Ops and Customer Support AI automation with human supervision

Your Ops and CX people are more valuable than ever supervising and reviewing AI's work

Hero Video

How it works

Prompt-based configuration

You tell Drafting AI what to do in different scenarios. There's no risk from it trying to figure out what to do on its own.

Define workflows

Define a separate workflow per scenario. The AI only acts within the workflows you define.
Define workflows

Workflows have a trigger and an action

  1. The trigger defines which emails trigger this workflow.
  2. The action is which URL to open and which form fields to fill when this workflow is triggered.
Workflows have a trigger and an action

Draft replies

Prompt the AI on how to reply to emails that trigger this workflow.
Prompting is more flexible than canned responses and AI writes fluent English.
The AI only writes the draft. You're in control of reviewing and sending it.
Draft replies

Develop your prompts

Our Prompt Studio lets you try different prompts on your emails. Validate your prompts, and develop better and better ones.
Develop your prompts

Why human-in-the-loop?

A browser based approach to automation

turns out to have lots of other benefits

Safe to deploy

No AI is 100% reliable yet, which is why industry is right to deploy it cautiously.
We design for AI unreliability by never taking action without confirmation from a human user.

Incrementally adoptable

Set up scenarios one by one, or parts at a time.
Get started in as little as 30 minutes. Start seeing time savings right away.

Works with the all the tools you already use

If it’s in a browser, Drafting AI can use it. Drops right in to your existing workflows.
It works with 3rd party SaaS like Shopify, Salesforce, and Stripe. It works with no-code tools like Retool and Airtable. And it works with custom internal web apps. On the email side it works with Gmail and Front, with others coming soon.
Airtable
Asana
Coda
Gitlab
Gmail
GoogleForms
Airtable
Asana
Coda
Gitlab
Gmail
GoogleForms
Airtable
Asana
Coda
Gitlab
Gmail
GoogleForms
Airtable
Asana
Coda
Gitlab
Gmail
GoogleForms
Intercom
Jira
Kustomer
Outlook
Retool
Salesforce
Intercom
Jira
Kustomer
Outlook
Retool
Salesforce
Intercom
Jira
Kustomer
Outlook
Retool
Salesforce
Intercom
Jira
Kustomer
Outlook
Retool
Salesforce
Shopify
Zendesk
Box
ClickUp
Docusign
Dropbox
Shopify
Zendesk
Box
ClickUp
Docusign
Dropbox
Shopify
Zendesk
Box
ClickUp
Docusign
Dropbox
Shopify
Zendesk
Box
ClickUp
Docusign
Dropbox
Github
Monday
Rippling
Stripe
Notion
Slack
Github
Monday
Rippling
Stripe
Notion
Slack
Github
Monday
Rippling
Stripe
Notion
Slack
Github
Monday
Rippling
Stripe
Notion
Slack

Improve with real-world usage

If people are intervening on a workflow, we can use their corrections to adjust the AI's instructions for next time. Track where what corrections or next steps are being taken, and build them into the automation.

Faster, more consistent execution

Save 90% of the time previously spent on routine emails.

The best reviewer experience

If the AI suggests making a change to Salesforce, review and approve that change in the Salesforce UI

Partial automation welcome

If you don’t have a fully defined process, Drafting AI can do steps 1-4 and hand back to a person to finish steps 5-8.

Easy integration

The AI uses the same website you do. No API or special access required. No login needed: the AI shares your browser, so it automatically has access to wherever you’re logged in. Get started in as little as 30 minutes.

Incremental Automation

We help you drive your organization from 0 to AI the right way, with value delivered and clear next steps at every stop along the way.

Partial automation changes the risk/ROI for implementing AI.

Without it, you have to map out an entire workflow and get 100% of the cases right before you can deploy. With partial automation, we suggest starting to automate a workflow by configuring the AI to do just the first few, well defined steps, then handing off to a person to finish. Then, gradually add more scope to the AI as you get real-world data about how your people are completing the task.

Because the AI is being trained to use the same tools as a person, there's no replatforming or rework as you grow it towards full automation.

Progress Report

Progress Report

Each workflow is configured separately. They range from fully human-handled to fully AI-handled.
WorkflowAutomation LevelCnt/WeekAvg TimeAccuracy
Refund request
50% automated
60%
150/wk5 min95%
Order status request
Task recognized
23%
300/wk3 min98%
Lost package
Worker finishes
82%
100/wk8 min90%
Double charge issues
Worker reviews
96%
50/wk10 min99%
Add items to order request
Full automation
100%
75/wk4 min100%
Other
Fully manual
0%
125/wk15 min0%

Typical journey for a single workflow

Each step of incremental automation delivers immediate incremental value.

Want to see how partial automation changes your AI implementation?

Schedule a call