AngelList customer story

How AngelList automated the work APIs couldn't reach

30–40 hrssaved every week
24/7running in their infrastructure

“We've saved dozens of hours every week on things where APIs don't exist. Stepwork is running 24/7 in our infrastructure, so it's working even when we're sleeping.”

Gabe WiederBusiness systems and automation lead
Alice WoChief of staff
Andres Ramirez JrInformation technology engineer

About AngelList

The platform behind a large share of US startup investing

AngelList is where funds, startups, and investors raise, invest, and manage capital. Behind that platform sit two teams that carry a constant load of repeatable work: an IT team handling access and provisioning for a fast-moving company, and a fund operations group handling the document work that private funds run on: incoming mail, agreements, certificates.

Both teams had the same problem in different clothing. Requests arrived cleanly. Finishing them still meant a person opening an app and clicking through it.

The problem

The queue described the work. A person still had to finish it.

Half the queue was access work, and connectors could not reach all of it.

We went through 3,837 service-desk tickets from October 2024 to June 2026 before automating anything. The pattern was consistent: most requests weren't questions, they were actions. Grant me access, change my role, remove me.

3,837Tickets analyzedOct 2024 – Jun 2026
1,968Access and provisioning51% of all tickets
503Hand-run by ITNo connector

Where identity tooling stopped

Identity tooling only reaches the apps that expose a connector. Everything else was hand-run by IT in the ticket queue: the same clicks, over and over.

Figures are approximate. Categories are derived from ticket subject, description, and tags; about 25% remain uncategorized.

44%No connector, hand-run by IT503 requests
56%SCIM-capable, auto-provisioned633 requests

Of those, 1,136 tickets named a specific app. This is how they split.

How it works

A ticket that resolves the work, not just describes it

Their ticketing tool stayed exactly where it was, as the place requests come in and get coordinated. Stepwork slotted in behind it as the layer that carries a request through to done, across APIs where they exist and straight through the interface where they don't.

Ticket received

Request validated

Work executed

Status returned

API

Human

API

Serval

Stepwork

TodayHanded to a person, who finishes the change by hand.

With StepworkHanded to Stepwork, which returns a governed result.

The same request, both ways: handed to a person at the point a connector stops, or carried through to done.

What runs today

Four flows, live in their own infrastructure

Each one is defined by its trigger, the work Stepwork completes, and the result handed back to the team. Two sit with IT, two with fund operations.

Lifecycle provisioning~11 runs/wkIT
Takes onboarding, role-change, and offboarding work from the ticket queue, validates the request, completes both API and non-SCIM actions across the required apps, then posts status back.
Systems
- Serval
- Okta
- Microsoft 365
- Figma
Office access and approvals~8 runs/wkIT
Reads the person, location, and access level out of the request, updates the building access system, handles the approval, and returns confirmation, without an operator opening the app.
Systems
- Slack
- Kisi
Incoming mail and document scanning~2 runs/wkFund operations
Twice a week, opens the document system, detects new mail, extracts what matters, routes each item, and stops early when there's nothing to process. Runs from a webhook, so there's no batch window to miss.
Systems
- Microsoft 365
Agreement processing~1,050 runs/wkFund operations
Starts from a document request, prepares and sends the envelope, tracks its status, and returns the completed agreement to the workflow that asked for it.
Systems
- DocuSign

How the rollout went

Stepwork installed
Sep 2025
Service accounts set up
Oct 2025
First automations tested
Nov 2025
Slack intake tested in their own environment
Jan 2026

Systems these flows run across

Serval
Slack
Okta
Microsoft 365
Google Workspace
DocuSign
Figma
Kisi
Front
Metabase
1Password

The results

Time back, without babysitting the queue

Work that once waited on a person now completes on its own, at any hour, in the systems the team already used. Every run leaves an auditable record.

30–40 hrsreturned every week on back-office work
1,050Procedures in seven daysOn the agreement flow alone
96%Success rate~2,100 successful runs to date
15×Faster on document scanning90 minutes a run, down to 6

What's next

The next set of flows is already scoped

Scoped with the team and not live yet. Listed here as direction, not results.

Certificate issuancePreparing and issuing certificates without repeated portal work. Around 60 a week today, handled by three people.
Non-SCIM provisioningThe applications identity connectors can't reach, which is where 44% of provisioning requests land.
The long tail of IT requestsThe repeatable remainder of the service queue, absorbed flow by flow.

The platform behind a large share of US startup investing

Both teams had the same problem in different clothing. Requests arrived cleanly. Finishing them still meant a person opening an app and clicking through it.

The queue described the work. A person still had to finish it.

Half the queue was access work, and connectors could not reach all of it.

3,837Tickets analyzedOct 2024 – Jun 2026
1,968Access and provisioning51% of all tickets
503Hand-run by ITNo connector

Where identity tooling stopped

Identity tooling only reaches the apps that expose a connector. Everything else was hand-run by IT in the ticket queue: the same clicks, over and over.

Figures are approximate. Categories are derived from ticket subject, description, and tags; about 25% remain uncategorized.

44%No connector, hand-run by IT503 requests
56%SCIM-capable, auto-provisioned633 requests

Of those, 1,136 tickets named a specific app. This is how they split.

Four flows, live in their own infrastructure

Each one is defined by its trigger, the work Stepwork completes, and the result handed back to the team. Two sit with IT, two with fund operations.

Lifecycle provisioning~11 runs/wkIT

Takes onboarding, role-change, and offboarding work from the ticket queue, validates the request, completes both API and non-SCIM actions across the required apps, then posts status back.

Systems

Serval
Okta
Microsoft 365
Figma

Office access and approvals~8 runs/wkIT

Reads the person, location, and access level out of the request, updates the building access system, handles the approval, and returns confirmation, without an operator opening the app.

Systems

Slack
Kisi

Incoming mail and document scanning~2 runs/wkFund operations

Twice a week, opens the document system, detects new mail, extracts what matters, routes each item, and stops early when there's nothing to process. Runs from a webhook, so there's no batch window to miss.

Systems

Microsoft 365

Agreement processing~1,050 runs/wkFund operations

Starts from a document request, prepares and sends the envelope, tracks its status, and returns the completed agreement to the workflow that asked for it.

Systems

DocuSign

How the rollout went

Stepwork installed
Sep 2025
Service accounts set up
Oct 2025
First automations tested
Nov 2025
Slack intake tested in their own environment
Jan 2026

Systems these flows run across

Serval
Slack
Okta
Microsoft 365
Google Workspace
DocuSign
Figma
Kisi
Front
Metabase
1Password

Time back, without babysitting the queue

Work that once waited on a person now completes on its own, at any hour, in the systems the team already used. Every run leaves an auditable record.

30–40 hrsreturned every week on back-office work

1,050Procedures in seven daysOn the agreement flow alone

96%Success rate~2,100 successful runs to date

15×Faster on document scanning90 minutes a run, down to 6

The next set of flows is already scoped

Scoped with the team and not live yet. Listed here as direction, not results.

Certificate issuancePreparing and issuing certificates without repeated portal work. Around 60 a week today, handled by three people.
Non-SCIM provisioningThe applications identity connectors can't reach, which is where 44% of provisioning requests land.
The long tail of IT requestsThe repeatable remainder of the service queue, absorbed flow by flow.