Executive summary
An add-in helps with the file that is open. A workspace holds the deal across six weeks and four people.
- Where the work lives. Model ML runs in Excel, PowerPoint, Outlook, its own app and your firm’s systems. Help arrives wherever you are, but the deal itself stays scattered across those places. In ArcaLabs the deal is the unit: one file tree, version history on every file, and your firm’s definitions applied throughout.
- Checks that stop the file. ArcaLabs finds hardcoded numbers where a formula should be and converts them, recalculates the workbook, and checks it for errors. Decks are rendered and compared against your template. A failed check stops delivery.
- Sourcing you can check after you send it. Every number carries a note saying where it came from, written into the workbook. Click a figure and the filing opens at that number. Model ML publishes no detail on how its outputs are sourced or audited.
Both put AI in front of an analyst. Only one of them can tell you where every number in the model came from.
Where the work lives
This is the real choice between the two, and it is not about features.
Model ML
Six parts: Workflows, Grids, Document Review, Chat, a Notetaker and dashboards, delivered inside Excel, PowerPoint, Outlook, its own app or your firm’s systems. The cost is that the deal has no single home. Versions, definitions and the record of what was sent live wherever each person left them.
ArcaLabs
Open a deal and you get the agent in the chat, live files in the editor and the deal tree on the right. Every file is versioned. Your firm’s definitions, like how you calculate adjusted EBITDA, are recorded once and applied to everything built after that. The same Microsoft tools connect: SharePoint, OneDrive, Outlook and Teams, and what comes out is real Excel, PowerPoint and Word. Six weeks later a new person opens the deal and sees the whole thing.
Checks before the file goes out
Any competent model will draft a workbook. The question is what happens between the draft and the version you send.
In ArcaLabs three checks run on their own, every time:
- Hardcoded numbers where a formula should be are found and converted.
- The workbook is recalculated and checked for errors.
- The deck is rendered and compared against your template.
Anything short of a pass stops delivery.
Sourcing and market data
In ArcaLabs every number carries a note saying where it came from, written into the workbook. Click a figure and the filing opens in the app at that number. Two rules keep the trail honest:
- If a company reported the number, the tag links to the filing, not to a data vendor’s copy of it. Diluted shares, total debt, cash, the LTM actuals under an EV bridge.
- Numbers that only exist on a data vendor, like a share price or a consensus estimate, name the vendor and the date instead, and are never dressed up as coming from a filing.
On data, ArcaLabs is FactSet’s Official AI Partner and the license is included in the seat price. Model ML connects to internal and external sources without naming a data partnership or saying who holds the license, so ask, and put the answer in your year-two numbers.
The deal record
Model ML reaches into Outlook and captures meetings with a notetaker, which helps the person doing the work. It is not the same as a record of the transaction.
In ArcaLabs the deal log works with the email you already use. Connect Gmail or Outlook and correspondence is matched to the right deal automatically, so the record builds itself without anyone forwarding anything. Each deal also gets its own address that counterparties can write to directly, and the agent drafts the reply there for you to approve. Every email, meeting and agent action lands in one deal log you can export.
The difference shows up when compliance asks what went to which counterparty and when. One answer is assembled from several inboxes. The other is already written down.
Where they differ
| Area | ArcaLabs | Model ML |
|---|---|---|
| Unit of work | The deal, with a file tree, version history and firm definitions | The document or task in front of you |
| Checks before delivery | Live formulas, workbooks recalculated and checked, decks compared against your templates | Not published |
| Sourcing | On every number. Click it and the filing opens at that figure | Not published |
| After you export | The notes are inside the workbook, so they survive the email | Not published |
| Reported vs estimated | A filing beats a data vendor’s copy. Vendor-only numbers name the vendor and date | Not addressed |
| Market data license | Data agnostic. All licenses (e.g. FactSet, CapIQ) included in the seat price | Connects to external sources. No data partnership named |
| Deal record | An email address per deal, drafted replies, one deal log you can export | Outlook and a meeting notetaker. No deal record |
Conclusion
An assistant in every window is not accountability for the file. The version that went out is wherever the last person saved it.
Files fail review in four places. A number was pasted in rather than calculated. The template drifted. An estimate was footnoted as if a company had reported it. Or someone asked where a figure came from and the answer took four minutes to find.
ArcaLabs checks all four before the file reaches you, keeps the sourcing inside the file so it holds up after you send it, and publishes exactly how each check works. Give both platforms three files from a deal you have already closed, ask for the same workbook and the same deck, then export them both and click a number.
Run the same test on your own deal.
Bring three files from a closed deal. Ask for the workbook and the deck, export them, and click a number.
Frequently asked questions
What is the difference between ArcaLabs and Model ML?
Model ML adds an assistant to the tools you already have open: Excel, PowerPoint, Outlook, its own app and your firm’s systems. ArcaLabs gives the deal one home, with one file tree, version history on every file and your firm’s definitions applied throughout, and checks the work before it leaves.
Does Model ML check a workbook before it goes out?
Model ML publishes no equivalent check. In ArcaLabs three checks run on their own every time: hardcoded numbers are found and converted to formulas, the workbook is recalculated and checked for errors, and the deck is rendered and compared against your template. Anything short of a pass stops delivery.
How is each number sourced?
In ArcaLabs every number carries a note saying where it came from, written into the workbook. Click a figure and the filing opens at that number. Model ML publishes no detail on how its outputs are sourced or audited.
Sources
What Model ML does is described here from that company’s public material, read from modelml.com on August 5, 2026, and put in our own words. Where this page says something is not published, it means we could not find it stated publicly. Vendor material changes often, so check anything that will decide your choice with them directly. ArcaLabs statements are verified against the platform before publication; single-tenant VPC deployment is available on request rather than standard.
Model ML: modelml.com · ArcaLabs: arcalabs.ai