Skip to content
Kit Hanson-Smith

5 min read

AI Pays Off in Sustainability Reporting, but Only for Leaders Who Steer It

AI's real job in sustainability reporting is to help stretched teams do the work well, not to replace them. Token pricing makes that help metered, so leaders must steer it.

This is the first proper post on this site, and it’s only fitting that it takes on the buzz topic of the moment: AI. As a sustainability consultant, I keep asking one thing: when AI starts to save sustainability teams real time, who’s the real winner?

Mark Zuckerberg said recently, “The primary purpose for AI is going to be the invention of new things, not automation.”1 He sells AI, so take that with a pinch of salt. But the point holds for reporting: the aim isn’t to automate the team away; it’s to empower them. That’s the lens I’d bring to sustainability reporting: AI enablement will equip typically overstretched teams with the tools needed to make their job easier.

Reporting teams are often juggling more than one set of rules. At the same time, the way we pay for AI is changing. Anthropic, for one, now bills Enterprise customers for every token they use,2 and a growing share of enterprise software is expected to follow.3 (A token is just a small chunk of text the model reads or writes.) AI’s help is real, in drafting and mapping across frameworks. But it only pays off for leaders who steer where it runs, what it costs and who signs off the numbers.

AI is already taking on the grunt work

A big slice of any reporting cycle goes on cleaning data and redrafting disclosures. That’s where AI already earns its keep. By Watershed’s own account, its agents turn messy activity data into workable, traceable data.4 That’s a vendor claim, not an independent measurement. Still, if it holds, every hour saved is an hour freed for the expert bit: explaining the numbers, testing assumptions and making the judgement calls.

The catch is accuracy. AI drafts and mappings can be confidently wrong, and checking them takes time vendors rarely count. So the time you save depends on who checks the output.

Mapping across frameworks is work AI suits

The same emissions figure or narrative often has to satisfy more than one set of rules. The real work is checking each disclosure, line by line, against each set of requirements. Official interoperability guidance covers some of this; start there.5 Where AI helps is the messier part: comparing narrative disclosures with requirements, flagging gaps and redrafting wording.

  • By 2027, companies in 18 jurisdictions will be issuing reports under ISSB Standards.6 More companies will be reporting under the global baseline, often alongside other rules.
  • In 2024, the EU’s reporting rules and the global ISSB baseline were judged highly aligned. The joint guidance pointed to “the high level of alignment achieved”.5
  • The EU’s revised rules, as drafted, offer only “limited alignment with ISSB frameworks”. And there’s no dedicated route for companies already reporting under the ISSB baseline.7

The two are drifting apart again, so the mapping needs redoing whenever either moves.57 That’s recurring, fiddly work: just what AI can help a small team do well. If the EU’s revised rules land as drafted, that’s another re-map.7

Token pricing puts AI on a meter

Finance is used to software as a fixed cost per head. Per-token billing changes that. Every redraft, re-mapping and agent re-run now lands on the bill.2

  • Anthropic’s Enterprise seat now “doesn’t include any usage”. Instead, “every token your team consumes is billed separately at standard API rates”, and older plans switch over at renewal.2 The seat itself reportedly costs $20 per employee per month.8
  • For some clients, the seat was “only about 20 percent” of the total bill. For them, usage, not seats, drives what they pay.8
  • “At least 40%” of enterprise SaaS spend is forecast to move to usage-, agent- or outcome-based pricing by 2030. It’s a market shift, not one vendor’s quirk.3

Here’s the thing: the bill probably won’t be huge. Claude Sonnet 5.5 lists at $2 per million tokens in and $10 per million out.9 So 100 million tokens in and 10 million out comes to $300. Even ten times that is $3,000. Not a budget-breaker, but enough that someone should own it. (There’s no published figure for a real reporting cycle’s token use, so this is a scale check, not an estimate.) The real risk isn’t the size of the bill; it’s that it’s metered and nobody’s watching it. The bigger hidden cost is expert time spent checking the output, and that’s on the meter too.

Steering comes down to three decisions

Steering takes three decisions from heads of sustainability, finance and procurement: 1) where AI runs, 2) what it costs and 3) who signs off. Sustainability teams rarely hold the AI contract, so you’ll want a seat at the table when IT and procurement set it up.

Where AI runs is a workflow choice. Point it at the high-volume work (drafting, framework mapping, gap checks) and keep it away from the judgement calls. What it costs is a budgeting choice. Give usage its own budget line, track spend per reporting cycle and ask vendors at renewal how they price it.23 Who signs off is an accountability choice, and the answer to the accuracy problem. A named person should be able to explain and validate every AI-assisted figure and draft before it goes into a compliance-critical disclosure. Each re-check adds spend too, so whoever signs off should know what the numbers cost.

So yes, the drafting and mapping gains are real, but they’re not self-executing. Teams that treat AI as a governed capability, with a budget line and a named owner for every figure, will get the time back. Teams that can’t harness AI or control spend will find the value harder to show and the cost harder to explain.

Steered well, the real winner is the AI-empowered team. AI won’t shrink it; it’ll help it do the work well.

Footnotes

  1. Heath, A. (2026) Mark Zuckerberg on Muse, Meta’s biggest AI bet yet. Sources with Alex Heath [podcast], 8 September. Available at: https://sources.news/p/mark-zuckerberg-meta-muse-ai-podcast-interview (Accessed: 3 November 2026). ↩

  2. Anthropic (n.d.) Usage-based Enterprise plans. Available at: https://support.claude.com/en/articles/11526368-usage-based-enterprise-plans (Accessed: 27 October 2026). ↩ ↩2 ↩3 ↩4

  3. Deloitte (2025) TMT Predictions 2026: SaaS meets AI agents (quoting a July 2025 Gartner forecast). Available at: https://www.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predictions/2026/saas-ai-agents.html (Accessed: 27 October 2026). ↩ ↩2 ↩3

  4. Watershed (2026) How leading sustainability teams drive business value with Watershed AI agents. Available at: https://watershed.com/blog/how-leading-sustainability-teams-drive-business-value-with-watershed-ai-agents (Accessed: 27 October 2026). ↩

  5. IFRS Foundation and EFRAG (2024) IFRS Foundation and EFRAG publish interoperability guidance. Available at: https://www.ifrs.org/news-and-events/news/2024/05/ifrs-foundation-and-efrag-publish-interoperability-guidance/ (Accessed: 27 October 2026). ↩ ↩2 ↩3

  6. IFRS Foundation (2026) Trustees announce five-year plan including new seat of the ISSB in Geneva. Available at: https://www.ifrs.org/news-and-events/news/2026/08/trustees-announce-five-year-plan/ (Accessed: 27 October 2026). ↩

  7. DLA Piper (2026) EU consultation on revised ESRS: key developments and implications. Available at: https://dlapiper.com/en-us/insights/blogs/environment-health-safety-and-product-compliance/2026/eu-consultation-on-revised-esrs-key-developments-and-implications (Accessed: 27 October 2026). ↩ ↩2 ↩3

  8. The Register (2026) Anthropic ejects bundled tokens from Enterprise plans. Available at: https://www.theregister.com/2026/04/16/anthropic_ejects_bundled_tokens_enterprise/ (Accessed: 27 October 2026). ↩ ↩2

  9. Anthropic (2026) Pricing. Available at: https://platform.claude.com/docs/en/about-claude/pricing (Accessed: 27 October 2026). ↩