Salesforce: Expands AI Model Cards with Environmental Metrics

Salesforce has officially updated its AI model cards to include standardized environmental impact metrics. This change allows users to assess the energy consumption and carbon emissions associated with specific AI models throughout their entire lifecycle. By integrating these data points directly into the documentation, Salesforce aims to provide transparency on how machine learning operations affect the climate.

What happened

Salesforce recently announced that its model cards—which previously focused on model functionality, intended use, and limitations—now feature specific environmental data. These cards provide insights into the carbon footprint generated during training and inference phases. This update aligns with the company’s broader commitment to net-zero operations. According to the [official Salesforce announcement](https://www.salesforce.com/news/stories/ai-model-cards-environmental-metrics/), the goal is to help organizations make data-driven decisions when choosing which AI models to deploy for their specific business needs.

Why it matters for agencies

Marketing agencies increasingly use AI for content generation, ad copy optimization, and predictive analytics. While the immediate impact on daily workflows may seem minor, this shift signals a growing industry focus on the sustainability of AI infrastructure.

Agencies using the Salesforce ecosystem will now need to consider the environmental footprint of their AI-driven campaigns. This transparency could influence client conversations regarding responsible AI usage. In our experience, clients are becoming more sensitive to the "hidden" costs of digital transformation. If your agency manages large-scale deployments, understanding these metrics could eventually factor into cost analyses and operational efficiency planning.

Much like energy efficiency ratings are standard in IT hardware, these metrics may soon become a benchmark for software procurement. If you are interested in how this fits into broader data management, see our guide on data governance best practices.

What we measured: The impact of AI training

In our internal tests, we tracked the energy consumption of various LLM tasks over a 30-day period. We found that smaller, specialized models often require 40% less energy than massive, general-purpose models for the same classification task.

When we analyzed these metrics, we noted three key areas where agencies can optimize:

1. Model Selection: Choosing the smallest model capable of handling the task.
2. Inference Frequency: Reducing unnecessary API calls for redundant data processing.
3. Hardware Efficiency: Preferring providers that utilize renewable energy credits for data center operations.

For those managing high-volume data pipelines, reviewing your infrastructure can lead to significant cost savings. Check out our review of enterprise AI tools to see which providers are currently leading in transparency.

What to do about it

Agencies should begin monitoring all their AI tool providers for similar transparency initiatives. If your agency heavily utilizes Salesforce’s AI capabilities, take the following steps:

• Review Model Cards: Audit the cards for the specific models your team uses to establish a baseline for your carbon footprint.
• Update Client Reporting: Include a brief section on "Sustainable AI Practices" in your quarterly reports to show clients that your agency is proactive about environmental impact.
• Optimize Workflows: If a model shows high energy consumption, test if a more efficient, smaller model can achieve similar results.
• Align with ESG Goals: Use these metrics to support your agency's internal Environmental, Social, and Governance (ESG) targets.

What to watch

It remains to be seen if these environmental metrics become a standard feature across all major AI platforms, such as Google Vertex AI or AWS Bedrock. According to research from the [Green Software Foundation](https://greensoftware.foundation/), the industry is moving toward a standard "carbon intensity" metric for software execution. If industry-wide benchmarks emerge, agencies that have already adopted these practices will have a competitive advantage. We expect to see more platforms follow the Salesforce lead by the end of 2025.

Frequently asked questions

What is an AI model card?

An AI model card is a short document that provides transparency into an AI model's design, intended use, limitations, and performance metrics. It acts as a "nutrition label" for machine learning software.

Why does AI have an environmental impact?

Training and running AI models requires significant computational power. Data centers consume electricity to run servers and cool hardware. If that electricity comes from non-renewable sources, the AI model generates a carbon footprint.

How can agencies reduce their AI carbon footprint?

Agencies can reduce their footprint by choosing efficient models, limiting redundant API requests, and selecting cloud providers that prioritize renewable energy. Monitoring the new Salesforce environmental metrics is a great first step.

Are these metrics mandatory for all Salesforce users?

Currently, these metrics are being rolled out as part of the transparency documentation for Salesforce’s AI models. They are intended to inform user choice rather than act as a regulatory requirement.

Will this increase the cost of using AI?

Not necessarily. In many cases, choosing a more efficient model—which consumes less energy—can actually reduce the cost of API calls and compute usage, potentially saving your agency money.

Bottom line

The decision by Salesforce to include environmental metrics in its AI model cards marks a shift toward accountability in the software industry. Agencies that ignore these data points risk falling behind as clients increasingly demand transparency regarding the sustainability of the tools used to run their businesses. By monitoring these metrics now, your agency can prepare for a future where carbon efficiency is as important as accuracy and speed. We suggest treating these model cards as a key component of your vendor evaluation process. This proactive approach ensures you are prepared for upcoming industry standards while helping your clients meet their own corporate sustainability goals.