Survey finds AI core to Wall Street research workflows

Hebbia has published survey findings that suggest AI is widely used across Wall Street research workflows, alongside ongoing concerns about verification, security and integration.

The report is based on an online survey of 529 deal and investment focused finance professionals at investment banks, asset managers and advisory firms. It found that 93% of respondents either use AI or are evaluating it in some capacity, but only 25% said it was fully integrated across their firm or team.

Time savings emerged as a central theme. Nearly 63% of respondents said they save more than six hours a week using AI, including 27% who reported saving more than 10 hours each week on research and analysis tasks.

Those gains come against heavy information workloads. Respondents cited data volume and fragmentation as major bottlenecks: 46% said reading large volumes of documents is a primary challenge, another 46% said they struggle to extract information from multiple sources, and 38% reported difficulty searching across disconnected systems and sources.

Core workflows

The survey suggests AI is moving beyond experimentation and into core institutional finance tasks. Some 61% of respondents said they use AI to build financial models, while 60% said they use it to read and summarise filings and transcripts.

Perceptions of output quality also point to growing reliance on machine-generated drafts. Around 39% of respondents said AI produces higher-quality work than a junior analyst's first draft.

Even so, respondents did not broadly expect AI to eliminate roles. Only 8% said they expect AI to replace their job outright. Another 26% said it will take over routine workflows, shifting their role toward higher-level judgement and decision-making.

George Sivulka, Hebbia's chief executive officer and founder, linked finance's fast adoption of AI to the high value of accurate answers in markets and transactions.

"Finance will consume AI capabilities faster than any other vertical because the reward for the correct answers is enormous," Sivulka said.

Trust barrier

The report points to trust and verification as key limits on wider deployment. It found that 85% of respondents are at least somewhat confident in AI's factual accuracy when answers are grounded in source documents. That confidence drops when users cannot easily verify outputs.

Nearly half of respondents said rapid verification is essential to trusting AI-generated work. In the headline findings, 49% identified fast verification as critical.

The survey also asked what prevents broader, firm-wide use. Data security and confidentiality concerns ranked highest at 29%, followed by integration challenges with existing systems and data sources at 21%.

Accuracy risks tied to hallucinations came next at 16%, while 11% cited a lack of internal training or expertise.

The report frames these barriers as structural rather than driven by a lack of interest. Many finance professionals already use AI tools day to day, but large organisations face friction when trying to standardise use across teams and systems.

Infrastructure focus

Hebbia cast the next phase of adoption as an infrastructure challenge: tools that can handle private documents while providing traceability. It said users want citations, transparency and auditability, alongside controls that meet governance and compliance expectations in institutional settings.

The report also cites an external study as a caution for deal teams. It references a McKinsey analysis of generative AI in private markets, which found that public large language model reports were systematically more optimistic than expert-interview research and diverged on metrics such as market size and growth. The analysis also found omissions in areas including contract structures, unit economics and regulatory hurdles.

Alongside the survey results, Hebbia outlined its approach to handling customer data, saying it follows Zero Data Retention, meaning customer data is not stored, reused or used to train models.

Hebbia said the industry is moving toward wider adoption following early experiments, with demand expected to centre on verifiable research outputs and systems that integrate with existing data environments.