Hebbia turns deal documents into structured financial analysis

Hebbia builds AI software for investors, bankers and other teams whose decisions depend on large document sets. Its Max analyst and Matrix workspace connect research with structured analysis and deliverables. The useful question is whether a team can trace the resulting numbers and conclusions back to the correct source, period and assumption.

01 / ProductMax and Matrix address different parts of financial work

Hebbia’s current product site combines firm documents, public filings and financial information. The site now uses hebbia.com; the earlier hebbia.ai address redirects there. Its company identity spans the products below, so buying access to Max is not a reason to treat Max as a separate vendor or research collection.

Max is presented as an analyst that can produce presentations, dashboards and financial models, including source citations in model cells. The same page describes interacting over email and converting firm processes into reusable agents. These are documented product claims; the quality of a specific model still depends on the inputs, instructions and review applied to it.

Matrix organizes analysis into rows and columns. The public example puts businesses beside source documents and analytical fields. That structure is useful when an analyst needs to compare the same question across a collection. It also provides a place to distinguish a missing answer from an unfavorable answer before either is compressed into a summary.

The two interfaces suggest complementary working styles. A team can use a structured comparison to establish the factual basis, then develop a narrative or presentation around the checked results. That is an editorial interpretation of the offer, not a claim that every customer has an identical integration between every interface. Confirm the actual account experience during a demonstration.

The investing page describes screening confidential information memoranda and virtual data rooms, synthesizing expert calls, maintaining company libraries and preparing investment-committee material. These examples explain the finance focus: documents are inputs to recurring decisions, not simply files to search. Each task needs its own definition of completeness and relevant evidence.

02 / AudienceMost useful when a research question repeats across a deal set

A private-market research team could evaluate Hebbia for a recurring first-pass comparison of authorized deal documents. An investment banking group could evaluate how it prepares a consistent company overview from filings and internal material. A corporate development team could compare acquisition criteria across a target list. In each case, experienced people must define the measures being compared.

This is a weaker fit for a person who only needs occasional answers from a handful of short files. The preparation and governance of a shared institutional workspace may outweigh that benefit. It is also unsuitable as a substitute for investment judgment: a well-presented sensitivity analysis cannot decide whether management assumptions, comparable companies or a proposed transaction are sound.

AlphaSense is a useful adjacent comparison when discovering external market intelligence is the starting need. Glean is relevant when the organization primarily wants to locate internal knowledge across applications. Hebbia is most interesting when the desired endpoint is structured financial work from a defined evidence set, rather than retrieval alone.

03 / WorkflowA proposed screening exercise keeps each financial basis visible

Consider a proposed evaluation using public filings from a small set of companies in one sector. The deliverable is a comparison table and a short briefing explaining the largest differences. This is a software evaluation rather than a recommendation to trade securities. Choose files that the team can review completely and retain the source inventory outside the generated work.

Define each column before loading the analysis. Revenue, adjusted earnings, debt and cash are not interchangeable measures, and similarly named disclosures can have different scopes. Require a reporting period, currency, unit and source reference beside every consequential number. A value stated in thousands should never enter a comparison alongside an unmarked value stated in millions.

Separate extracted figures from calculated figures and analyst assumptions. For example, a margin computed from two reported values should show the chosen numerator and denominator. A management forecast should retain its forecast label. An estimate supplied by a licensed data provider should carry that provenance. These distinctions help the reviewer identify what must be checked and what requires judgment.

Use a Matrix-style comparison to inspect the factual rows before requesting a briefing. Include deliberately awkward cases: a changed fiscal year, an acquisition that alters comparability, a missing segment disclosure and an amended filing. They reveal whether the workflow preserves uncertainty or quietly fills gaps. A graceful missing-data result is more useful than an unsupported number that looks complete.

If the account includes Max’s financial-model or presentation capabilities, ask for a deliverable based on the approved table. Inspect the exported output separately. A citation that works in the application may be less convenient after a file is shared, while formatting can conceal units or truncated notes. Another analyst should be able to reproduce the important calculations without consulting the original prompt.

Finally, repeat the exercise with one new filing and track what changes. Record whether the system updates the appropriate period, preserves checked historical values and flags changed definitions. Measure accepted fields and review effort rather than the number of generated pages. The pilot succeeds when a reviewer reaches a reliable comparison more efficiently, not when the longest briefing arrives fastest.

04 / PricingInstitutional access starts with a scoped demonstration

The public demo route offers a discussion with Hebbia’s team. The reviewed pages did not establish a numerical list tariff. A useful proposal should distinguish access to Max and Matrix, users, document scope, integrations and any usage measures. Do not infer a per-seat rate from funding announcements, customer size or prices quoted by unrelated comparison sites.

Licensed financial sources need particular attention. The homepage displays several market-data and content providers, but a provider logo does not establish the rights included in every Hebbia contract. Ask which subscriptions the customer must already hold, what can be exported and how many users may use the resulting work. A seemingly inexpensive workflow can depend on separate data entitlements.

05 / DistinctionsThe intermediate table matters as much as the final answer

The most useful distinction is the bridge between document reading and institutional deliverables. A research assistant that supplies a paragraph can be helpful, but a comparison table exposes uneven source coverage. It lets the reviewer see which business lacks a figure and which figure rests on a different definition. That makes the shape of the work itself part of quality control.

Max’s stated support for firm-styled decks and models connects analysis to formats that finance teams already circulate. Evaluate whether the generated files remain editable and whether changes survive another iteration. The product’s advantage will be smaller if reviewers must rebuild formulas, citations or layouts before the work can enter their normal approval process.

Hebbia’s security page states that documents, prompts and outputs are not used to train its models or its model providers. It describes isolated customer environments, access records, regional storage and inference in the US and EU, and dedicated-tenant options. Treat these as vendor-described controls whose contractual and configuration details need checking for the intended data.

06 / QuestionsResolve source rights, numerical fidelity and update behavior

The first open question is whether the available source set matches the proposed research universe. A demonstration built around familiar public filings may not establish support for a specialist document format, a particular licensed dataset or an incomplete data room. Ask to run the actual document types and retain the failures rather than replacing them with easier examples.

The second is how numerical errors become visible. Source citations help locate evidence, but the citation alone cannot prove that the correct period or accounting definition was selected. Include reconciliations and known answers in the sample. Decide whether the reviewer can correct a single value, annotate the reason and prevent that correction from being silently overwritten by a later run.

The third is permission continuity. A deal workspace may have stricter access than the firm’s general research library. Verify access using representative users, including someone outside the deal team, and check the exported deliverable as well as the source file. This public review did not inspect an authenticated deployment, contractual data licences or enterprise access settings.

07 / DecisionStart with a comparison that another analyst can reproduce

Hebbia merits evaluation for teams that want to turn document-heavy financial research into organized, reusable work. Begin with a narrow comparison whose important numbers can all be checked. Broaden the scope only after the team understands source access, numerical correction and the cost of producing an accepted deliverable.