How we use automation in Customer Success
Contents
How we use automation
Customer Success at PostHog means managing ~30 accounts per CSM while maintaining deep, meaningful relationships with each customer. Automation and AI tools can help surface important signals and streamline repetitive tasks, allowing CSMs to focus on strategic guidance and relationship building.
Automation should never be used as a replacement for human connection and interaction, but mainly as a tool to help a CSM be better prepared, informed, and effective.
Current automation stack
PostHog CS leverages several integrated tools to monitor account health and identify opportunities:
Core monitoring systems:
- Customer Analytics: Tracks usage patterns, billing changes, health scores, and engagement metrics.
- Health scores update regularly as a composite of multiple metrics with different weights
- Data is synced in from Salesforce, PostHog, BuildBetter, and Stripe
- PostHog pipelines: Alerts for usage milestones, new product adoption, and behavioral changes. These are sent to Salesforce or Slack via PostHog CDP for alerts and data updates.
- BuildBetter: Analyzes customer calls for feature requests, pain points, and sentiment.
- Notes from calls are automatically synced to Salesforce and Customer Analytics
- feature requests and painpoints are automatically added to Customer Analytics and sent to #feature-request-feed channel
- Gong: Records and transcribes customer calls.
- Gong Streams push call recordings, transcripts, and summaries for a specific account into that account's internal Slack channel
- Zapier: Used for numerous automations such as:
- Renewal reminders
- Stale Slack channel notifications
- Billing updates
How to surface calls for shared accounts
On larger accounts, a TAM and a CSM share ownership. Both need visibility of every customer call, not just the ones they attended.
For each shared customer, we keep a dedicated internal Slack channel named customer-name-internal. This sits alongside the customer-facing Slack Connect channel, but stays internal only.
To surface calls there, set up a Gong Stream:
- In Gong, open Your Library and find Streams (ask Simon for access)
- Configure a stream for the account
- Point it at that account's
customer-name-internalchannel Once live, the stream posts each call to the channel after it ends. Everyone working the account is notified once the call is done, with the transcript and summary to hand.
Key automated workflows
Account monitoring triggers include:
- MRR changes exceeding certain thresholds generate investigation tasks
- inactivity periods to flag engagement reviews
- The first real use of a paid product that the account did not use before creates a Customer Analytics task for the account owner (see managed customer scouts)
- Churn risk signals, a new admin or owner, and a team rollout create a Customer Analytics task for the account owner (see managed customer scouts)
- Overdue invoices and failed card payments send Slack alerts to the assigned AE, CSM, or TAM (see failed/late payments)
- Paid credit that will run out at least a month before it expires sends Slack alerts to the assigned AE, CSM, or TAM, and creates a Customer Analytics task for the account owner (see when they don't have enough credit)
- Health score changes can trigger Customer Analytics tasks or PostHog automation
- Annual renewal dates trigger preparation workflows in advance and a ping in Slack
Managed customer scouts
Five scouts watch every account that has a CSM, TAM, or AE in Customer Analytics. When a scout finds something, it creates a Customer Analytics task for the account owner. Most tasks are due in 3 working days. A HIGH churn risk task is due in 1 working day, and a new stakeholder task is due in 2 working days. The scout also tags the account owners in #project-managed-customers-consolidated-scouts.
| Scout | What it finds | Task name starts with |
|---|---|---|
| Billing intent | A billing page session where the customer is quota limited, struggled, changed a setting, looked into a cost change, or checked their limits several times. Replay Vision reads the session. | Billing: |
| New tool adoption | The first real use of a paid product that nobody at the account used in the last 400 days. | New tool: |
| Survey responses | An in-app survey answer that needs action, for example a low score, a bug, a cost concern, or a request for help. Replay Vision adds what the person did before and after the answer. Each owner also gets a weekly digest on Mondays with all other answers from their accounts. | Survey: |
| Churn risk | A decline that nothing explains, for example a sudden usage drop on a product that is a large part of the bill, active users that fall by half, a cancelled product or a lower billing limit, a champion who left, or revenue that fell. Before it creates a task, it removes drops that a billing limit, a change that an owner recommended, or a move to the API or MCP explains. Each weekday it scores one fifth of the accounts, so it scores each account once a week. | Churn risk (HIGH): or Churn risk (MEDIUM): |
| People changes | A new admin or owner joins the account, or 5 or more people join an account with $20k or more ARR in 14 days. Each owner also gets a weekly digest on Mondays with all the people who joined, left, or became an admin at their accounts. | New stakeholder: or Team rollout: |
The churn risk scout does not add the churn risk tag. If you agree that the account is at risk, add the tag yourself.
Close the task when you act on it. Reply in the Slack thread when a task was useful or was noise, so that we can tune the scouts.
Friction watches for your own book
Each friction watch skill catches a person who gets stuck in an area where a timely intervention from you can unblock them. These friction watches are low noise by design, and nothing is ever sent to a customer until you turn it on. Matches post to #project-managed-customers-consolidated-scouts and tag you.
To set one up, ask your agent with the prompts below. The setup skill creates your own copy of the scout or workflow, and your own draft email workflow that sends from your address.
| Friction watch | What it catches | What you get | Limit | Ask your agent |
|---|---|---|---|---|
| Flag and experiment friction | A person starts a feature flag or experiment, doesn't save it within 48 hours, and reads the docs for it. | A post that tags you when there are matches. After you turn on the email workflow, each match gets an email from you that offers help. | One email per person every 90 days | "Set up the flag friction watch for me" |
| Warehouse friction | A data warehouse source fails to connect and is still not connected 45 minutes later. | A post that tags you, with the error the person saw and a proposed email from you. Send it yourself, or turn on the email workflow. | One email per person every 90 days | "Set up the data warehouse friction watch" |
| Query friction | A person keeps hitting slow or failed queries, in the app or through the MCP. | After you turn on your workflow, an email from you to each person it finds, and a post that tags you each time an email goes out. | One email per person every 30 days. Each account hears from one owner only, in this order: CSM, TAM, AE. | "Set up query friction outreach for me" |
Before you turn on an email workflow:
- Rewrite the default email so it sounds like you.
Query frictionhas two emails, one for app queries and one for MCP queries. - Send yourself a test email to confirm everything looks good.
Human-first automation philosophy
Every automated workflow includes deliberate human decision points. For example, when an account begins using session replay, Customer Analytics can create a task suggesting outreach about their use case - but the CSM determines whether and how to engage based on the account relationship and context.
This approach ensures automation enhances rather than replaces the human elements of customer success.
Working effectively with automations
Best practices:
- Review automated tasks such as Customer Analytics tasks, Slack alerts, and customer health scores at least twice weekly.
- Treat automated insights as starting points for investigation, not final answers!
- Set up individual alerts in Customer Analytics or PostHog for signals that the managed customer scouts don't cover. Start with the friction watches. Do not copy the managed scouts for your own book, because they already watch every owned account.
- As an example, setting up your accounts as a Cohort in PostHog and then setting up CDP alerts/notifications to Slack based on product usage and activity.
What remains purely human:
- Initial customer responses and relationship building
- Renewal negotiations and strategic planning
- Technical implementation guidance
- Complex problem-solving and consultative conversations
Requesting new automations
CSMs are encouraged (as are all PostHog employees) to experiment and surface new ideas frequently in Slack or team stand-up. Examples of areas where automations could be useful include, but are not limited to:
- Time savings versus implementation complexity
- Impact on customer experience
- Requirement for human judgment
- Scalability across the team