or "what is this company's share of all knowledge created by the private sector". Therefore, the model is able to produce estimates on impacts even in the absence of natural absolute units of measurement.
Producing impact metrics beyond net impact
Upright's net impact model has been primarily built for measuring net impact, which requires comprehensive quantification of all essential impacts a company creates.
Given that the model's internal representation captures impact comprehensively, it has been possible to quickly adapt the model to also produce impact metrics used by other frameworks, such the UN SDGs, the EU taxonomy, EU SFDR Principal Adverse Impacts, or CSRD DMA.
Last updated
