An AI-native alternative to FactSet, built for how your analysts actually work.
FactSet is a genuinely useful research workstation. It is also priced per seat, generic across every client, and structurally incapable of doing the one thing that matters most in 2026 — reasoning over your fund's private thesis process. There is a better shape for research spend.
1 — What FactSet actually is
FactSet is an integrated research, analytics and portfolio-analysis platform used by tens of thousands of finance professionals worldwide. It bundles high-quality fundamentals data, consensus estimates, ownership data, quantitative screening, portfolio analytics, charting and increasingly AI-driven document search. Vendr's marketplace data puts the average FactSet contract value at roughly $25,000 per user per year, in line with industry reporting.
For sell-side equity research, banking, corporate strategy and long-only asset management, FactSet is often the tool of choice — priced meaningfully below Bloomberg but with a workstation experience that fits analytical work well.
For a hedge fund in 2026 the question is different. Analysts spend a huge share of their time on tasks — reading transcripts, comparing filings, summarising expert calls, cross-checking consensus vs. proprietary views — where the productive work is not "look up a fact in FactSet" but "reason across everything my firm knows and everything the world knows." That is a different problem, and it is not one FactSet was designed to solve.
2 — What FactSet costs you
FactSet is more transparent than Bloomberg but the real cost varies significantly with configuration. Public data points give a defensible range.
| Line item | Order-of-magnitude cost | Note |
|---|---|---|
| Average FactSet contract value (per seat, per year) | ≈ $25,160 (Vendr marketplace, 2026) | Aggregate across configurations; not a single list price. |
| Workstation with Estimates / Analytics | Custom / per user / year | FactSet does not publish list prices publicly. |
| Data feed licences | Not publicly disclosed | Separately priced by data category. |
| Portfolio analytics / risk add-ons | Not publicly disclosed | Module-driven. |
| Compliance archiving / integration | $500 – $2,000 / user / year | Depending on stack. |
| Analyst time still spent on manual synthesis | Hidden but material | The largest cost is rarely on the invoice. |
For a 30-analyst research team, direct FactSet spend of $750k–$1.2m+ per year is unremarkable. What almost no fund models is the labour cost of analysts doing manual synthesis on top of FactSet — reading transcripts, cross-comparing filings, writing summary memos. That is the cost an AI-native research layer collapses.
3 — What FactSet can't do for your firm
FactSet's AI features are improving. So is Bloomberg's. So is every vendor's. The structural problem is not "FactSet's AI isn't good enough" — it is that any vendor-hosted AI is, by definition, generic across every fund, and cannot reason over your private research process.
It cannot reason over your private thesis process
Your analysts' memos, pre-mortems, IC notes and internal Slack are the highest-signal research asset in your firm. FactSet has no legitimate access to it. A firm-owned research layer treats that corpus as central.
It cannot draft in your firm's voice
Every fund's research notes and IR letters have a house style. A vendor model — trained on the entire internet — writes in nobody's voice. A firm-owned model, grounded on your own memos, writes in yours.
It cannot connect a thesis to its evidential trail
"When did we first form this view, what evidence supported it, how has it evolved?" is a first-class question for a hedge fund. FactSet can search public documents; it cannot maintain the firm's evidential memory.
It is priced per seat rather than per workflow
AI-native research replaces analyst hours, not analyst seats. A per-seat pricing model is misaligned with the direction of the work: fewer, higher-leverage analysts, each supported by systems that do the manual synthesis in seconds.
It cannot be extended by your engineers
FactSet has APIs, but it is fundamentally a workstation. A firm-owned research layer is a codebase your team owns, extends and evaluates continuously.
4 — Why 'your own version' is viable in 2026
Frontier LLMs are now genuinely capable of the kind of financial reasoning that used to define a mid-level analyst's job — reading and synthesising transcripts, cross-comparing filings across years, extracting structured data from long PDFs, generating first-draft memos in a defined house style. When those capabilities are combined with your firm's private corpus and a governed evaluation harness, you get something no vendor workstation can be: a research layer that is uniquely yours.
This is not "AI will replace analysts." It is the opposite. It is analysts operating at the leverage of a team three times their size, with the boring parts of the job compressed and the interesting parts amplified. That is the specific shape of edge available in 2026 — and it is why the smartest funds are quietly redirecting research spend from more workstations to fewer, higher-leverage analysts plus a firm-owned AI layer.
5 — Reference architecture
Firm-owned AI research layer above FactSet (or a slimmer research stack)
FactSet becomes one source among several — you keep the seats where FactSet is genuinely the best fit for a workflow and you retire the seats where the AI layer has replaced the workflow.
The private corpus is the point. Every analyst memo, IC decision and pre-mortem becomes retrievable, groundable evidence that the AI layer reasons over. Nothing an analyst reads on a surface comes from a hallucination — every generation is anchored to specific documents in your corpus.
The surfaces are deliberately narrow. An analyst copilot that drafts the first version of every memo in the firm's voice. A morning brief tailored to each analyst's coverage. A thesis diff viewer that shows how a view has evolved. An IC prep pack generated in minutes rather than days.
6 — Build vs. rent: 3-year TCO
Three-year illustration for a 30-analyst equity research team currently on full FactSet coverage.
| Dimension | Rent (FactSet) | Build (KJ Capital) |
|---|---|---|
| FactSet seats | 30 × ~$25k = $750k / yr | 12 × ~$25k = $300k / yr |
| Firm-owned research layer | N/A | £400k – £700k build; ~£250k / yr run |
| Analyst time on manual synthesis | Large hidden cost — hours per analyst per week | Compressed by 60–80% on target workflows |
| House-voice drafting | Every analyst writes from scratch | First draft generated in seconds in firm style |
| 3-year total (illustrative direct) | ≈ $2.25m + opportunity cost | ≈ $1.5m – $2.0m + owned research asset |
| IP owned at year 3 | Zero — you paid rent | A compounding, evaluated, firm-specific research layer |
The point of the exercise is not that FactSet is expensive. It is that the shape of research spend is shifting. Every year you delay building the layer, your competitors compound theirs.
7 — The three honest risks of building your own
"Our analysts will resist AI-drafted memos."
How we solve it —Every fund we work with is initially sceptical. The trick is drafting, not deciding. The AI writes the boring first draft grounded on the firm's evidence; the analyst edits, adds judgement, and signs off. Analysts stop resisting the moment they realise they get their evenings back.
"We can't evaluate whether the AI is right often enough to trust it."
How we solve it —Correct — and that is why the evaluation harness is a first-class part of the build. A curated golden set of research questions, run continuously, is the difference between an AI system a PM will actually use and one they will quietly ignore.
"If we build it, we own the pager."
How we solve it —That is exactly what our Operator engagement is for. We run production for the first 12–18 months with your team on the rota; by month 18 they own it. This is a build-then-transfer model, not permanent dependency.
8 — The CTA ladder
The gap between funds that build an AI-native research layer in 2026 and funds that don't will widen every year. Not because the tooling is exotic — because it isn't — but because the compounding advantage of a firm-owned research asset is real.
Readiness → Audit → Blueprint. In two weeks you'll know exactly what a firm-owned research layer would look like on top of a rationalised FactSet footprint.
Free Readiness Score
5 minutes. Honest score of whether your firm is in a position to build.
AI Diagnostic — £15k / 2 weeks
Founder-led. Map the specific systems where you'd get compounding return.
Blueprint conversation
30-minute call with Kasim to scope the Build.
Five questions we get asked most.
Are you saying we should drop FactSet?+
No. FactSet is a good research workstation and for many analysts remains the right tool. The argument is to rationalise the seat count to the roles that genuinely depend on it, and to redirect the freed spend into a firm-owned AI-native research layer.
How much does FactSet cost per user per year?+
Vendr's marketplace data puts the average contract value at approximately $25,160 per user per year, though FactSet does not publish list prices publicly and actual per-seat cost varies significantly with configuration.
Will AI-drafted memos really be trusted by the IC?+
Only if the AI drafts and the analyst decides. Every memo is grounded in the firm's own evidence with source citations, and analysts edit, add judgement and sign off. The IC never sees a memo the analyst has not owned.
How long until this is live?+
Two-week Diagnostic, 6–12 weeks initial Build, first production surface (usually analyst copilot) live within a quarter of signing.
What about FactSet's own AI features?+
They are useful for generic tasks. What they cannot be — because FactSet does not have access to your private corpus and cannot legitimately train on your firm's proprietary process — is a research layer that reasons in your firm's voice over your firm's evidence.