Why Investment Research Software Fails to Scale, and How the Major Platforms Compare in 2026

Your research team added headcount last year, renewed the terminals, and still cannot cover the universe your portfolio managers are asking for. The shortfall is rarely effort and it is rarely budget. It sits in the architecture underneath the software.

This piece does two things. The first half sets out the nine structural reasons traditional investment research platforms stop scaling once coverage, speed, and compliance demands rise past a certain point. The second half turns those nine reasons into an evaluation framework and runs the platforms institutional teams actually shortlist against it: Bloomberg Terminal, FactSet, S&P Capital IQ, AlphaSense, Hebbia, and Orbit Insight.

We build one of those platforms, and we have been specific about what each of the others does well, because a shortlist built on vendor claims is a shortlist that gets revisited in eighteen months.

Key takeaways

Traditional research platforms fragment data across disconnected systems, which makes systematic analysis across a growing coverage universe progressively harder rather than progressively easier.

Manual document review imposes a hard ceiling on how many names one analyst can cover with real depth, and no amount of interface improvement moves that ceiling.

Per seat licensing ties research cost to headcount, which means the economics work against you at exactly the moment coverage needs to expand.

Closed architectures limit how teams customize workflows, connect internal research, or choose their own models, and they make migration expensive by design.

The category has split into platforms that own a document corpus, platforms that reason over documents you supply, and a small group doing both. Which one you need depends on which of the nine failures is actually costing you.

Part one: nine reasons research platforms stop scaling

1. Data silos fragment the workflow

Most institutional research functions run across several disconnected systems. Exchange filings sit in one platform. News arrives through another. Broker research lands in email. Internal notes scatter across shared drives, meeting tools, and individual workstations.

Fragmentation compounds with scale. Every additional source adds a login, an interface, and another place where something material can be missed. Analysts spend a meaningful share of the week reconciling systems instead of forming views.

The test is simple. Can you ask one question and have it answered across public filings, licensed third party content, and your own internal research at the same time? For most firms the answer is no, and the workaround is an analyst doing it manually.

2. Manual document review creates a coverage ceiling

A single annual report can run past five hundred pages. Multiply that by quarterly reporting, earnings transcripts, sustainability disclosures, and regulatory filings for every name in the universe, and human review stops being a scaling strategy.

The convention in equity research is that one analyst covers somewhere between ten and twenty names with genuine depth. That number has been stable for decades because it reflects reading speed, not tooling. Traditional software assumes throughput scales linearly with headcount. It does not, and the ceiling arrives well before the coverage mandate is satisfied.

3. Per seat licensing ties research cost to headcount

This is the constraint that does the most quiet damage, because it shapes behavior long before anyone calls it a problem.

A Bloomberg Terminal seat lists at $31,980 a year in 2026, falling to $28,320 on multi terminal contracts, typically on a two year minimum. AlphaSense does not publish pricing at all, and reported enterprise seats cluster near $18,000 a year with content add ons that buyers commonly describe as another twenty to forty percent of contract value. FactSet and S&P Capital IQ both sit in wide negotiated bands running from roughly $12,000 to the high tens of thousands per user.

Under any of those models, adding coverage means adding seats, and adding seats means a budget conversation. Research capacity becomes a function of what procurement will approve rather than what the portfolio requires. Teams ration access, junior analysts share logins, and the platform quietly becomes something people use less than they should.

Usage based pricing breaks that link. When cost tracks queries rather than headcount, an analyst who runs one search a week costs approximately one search a week, and the decision about who gets access stops being a budget decision.

4. Filing latency costs alpha on event driven strategies

When a company files material information, the market reprices within minutes. Research platforms that deliver filings on a batch cycle hand you the document after the move has happened.

For fundamental long only work this rarely matters. For event driven, merger arbitrage, and systematic strategies reading filings as a signal, it is the whole game. The question to ask a vendor is not whether they have the filing. It is how long after publication it becomes queryable, and whether that latency is contractual or aspirational.

5. Global coverage gaps leave blind spots

North American and Western European filings are table stakes. The gaps open elsewhere.

China A-Share disclosure is the clearest example. Mainland listed companies publish material that has no equivalent in Western markets, including mandatory disclosure of institutional on site research meetings, where the company records the questions asked by visiting brokers and asset managers. That content is published in Chinese, on exchange systems built for domestic participants, and most global platforms either skip it or surface a thin translated subset.

Emerging market bond documentation, smaller European exchanges, and APAC regulatory filings show similar patterns. A global mandate assessed on partial coverage is a global mandate assessed on partial evidence, and the gap does not announce itself. It shows up as an absence in a screen that looked complete.

6. Opaque scoring fails compliance review

Sustainability analysis moved from optional to regulated. CSRD, SFDR, and the UK Sustainable Disclosure Requirements all demand that a firm explain its reasoning.

Many platforms answer with a proprietary rating and no visible methodology. When an auditor asks how a sustainability assessment was reached, the honest answer is that the calculation happens inside a vendor system nobody at the firm can inspect. That fails review, and it fails it late, after the position is taken and the report is published.

Firms with their own methodology have a second problem. A vendor score they cannot decompose is a vendor score they cannot reconcile against their own framework, so they end up paying for a number they then rebuild by hand.

7. Unstructured content stays unread

The material insight usually sits in unstructured form. Management commentary on a call. A risk factor that changed wording between filings. A tone shift across four quarters of disclosure. Tables buried in a PDF appendix that never reach any structured feed.

Traditional platforms handle structured data well and treat everything else as documents to be retrieved and read by a person. That is exactly the layer where scale breaks, because retrieval does not reduce reading time. The platforms that solve this parse the document rather than index it, extracting tables, financial language, and section structure into something queryable before an analyst ever opens it.

8. Scaling infrastructure means scaling headcount

Follow reasons two and three to their conclusion and you get the core economic problem. Doubling the coverage universe under a traditional model means roughly doubling the analyst team or accepting materially shallower work across the whole universe.

Neither is acceptable, so most firms do a third thing. They quietly narrow the universe and call it focus.

9. Integration is treated as an afterthought

Modern research teams need the research layer to connect to portfolio management systems, risk engines, internal knowledge stores, and increasingly to their own AI assistants. Legacy platforms were built as destinations, so connectivity arrives as custom development priced per project.

This has become the fastest moving part of the market. Through late 2025 and 2026, several vendors shipped Model Context Protocol servers, which let an AI assistant query the vendor's data directly rather than through a person copying answers between windows. FactSet announced a production grade MCP server in December 2025 covering nine datasets and extended it to portfolio analytics in June 2026. S&P Global exposes Capital IQ financials and transcripts through a hosted endpoint built by Kensho. Bloomberg helped fund the foundation that now governs the protocol and has not shipped a production implementation for the Terminal.

Two questions separate real openness from announced openness. Is the server hosted and operated by the vendor, or is it a code sample you run yourself? And is it included in the product, or gated behind an enterprise tier and a separate conversation with an account team?

Part two: turning nine failures into an evaluation framework

Those nine reasons map onto six questions worth asking every vendor on a shortlist.

Coverage. How many companies, how many countries, how many exchanges, how far back, and specifically what happens outside the United States and Western Europe?

Latency. How long after publication does a filing become queryable, and is that a commitment or an average?

Transparency. Can every answer be traced to a source document, and can a proprietary score be decomposed into its inputs?

Pricing model. Is cost tied to seats, to usage, or to data volume, and is the price published?

Openness. Is there a real API, a hosted MCP server, a choice of model, and a deployment option that satisfies your security team?

Internal data. Can the firm's own research be indexed alongside external content, and under what data governance terms?

Part three: how the major platforms compare

The category divides into three groups, and the group a vendor belongs to predicts most of its answers to the six questions above.

Group one: the terminals and data incumbents

Bloomberg Terminal is the industry default and the network effect is real. Much of the buy side and sell side communicates and transacts over it, which means the value is not only the data.

It scales on breadth across asset classes, real time market data, execution, and messaging. Nothing else consolidates that much into one product.

It stops scaling on economics and on documents. Pricing is uniform per seat regardless of how much of the product a user touches, at $31,980 a year in 2026 with contract minimums typically running two years, so the bill is identical whether a user relies on two functions or two hundred. Document intelligence over unstructured filings is not what the Terminal was built for. Despite backing the foundation that governs the Model Context Protocol, Bloomberg has not shipped a production MCP implementation for the Terminal, so programmatic access into AI workflows remains constrained.

Best fit: trading desks where messaging, real time data, and execution are the primary jobs.

FactSet has moved faster on AI connectivity than any other incumbent. It serves more than nine thousand clients and over 241,000 users, and announced the first production grade MCP server in financial data in December 2025, spanning nine datasets including fundamentals, global M&A, and supply chain. A portfolio analytics MCP followed in June 2026.

It scales on estimates, consensus, portfolio analytics, and performance attribution, and the MCP work means that data now reaches an AI agent without a custom integration project.

It stops scaling on the same seat economics as the rest of this group, with pricing negotiated per user in a band commonly running from around $12,000 into the tens of thousands. Coverage of unstructured disclosure outside major markets is thinner than the structured data coverage suggests, and the MCP surface covers curated datasets rather than the underlying document corpus.

Best fit: buy side firms whose primary need is portfolio analytics and estimates.

S&P Capital IQ brings depth in fundamentals, credit, private company data, and commodities, and exposes Capital IQ financials and earnings transcripts through a Kensho built endpoint that AI assistants can query directly.

It scales on financial history, credit analytics, and private company coverage, which are difficult to match.

It stops scaling on per user pricing that typically lands between roughly $12,000 and $30,000, and on a platform that still reflects its origins as a set of separately built products. Coverage of unstructured non financial disclosure is not its strength.

Best fit: credit teams and private markets analysis.

Group two: search and reasoning platforms

AlphaSense is the largest pure play in this category. The company raised $350 million in June 2026 at a $7.5 billion valuation, on roughly $600 million of annual recurring revenue and more than seven thousand enterprise customers. Its library exceeds 500 million documents, largely licensed from third parties, and the Tegus acquisition brought a substantial expert call inventory.

It scales on search quality across Western filings, transcripts, and broker research. The expert call library is a genuine asset with no equivalent at most competitors.

It stops scaling on pricing, on customization, and on coverage outside its core markets. Pricing is quote only with no public list and no self serve tier, with reported enterprise seats near $18,000 a year before add ons that buyers commonly describe as another twenty to forty percent of contract value. Seats are sold in bands, so adding a small number of users can move a firm into a higher tier. On connectivity, AlphaSense offers an Agent API and MCP tooling, though its own developer documentation describes the MCP server as a locally hosted reference implementation rather than a production service it operates. Teams wanting agents built to their own logic will find the platform prescriptive, which is the reason firms with specific methodologies, including CANDRIAM, have selected alternatives on customization grounds.

Best fit: firms that want strong search across Western filings and expert content, and that can absorb per seat pricing.

Hebbia is built around the Matrix interface and a multi agent approach the company calls iterative source decomposition, backed by roughly $161 million from Andreessen Horowitz, Index Ventures, and Google Ventures.

It scales on running one analytical question across hundreds or thousands of documents at once and returning an auditable grid rather than a paragraph. For diligence, data room review, and credit agreement work, that shape fits the task well.

It stops scaling at the data layer, because there is not one. Hebbia reasons over documents you supply or content licensed from others, which is why it has announced integrations with FactSet, Third Bridge, and Preqin. If the problem is that you cannot reach the underlying documents in the first place, particularly outside the United States, a reasoning layer does not solve it. You buy the data separately and pay twice.

Best fit: investment banking and private markets teams working primarily on documents they already hold.

Group three: platforms that own both the data layer and the reasoning layer

Orbit Insight is an award winning AI investment research platform built for institutional buy side teams. It is the one platform on this list that built the document infrastructure and the reasoning layer together, which is what allows it to answer the six framework questions without pointing at a partner.

On coverage, the platform processes 70 million documents a year across more than 75,000 companies, 120 countries, and more than 80 exchanges, spanning 308 document types parsed at better than 99 percent accuracy. That includes 5,500 China A-Share names with the institutional on site research transcripts described in reason five, which are not available through any other vendor, alongside global earnings transcripts, sustainability disclosure, and regulatory material across more than 250 regulatory bodies in 65 languages.

On latency, filings are monitored across more than 100,000 sources and become queryable within minutes of publication rather than on a batch cycle.

On transparency, every answer traces to its source document, and Orbit delivers extracted metrics rather than opaque scores, so firms produce their own numbers against their own methodology. That is the difference between passing a CSRD or SFDR audit and rebuilding a vendor score by hand, and it is why sustainability teams at large asset managers use Orbit to produce reported metrics rather than to buy someone else's rating.

On pricing, Orbit publishes its prices. Plans run from £75 a month through £200 and £500 tiers, priced by credits consumed rather than by seats, with API and Orbit MCP access included at every tier. Institutional agreements are modular, so firms pay for the capabilities they use instead of an all or nothing bundle. This is the direct answer to reason three, and it is the reason a growing number of firms run Orbit alongside an incumbent rather than replacing one expensive bundle with another.

On openness, Orbit operates a hosted MCP server included at every tier, a full REST API covering entity lookup, document discovery, parsed content, and question answering, and supports cloud, private cloud, and on premises deployment for firms whose security review rules out anything else. Orbit is the first partner agent on SimCorp One's Agent Launchpad. In benchmarking, an AI assistant connected to Orbit MCP scored materially higher on financial research tasks than the same assistant without it.

On internal data and customization, Orbit Agent Builder lets analysts create their own research agents with no code and no engineering support, defining what an agent reads, what it extracts, and how it reports, then running it on demand, on a schedule, or on a filing trigger. The Agent Marketplace supplies prebuilt agents that can be opened and edited rather than accepted as given. Internal documents including broker research, meeting notes, and investment memos are indexed alongside external content in the same queryable layer.

Orbit is deliberately not a trading system. There is no real time market data, no execution, and no messaging network, because those are a different job and the Terminal does them well. Orbit is built for the research layer.

Best fit: buy side research teams whose bottleneck is unstructured document coverage and workflow automation, firms with global or APAC mandates, quant and data teams that need programmatic access, and sustainability teams that need to produce their own metrics from primary disclosure.

At a glance

Published pricing. Orbit publishes its prices. Bloomberg, FactSet, S&P Capital IQ, AlphaSense, and Hebbia do not.

Pricing model. Orbit charges by credits consumed. Bloomberg, FactSet, S&P Capital IQ, and AlphaSense all charge per seat. Hebbia prices by enterprise agreement.

Indicative cost. Orbit starts at £75 a month. Bloomberg lists at $31,980 per seat per year. AlphaSense enterprise seats are reported near $18,000 before add ons. FactSet and S&P Capital IQ both negotiate in bands from roughly $12,000 per user upward.

Proprietary document corpus. Bloomberg, FactSet, S&P Capital IQ, AlphaSense, and Orbit each maintain one. Hebbia does not and relies on licensed partners or documents you supply.

China A-Share depth. Orbit carries full coverage including institutional on site research transcripts. Every other platform on this list carries partial coverage or accesses it through licensed sources.

Hosted MCP server. FactSet, S&P Global, and Orbit operate one. Orbit includes it at every pricing tier. AlphaSense provides a locally hosted reference implementation. Bloomberg and Hebbia have not shipped one.

Custom agent building. Orbit and Hebbia support building agents to your own logic. The others are limited to configured workflows.

On premises deployment. Orbit supports cloud, private cloud, and on premises. AlphaSense supports deployment into AWS and GCP under its enterprise tier. The rest are effectively cloud only.

Real time market data and execution. Bloomberg leads and no AI research platform on this list competes with it.

Expert call library. AlphaSense leads through the Tegus acquisition. Hebbia accesses expert content through partners.

Part four: choosing between them

If the problem is real time data, execution, and market communication, the Terminal remains the answer and no AI platform on this list changes that.

If the problem is estimates, consensus, and portfolio analytics, FactSet or S&P Capital IQ solve it directly, and both now expose that data to AI agents.

If the problem is reasoning across documents you already hold, particularly in deal work, Hebbia fits the shape of the task.

If the problem is search across Western filings and expert calls, and per seat pricing is workable, AlphaSense is the established option.

If the problem is that the documents themselves are missing, that coverage stops at the edge of developed markets, that research capacity is capped by how many seats finance will approve, or that your team needs to build its own agents and pipe research into its own systems, that is the problem Orbit was built to solve, and it is the only platform here that addresses all four at once.

Most institutional teams run two of these rather than one. The useful question is which two, and which is doing the heavy lifting.

Frequently asked questions

What causes investment research software to fail at institutional scale?

Three things compound. Data sits in disconnected systems, so nothing can be asked across the whole evidence base at once. Document review depends on human reading speed, which caps coverage per analyst regardless of interface quality. And per seat licensing ties research capacity to headcount, so expanding coverage requires a budget approval rather than a configuration change. Platforms that parse documents into a queryable layer and price by usage rather than by seat remove all three constraints at the same time.

How many companies can one analyst realistically cover?

The long standing convention in equity research is ten to twenty names with genuine depth. That figure has been stable for decades because it reflects how fast a person can read and form a view, not what software is available. Automated monitoring changes what falls to the analyst rather than raising their reading speed, so the gain comes from what gets surfaced for attention rather than from faster reading.

Why does per seat pricing become a problem as a research team grows?

Per seat models make the marginal cost of another person looking at the data equal to a full license. At Bloomberg's 2026 rate of $31,980 a seat, or AlphaSense's reported figures near $18,000 before add ons, teams start rationing access. Credit based models decouple cost from headcount. Orbit publishes plans from £75 a month priced by credits consumed, so a junior analyst running one query a week costs approximately one query a week.

What filing latency should institutional teams expect?

Modern platforms process filings within minutes of publication rather than on a batch cycle. Orbit monitors more than 100,000 sources and makes filings queryable within minutes. Ask any vendor for the measured time between publication and the document becoming queryable, and ask whether that figure is a commitment or an average.

Which investment research platforms offer an MCP server?

FactSet announced a production grade MCP server in December 2025 covering nine datasets, extended to portfolio analytics in June 2026. S&P Global exposes Capital IQ financials and transcripts through a Kensho built hosted endpoint. Orbit operates a hosted MCP server and includes it at every pricing tier rather than gating it behind an enterprise agreement. AlphaSense provides an Agent API with MCP tooling, though its documentation describes the MCP server as a locally hosted reference implementation. Bloomberg helped fund the foundation governing the protocol and has not shipped a production implementation for the Terminal.

How do compliance requirements affect research platform choice?

Regulation across CSRD, SFDR, and the UK Sustainable Disclosure Requirements demands that a firm explain its reasoning with an audit trail. Proprietary scores that cannot be decomposed into their inputs fail that test. Platforms with full source attribution and extracted metrics rather than opaque ratings meet the requirement without manual documentation, and let firms with their own methodology reconcile against their own framework instead of rebuilding a vendor number by hand.

What China coverage should a global mandate expect?

Beyond standard exchange filings, mainland listed companies disclose institutional on site research meetings, recording the questions asked by visiting brokers and asset managers. Between seventy five and eighty percent of covered companies publish that material within three days of the meeting. Most global platforms do not carry it. Orbit covers 5,500 A-Share names with history from 2012, including those transcripts in original Chinese and professional English translation. For any mandate with meaningful A-Share exposure, this is the coverage question to ask.

Should we build an internal research system instead of buying one?

The reasoning layer is now the easy part. What is hard, and what most internal projects underestimate, is acquiring and maintaining the document corpus: monitoring filing sources across dozens of exchanges, parsing inconsistent formats at reliable accuracy, resolving entities across markets, and keeping all of it current every day. Firms that build usually end up licensing the data anyway, at which point the calculation is whether the interface layer justifies the engineering team maintaining it. Platforms that expose their corpus through an API and an MCP server let teams build their own interface on top of infrastructure they do not have to run.

Can AI research platforms be deployed on premises?

It varies and it is worth asking early, because security review is where deals stall. Orbit supports cloud, private cloud, and on premises deployment. AlphaSense Enterprise Intelligence supports deployment into AWS and GCP. Most other platforms on this list are effectively cloud only.

Can analysts build their own research agents?

On most platforms, no. Configured workflows and prebuilt templates are common, and genuine agent building is rare. Orbit Agent Builder lets an analyst define what an agent reads, what it extracts, and how it reports, with no code and no engineering support, then run it on demand, on a schedule, or triggered by a new filing. Every agent in the Orbit Agent Marketplace can be opened and edited rather than accepted as supplied.

Where this leaves you

The category divided into three groups. Incumbents added AI to terminals built for a different job. Reasoning platforms built strong interfaces without owning the documents underneath. A smaller group built both.

Which one you need depends entirely on which of the nine failures above is actually costing you. If it is execution and market data, buy a Terminal. If it is reasoning over documents you already hold, buy a reasoning layer. If it is that the documents are missing, that coverage stops at the edge of the developed markets, or that your research capacity is capped by how many seats finance will sign off, you need infrastructure rather than another interface.

These are conversations that matter deeply to us at Orbit. If you want to test any of the claims in this piece against your own coverage universe, we would rather you did that than take our word for it.