Choose Claude if
You want a general assistant across the firm at a published per-seat price, you have people to drive it, and you would rather supply the documents it works from.
Starch vs Claude
Claude assists the work.
We finish it.
Claude for Financial Services is a serious product with published pricing and real connectors; we are arguing about what an assistant can finish.
Facts about Claude checked
Choose Starch if
You want one recurring desk process finished and running, with the review path and the upkeep owned by someone other than your bankers.
An assistant and an operated workflow are different products.
| Axis | Starch | Claude |
|---|---|---|
| What it is | One desk's process, finished and operated by us. | A general assistant with a financial-services layer: connectors, skills, enterprise deployment. |
| The general workspace | Included. Agents, automations and apps, with a library and an inbox. | Is the product. A general assistant. |
| Built for | Coverage groups and deal teams: public finance, M&A, financial institutions. | Any organisation. Anthropic names Citi, RBC, BNY among financial-services users. |
| How it is priced | Scoped to the workflow and the desks it runs on. A build, then the running of it. | Yes. Claude Enterprise is $20 per seat plus consumption, billed annually. |
| Knowledge base | Ours to wire and keep current, from your precedent and formats. | Yours to supply. Claude connects to LSEG, FactSet, S&P Global, Morningstar and PitchBook. |
| Recurring execution | Is the product. Workflows run on a cadence unattended. | Not the shape of it. Claude runs when someone asks. |
| The output | A sourced draft in the firm's format, routed for approval. | An answer or an Office artifact, returned to whoever asked. |
| Audit and approval | Built in. Once routed for approval, an artifact is read-only. | Anthropic stages advisor outputs for human approval, the right instinct. |
| Deployment | Single-tenant, your cloud or ours. SAML SSO and SCIM provisioning, never pooled. | Enterprise deployment, on Anthropic's infrastructure or through a cloud provider. |
Claude is the strongest general model we have used. This page is about the part of a desk's week that is not a model call.
Where Claude is genuinely ahead.
It publishes its price
Claude Enterprise is $20 per seat plus consumption, billed annually, and the team tiers are public.
The connector estate is unusually deep
It reaches LSEG, FactSet, S&P Global, Morningstar, PitchBook and Moody's, plus Claude for Excel.
One instrument across the whole firm
It is a general assistant, not a finance product, so legal, HR, engineering and ops share the same seat.
Past the table
Four questions buyers ask next.
Short answers to the four questions that come after any comparison, written for this one. The last one is where we point you at Claude instead of us.
What onboarding actually involves
It is white-glove from the first meeting. We learn how the team works, which processes come round every month, and who runs them, and we use that to judge fit. If we are not the right people for it, we say so then.
If we are, we build that process, tailor it to the desk and deploy it inside your own environment, with the security work done as part of that deployment. We are forward-deployed: the same team carries the integration, the changes and the upkeep. Your bankers are not asked to run any of it.
Where your data sits
Your tenant holds the data, and it stays there. Our access is kept small. It covers what the work needs, for example redacted logs, and it can be permission-based.
The arrangement does not change from one process to the next. Everything the work touches sits inside that boundary, and that is the whole of it.
Does it differ from desk to desk?
Less than you would expect. A FIG team, a public finance team and an M&A team all get the same white-glove onboarding, the same pattern of deployment, and the same people answering when something needs fixing.
What changes is the product. Different desks do different jobs, so each gets workflows built around its own work, and that accounts for most of the difference between one desk and another. The service around them stays the same.
When not to pick us
There are two cases where we are the wrong call. The first is a workflow we do not already serve, one that does not sit on the same machinery as the processes we run. We will say so and stop there, rather than build something we do not believe will hold up.
The second is a team whose week is mostly writing documents and running research. For work like that, the right tool is a general assistant, and Claude is the best one we have used. We run one inside the Workspace, and Claude is better at it. If that describes your team, take it. Starch is for the rest of the process: pulling together the inputs a draft is built from, sourcing the numbers, sending the finished work to a named approver, and doing it all again next month without a person starting it.
Asked about Claude
Can I just use Claude instead of Starch?
For drafting and one-off research, yes. Where it stops is the repeatable process: qualifying work, assembling it against your precedent, routing it for approval, then doing it again. Independent banker testing found frontier models failing roughly half of banker-defined criteria.
Does Starch use Claude under the hood?
Starch builds on frontier models rather than training its own. Which model runs a given workflow is an implementation detail we choose per workflow and revisit as the models change, so your desk is not locked to one vendor's releases.
How much does Claude cost compared to Starch?
Claude publishes its pricing: Claude Enterprise at $20 per seat plus consumption, billed annually. A Starch engagement is scoped to the workflow and the desks it runs on, so the two are not the same kind of purchase.
Who reviews what the AI produces?
With Claude, whoever asked checks the output, and Anthropic stages advisor workflows for human approval. With Starch the approval path is part of the workflow: nothing reaches a client, issuer or regulator without a named person approving it.
Where does our data live?
Starch deploys single-tenant, in your cloud or ours. Your data is never pooled with another firm's and is never used to train a model, with access scoped per person and per team.
See it run
Keep the assistant.
Let us finish the process.
Bring the workflow your desk repeats every month, and we will show you what it looks like finished.
Book a demoSources and methodChecked 2026-09-16
Everything stated about Claude on this page comes from the material below, checked on 2026-09-16. Pricing and product detail change; confirm on their site before deciding.
- Claude pricingpublished seat and consumption pricing
- Claude for financial servicesskills, connectors and named customers
- Claude for Financial Advisorsadvisor workflows and human approval staging
- BankerToolBenchindependent benchmark built with 502 investment bankers
- BankerToolBench repositorytasks, rubric and results
Claude is a trademark of Anthropic. Starch is not affiliated with, endorsed by or partnered with Anthropic. Statements about Claude on this page come from Anthropic's own public materials on the date shown and may since have changed; confirm with Anthropic before deciding. The benchmark cited is independent research about model performance, not about Anthropic's finance product specifically. Nothing here describes what Claude cannot do.