2026 Decision Intelligence Benchmark — Special AI Report
How ESG and sustainability leaders are adopting AI — or not — to achieve their business objectives
Table of ContentsESG and sustainability leaders present one of the more unusual AI profiles in the benchmark. For example, their teams use public generative AI daily at a rate that exceeds stated expectations — something unique across Assemble’s benchmark — made possible because leaders hold some of the lowest expectations for daily AI use of any function surveyed, ranking 10th out of 11 functions on that vector.
At the same time, this cohort rates its teams’ enterprise AI proficiency as largely beginner-level, trusts public AI tools less than most (10th of 11 on trust), and sees only modest disruption risk if AI disappeared tomorrow (9th of 11 on dependence). The creates an image of a function that has started using AI for specific tasks — without yet having formed the organizational ambition or infrastructure to scale that use. Based on adoption rates, the internal innovators already exist; the job now is to find them and formalize what they are doing.
The result? ESG teams are using AI. Experimentation is real. But the value isn’t showing up yet in how the work gets done. One reason could be the very nature of ESG work that is rooted in being accurate: 80% of leaders flag hallucination and inaccuracy as trust concerns for public AI — the highest rate in the benchmark — and data privacy and policy misalignment aren’t far behind.
Leaders know the risks. What they need now is a path from “we’re trying it” to “we trust it enough to build on it.”
For the purposes of this report, and to better understand how leaders think about different kinds of AI, we grouped tools into three categories:
ESG teams are using AI more than their leaders expect – that’s unique across Assemble’s benchmark.
ESG leaders hold some of the most conservative expectations for AI use of any function in the benchmark. Averaged across all three AI types, only 45% of ESG leaders expect daily use — 13 points below the cross-functional average of 58% and the second-lowest of all 11 functions surveyed. The gap is widest on company-owned AI, where ESG’s 29% expected daily use sits 22 points below the benchmark average.
The role-level picture adds a notable dimension. Managers hold higher expectations for daily use of Enterprise AI at 60%, compared with 50% for both directors and senior leaders. In most functions, expectations are highest at the senior level.
Actual use tells a different story than expectations alone. Across all three AI types combined, ESG ranks 3rd of 11 functions on actual daily use — a stark contrast to its 10th-place position on expected usage. Public generative AI leads at 57% daily use — the only category where ESG teams outrun their own leaders’ ambitions. Enterprise AI daily use sits at 42%, above the cross-functional average of 36%. The 27% actual daily use of company-owned AI slightly trails the average of 29%.
Senior leaders report the highest actual public GenAI use at 67% daily — more than any other level within the function. That senior-level engagement with the most accessible AI type, alongside their relatively low trust in its outputs, points toward interest from this cohort, but these leaders may not yet have a clear picture of how AI can reshape their operations.
The expectations-versus-actuals story for ESG is nearly the inverse of most functions. Where most functions show actual use trailing expectations — sometimes by wide margins — ESG leaders routinely see usage at or above what they anticipated. Public GenAI shows a rare positive gap (+2 points). Enterprise AI shows a smaller-than-average negative gap (52% expected vs. 42% actual, a 10-point shortfall versus the wider divergence seen across the benchmark). Company-owned AI shows a minimal gap (−2 points), though the expectation itself was already the lowest in the benchmark.
The practical implication is that ESG leaders may be underestimating their teams’ AI engagement. They may not be the tip of the spear, but the group is curious. A function that ranks 3rd on actual adoption but 10th on expected usage has a lot of room to grow.
ESG holds the lowest average AI use expectations in the benchmark — 10th of 11 functions — yet teams are using public AI at rates that exceed those expectations, ranking 3rd of 11 on actual adoption. The function seems to have bought into AI practice informally before its leadership has built the ambition framework to match. Members describe a state of experimental engagement: AI is in the workflow, but not yet in the system. The risk is not that teams resist AI; it is that informal use continues without the governance, standards, or review processes that the function’s data accountability requirements demand.
ESG leaders’ dependence on AI is among the lowest in the benchmark, and their automation outlook is the most conservative of any function. That combination suggests the function is still building the case and the confidence to make AI a core part of its work.
ESG leaders report that their teams would largely carry on if AI disappeared tomorrow, with 27% saying it would be business as usual, and another 56% expecting only slight disruption. Just 17% anticipate moderate or worse disruption — placing ESG 9th of 11 functions on this metric, compared with a cross-functional average of 27%.
Within the function, senior leaders show the lowest dependence: none report moderate or worse disruption, with responses split evenly between business as usual and slight disruption. Directors and managers show slightly higher, though still modest, dependence at 17% and 19% respectively.
ESG & CSR Board Membership Directors describe this pattern accurately: heavy experimentation is underway, but AI has not yet become essential for ESG workflows. The function is using AI; it has not yet built its critical paths through it.
ESG leaders hold the most conservative automation outlook in the benchmark. Nearly two-thirds (64%) expect AI to automate 20% or less of the function’s work over the next 24 months. Another 30% see automation potential in the 21–40% range. Only 6% expect 40% or more — last of all 11 functions, tied with CSR and Employee Experience at that threshold while sitting below the 9% cross-functional average.
This conservatism likely reflects a genuine assessment of the function’s work and aligns with other functions like CSR. The function’s skepticism about hallucination and output accuracy, detailed in Part 4, reinforces that leaders do not yet see AI as reliably trustworthy enough to bear responsibility for the high-stakes outputs ESG requires.
ESG’s 9th-place ranking on dependence and last-place finish on high-end automation outlook reflect a function with a sober view of what AI can and cannot currently do in a domain defined by regulatory scrutiny, data auditability, and public accountability. Members describe AI as actively in use but not yet trusted with the work that matters most.
Across all three AI types, most ESG teams land in the competent range. ESG ranks 9th of 11 functions on overall team proficiency — but still close to the average. Senior leaders assess their teams’ enterprise AI proficiency more critically than directors or managers do.
Across all three AI types, ESG teams cluster in the competent range — indicating an ability to use the tools for current tasks but not yet at the level of workflow redesign or systematic value generation. ESG’s 60% competent-or-above rating places the function 9th of 11 on this metric, above only CSR (56%) and supply chain (31%).
Public generative AI shows the strongest competence base: 61% rate their teams competent and 11% advanced, reflecting greater access and self-directed practice. Enterprise AI tells a more demanding story: 45% competent but 44% beginner, and 11% advanced. Company-owned AI has the largest advanced cohort at 22%, concentrated among directors and senior leaders.
Advanced proficiency is rare and uneven across all three types. The function has pockets of advanced expertise, but not yet at sufficient scale to treat any of these AI categories as integral to their workflows.
The sharpest role-level finding is senior leaders’ assessment of enterprise AI proficiency. 71% of senior leaders rate their teams as beginner — compared with 42% of directors and 31% of managers. This discrepancy suggests different perceptions of what effective enterprise AI usage looks like.
Directors and managers rate their own teams more generously on enterprise AI — 45% and 56% reaching competent, respectively.
The divergence across levels suggests senior leaders may have greater visibility into the gap between tool capabilities and actual workflow integration, or that their expectations for enterprise AI performance are simply higher than those of the practitioners doing the work.
ESG’s overall proficiency sits near the benchmark average — essentially at the midpoint — but the senior leader signal on enterprise AI is the finding worth acting on. The function likely cannot close this with adoption campaigns alone. It may require role-specific training, visible use cases demonstrated, and honest calibration of what competent enterprise AI use looks like inside ESG workflows.
ESG presents a distinctive trust gradient: higher trust in the AI that is more governed, lowest trust in the AI that is most accessible. The function ranks 8th of 11 on overall trust. The mistrust drivers explain why.
ESG’s trust profile is the inverse of its usage. Rather than trusting the AI they use most (public generative AI), ESG leaders extend their highest trust to enterprise platform AI (50% moderate or significant trust) and company-owned AI (42%). Trust in public GenAI is only 23% — 10th of 11 functions. The function ranks 7th of 11 on enterprise AI trust at 50%, demonstrating that skepticism is tool-specific.
The function ranks 4th of 11 on enterprise AI trust at 50%, demonstrating that skepticism is tool-specific. It makes sense: enterprise and company-owned AI tools are typically governed, auditable, and subject to organizational policy controls that public tools are not. ESG work — emissions data, materiality assessments, regulatory disclosures, stakeholder reporting — is publicly scrutinized and legally consequential.
Interestingly, senior leaders actually show the highest trust in public AI at 25%, slightly above directors (20%) and managers (24%), though trust remains low across all role levels.
Inaccurate or hallucinated outputs concern 80% of ESG leaders around public AI — the dominant mistrust driver in the function and the highest rate in the entire benchmark. Data privacy or security concerns follow at 71%, and misalignment with internal policy at 59%. These three concerns point toward a specific cluster: AI whose outputs cannot be verified, whose data handling cannot be confirmed, and whose use may not yet be sanctioned by organizational policy.
Bias in training data was flagged by 45% for public AI, and lack of explainability by 41%. Both are substantive concerns for a function that relies on standardized reporting frameworks.
Trust and use are not aligned in ESG — and that tension is productive, not pathological. Teams are using public GenAI at daily rates that exceed expectations, yet the function ranks 8th of 11 on overall trust. That combination describes a function where experimentation is outpacing governance: practitioners are finding utility in tools that have not yet been sanctioned, reviewed, or validated for the function’s specific accountability requirements. The goal is not to close the trust gap by restricting use; it is to build the review and validation infrastructure that makes trust possible.
Compared with 10 enterprise functions, ESG’s cross-functional position is defined by a gap between ambition and behavior. On the metrics that reflect stated intent — expected usage, automation outlook, dependence — ESG ranks near the bottom. On the metric that reflects what teams are actually doing, it ranks near the top.
ESG’s benchmark position is defined by two diverging signals. On the ambition metrics — expected usage (10th/11), automation outlook (tied 7th/11), dependence (9th/11) — the function ranks near the bottom of all surveyed groups. On the behavior metric for actual daily adoption, it ranks 3rd.
That gap is the function’s defining characteristic: the lowest stated ambition in the benchmark paired with near-leading actual engagement. On proficiency, ESG sits close to the cross-functional average (60% vs. 61%). On trust, ESG ranks 8th of 11. On automation outlook, ESG’s 6% high-end expectation ties with CSR and employee experience for last place.
ESG’s cross-functional position — 3rd on actual adoption, 10th on expected usage — is the function’s paradox in this benchmark. Teams have bought into AI practice informally before leadership has built the framework to match. Members describe a state of active but unstructured engagement: real daily use, real hallucination concerns, and a sense that the unlock for ESG-specific AI value has not yet arrived. The opportunity is to build the governance and validation infrastructure that converts informal adoption into systematic, auditable practice the function can own and defend.