Name: Watershed Climate Platform
Brand: Watershed

Question 1

Which AI technologies are used in Watershed today?

Accepted Answer

In our use of AI, we optimize for performance and efficiency and regularly evaluate the field of technology options. We use various machine learning and generative AI techniques, from classical statistical methods like regression to recent models like LLMs, including Anthropic, OpenAI, and Google Gemini.

Question 2

How does Watershed test and ensure AI quality?

Accepted Answer

Watershed AI products are tested on Watershed-crafted evaluation sets, representative of customer experience. We monitor performance of all deployed models. Every time we make an iteration to the agents, we give them hundreds of tests. For example, our agents had more than 50,000 evaluations ahead of rollout of AI-accelerated Product Footprints.

Question 3

Does Watershed’s use of AI require access to personal data?

Accepted Answer

No personal data is ingested or used by any generative AI feature on Watershed.

Question 4

How does Watershed ensure the security of  AI models and processed data?

Accepted Answer

Watershed uses enterprise-level security across its platform to ensure high levels of protection. Read more about our security protocols at trust.watershed.com.

Question 5

How does Watershed process personal data for use in AI?

Accepted Answer

We only use personal data for account management (username, which is often a person’s name, and email address). This is not used in the context of AI. For our privacy-related policies, please consult trust.watershed.com.

Question 6

Does Watershed AI allow for human review?

Accepted Answer

Yes. Watershed AI surfaces all key outputs for human review. With data cleaning, Watershed agents surface actions for review before executing to ensure there's full visibility into changes, and then keeps track of any changes applied to ensure auditors or others in the future can understand the process. With other Watershed AI like emissions factor mapping, every mapping comes with a match score to surface places where there might not be an existing, applicable EF and human review could be used to decide which EF comes closest. Across Watershed, since every calculation is transparent and traceable, you can see exactly how the AI reached its conclusions — not just what it decided.

Question 7

What steps do you take to safeguard against errors and hallucinations in model responses?

Accepted Answer

Watershed builds in a number of safeguards to minimize errors and hallucinations in the AI outputs and every AI output is surfaced for human review. Among our technical safeguards: We take a multi-agent approach to create checks and balances in the process. Judge agents evaluate the outputs of mapping or reporting agents and are kept separate to reduce confirmation bias. Every agent iteration goes through 50,000+ evaluations, tests designed for consistency and accuracy.

Question 8

Who works on AI at Watershed?

Accepted Answer

Our AI engineers, software engineers, and climate and data scientists work side by side to develop and evaluate systems and approaches, integrating and incorporating open source SDKs, libraries, and third-party APIs. We have 21 climate scientists and AI engineers working together to develop and maintain Watershed’s AI systems.

Question 9

Can I opt out of using AI in my Watershed instance?

Accepted Answer

We can work with customers to opt out of using AI solutions. Because you will lose some AI-dependent functionality when you opt out, it's best to have a conversation with the team to fully understand the opt-out paths. To request an opt-out, speak with your account manager at the point of purchase or your customer success manager at any point as a customer.

Question 10

What are Watershed's long-term plans for AI in the platform?

Accepted Answer

We will continue deploying dedicated engineering, product, and data science resources to build out sustainability AI use cases in a safe, responsible, and scalable manner. Future use cases will reflect the best practices above.

Stop wrangling data. Start driving change.

Any AI can make big claims. Watershed AI gets you results.

Give your sustainability team superpowers

Process utility bills in seconds to analyze spend & emissions

Draft reports and disclosures in moments

Clean messy data automatically

Leverage expert-built skills to tackle common challenges

Edit PCFs in an instant to reflect your supply chain

Automate PCFs across your portfolio

The leading platform for sustainability AI

Agents to accelerate work

Intelligence to improve results

Full controls for peace of mind

Become the AI expert your sustainability team needs.

Not all sustainability AI is created equal

Encoded sustainability intelligence

Safeguards for greater accuracy

Transparency + explainability

Human control and refinement

Advancing the field of sustainability AI

Criteria for credible AI-assisted carbon footprinting systems: The cases of mapping and lifecycle modeling

Built with sustainability intelligence

An LLM-based approach to creating a carbon emissions reduction levers library at scale

How Watershed uses AI