Build in Public
Explain a SaaS Rollback While Building in Public
Explain what a SaaS rollback restored, what work remains pending, and what outputs already reached users. Use a fictional reminder-service example.
By Uriel Bitton · · 9 min read

The short answer
Explain a SaaS rollback by naming what you reverted, the behavior you have verified, and the effects still under review. Check live software, pending work, and earlier outputs separately when relevant. A rollback can be the full fix, but describe the customer state the evidence establishes.
Explain a SaaS rollback in public by saying what you reverted, which behavior you have verified, and which effects still need attention. Separate the live software from pending work and results already produced. A completed rollback can be an important recovery step, but your public update should describe the customer state you have checked rather than leave readers to infer it from a deployment action.
This matters when you build in public because an earlier feature announcement remains part of the product’s story. People may have seen the launch, tried the new behavior, or relied on its output. Your rollback update should help them understand what is available now and what remains under review.
Explain what “rolled back” means in this update
Here, a rollback means returning software or configuration to an earlier version. Name the affected part in language a customer can connect to a task: the reminder scheduler, the export service, or the new editor. An unexplained commit identifier does little to tell someone what they can use.
Also separate the attempted action from its verified result. Starting a rollback, completing a deployment, and confirming that the intended behavior is working are different observations. Report the stage you have actually reached. If verification is still underway, include that limit beside the action.
Amazon’s rollback-safety guide explains why moving backward between versions requires care: an older version may be unable to interpret data or protocols introduced by a newer one. The engineering lesson is that rollback compatibility needs verification. The communication implication is narrower: the identity of a previous release alone does not prove the current service is healthy.
For notification channels, update cadence, and an overall incident timeline, use the guide to communicating a SaaS outage while building in public. This article focuses on the state after a release has been reverted, particularly when work created during that release can outlast the code that created it.
Give live behavior, pending work, and past outputs separate attention
Before writing the public explanation, ask three questions. What is running now? What work is still waiting to run? What results have already reached users or another system? These questions help identify the evidence needed for each sentence in the update.
The answers depend on your product. A calculation performed entirely in the browser may leave different effects from a scheduler that stores future jobs. A generated file, a saved record, and a delivered message can each require a different assessment. Do not assume that every release creates all three kinds of work.
Keep a private note connecting each relevant state to its latest observation. If a queue has not been reviewed, say that internally before using broad recovery wording externally. If a particular task has passed a check, describe the task and scope the check actually covered.
This is a way to organize the explanation, not a procedure for choosing a rollback, repairing records, or resending work. Those operational decisions need the product’s actual design and recovery process. The public statement should follow the checked state produced by that work.
A fictional reminder service shows why the distinction matters
Tutor Vale is an invented SaaS for tutors to organize lesson reminders. Its architecture, failure, observations, and example updates are fictional. No deployment, rollback, incident response, or customer-data review was performed for this article.
In the imagined design, a scheduler stores reminder jobs and a separate worker dispatches their emails. A new scheduler release creates some jobs with an earlier sending time than intended. The previous scheduler remains compatible with the stored job format, but redeploying it does not rewrite jobs already stored. The example has no mechanism for recalling delivered emails.
The founder returns the scheduler to its previous release and separately pauses reminder dispatch. A check of a new example job produces the expected sending time. The jobs created during the affected release still need review, and the scope of emails already dispatched has not yet been established.
Those assumptions produce the following account at this point in the fictional recovery:
Part of the product state | What the founder has established | What remains unresolved |
|---|---|---|
Live scheduler | The previous release is running; a new example job has the expected time. | The single example does not establish every scheduling case. |
Pending reminder jobs | Dispatch is paused; stored jobs remain available for review. | Which jobs need correction and when dispatch can resume. |
Earlier email outputs | Some reminders were sent earlier than intended. | The affected scope and any follow-up needed for recipients. |
The rollback is real within the exercise, but it answers only part of the recovery question. The worker pause is a separate action. Reviewing stored jobs is separate work. Earlier messages remain earlier messages even after the scheduler version changes.
This does not mean every SaaS rollback leaves a queue or output problem. It means that a public explanation should follow the architecture and evidence of the particular product. The three rows help the fictional founder avoid compressing several unfinished questions into “everything is back.”
Write the public update from the verified state
Lead with what changed for the affected task. Then identify the remaining review and where current instructions will appear. Readers should be able to understand the situation without knowing how the scheduler and worker are deployed.
The following is proposed wording for Tutor Vale’s fictional situation. It reports no real incident or customer outcome:
We have returned the reminder scheduler to its previous release. A new example job now has the expected sending time. Reminder dispatch is paused while we review jobs created during the affected release.
The rollback does not recall reminder emails already delivered. We are checking the affected scope and what follow-up is needed. We have not marked the incident resolved. Current instructions and the next checked update will remain in the incident notice.
The wording names the action and its evidence without presenting one successful example as a complete recovery test. It also states the delivery pause separately, so readers do not have to assume that replacing scheduler code stopped a different component.
A real notice needs the actual observation time, relevant timezone, affected task, and next communication commitment. Use the incident notice as the dependable current record, with a shorter public founder update pointing there. The build-in-public post can explain the decision while keeping operational information coherent.
Check the scope before telling customers to repeat a task
A rollback update often creates an immediate practical question: should someone try again, wait, or check an earlier result? Give a recovery instruction only when you have established that it fits the affected action and audience.
In the Tutor Vale example, a blanket request to resend every reminder would introduce a new action before the earlier dispatch history has been understood. The fictional founder instead keeps that decision open while reviewing the relevant state. The public notice can explain the ongoing pause and provide an appropriate support route for questions.
Apply the same reasoning to other outputs. A failed export that produced no file differs from an export whose completion is uncertain. A saved change that needs repair differs from a request that never reached the service. Your communication should reflect the case you have verified rather than borrow a generic “please retry” sentence.
Keep any individual review in the appropriate private channel. The public update can explain the affected task and the next step without reproducing account records or a recipient’s message. Ask for only the context needed by the actual support process.
Keep the earlier launch claim connected to the rollback
An accurate rollback notice can coexist with an old launch post that still suggests the new feature is available. Someone arriving through that post needs a route to the current state. Add an appropriate dated update where the original claim remains relevant, and keep it linked to the maintained notice.
Distinguish a later change from an earlier factual error. If the feature was available when announced and was subsequently reverted, record that later event. If the announcement overstated availability from the start, correct that claim as well. The guide to correcting a build-in-public update covers how to make those changes visible where the claim traveled.
You do not need to rewrite the history into a seamless success story. Preserving the earlier decision, the reason for the reversal, and the checked current state can make the product’s development understandable. Keep the technical detail proportional to what helps the reader use the product or assess your reasoning.
Match resolution wording to the remaining impact
A rollback can be the complete fix when the evidence supports that conclusion. Google Cloud’s incident lifecycle includes rollback as a possible fix when the change is expected to end the impact. It also distinguishes mitigation, which reduces impact or scope, from a fix. Those are Google’s operational definitions, not a required vocabulary for every small SaaS.
For your public account, choose status wording that matches what you have established. If new requests work but earlier jobs still need repair, make both facts visible. If the relevant recovery checks are complete and no remaining customer impact is identified, explain the basis for closing the incident.
In the fictional example, the founder needs an answer about affected stored jobs, dispatch readiness, and earlier messages before making a complete recovery claim. The article does not supply those answers. A later update should report the actual findings and the customer action they support, rather than assume a favorable result because the previous release is running.
Google Cloud’s guidance also describes ongoing updates about scope, mitigation progress, and changes in status. That is a useful communication model: each update should change the reader’s understanding of the current situation using checked information.
Make the later build-in-public lesson specific
After the relevant facts are established, explain what the rollback revealed about the product and its release process. A useful lesson could concern a missing compatibility check, a queue that needed separate handling, or an announcement that described recovery too broadly. Claim only the lesson supported by the actual event.
Separate completed improvements from proposed ones. “We added this case to the release check and verified it” needs evidence of that work. “We plan to add this case” describes a commitment. Both can belong in the public record when their status is clear.
For your next release, prepare a short private explanation of what reverting the code would change and what it would leave behind. If you later need a rollback update, that record gives you specific questions to verify. Publish the answers you have, keep unresolved effects visible, and return with the findings that change them.
Questions founders ask next
Should I publish the commit hash or deployment log?
Use the detail that helps the reader understand the affected task and current behavior. A technical audience may benefit from a relevant public release reference. Customers usually need the practical scope and next step. Keep private logs, identifiers, and operational access details out of an ordinary public update.
What if the rollback itself does not restore the task?
Describe the rollback as an action taken and state that the task remains affected. Keep the next investigation and communication commitment current. Do not describe an attempted recovery step as a successful outcome when the relevant behavior has not been restored.
Can I publish the root cause before the review is finished?
Separate a working explanation from an established finding. You can describe the confirmed symptom, actions, and remaining questions while the cause is being investigated. Add the reviewed explanation later, and correct any earlier material claim that turns out to be wrong.
Sources
Frequently asked questions
Should I publish the commit hash or deployment log?
Use the detail that helps the reader understand the affected task and current behavior. A technical audience may benefit from a relevant public release reference. Customers usually need the practical scope and next step. Keep private logs, identifiers, and operational access details out of an ordinary public update.
What if the rollback itself does not restore the task?
Describe the rollback as an action taken and state that the task remains affected. Keep the next investigation and communication commitment current. Do not describe an attempted recovery step as a successful outcome when the relevant behavior has not been restored.
Can I publish the root cause before the review is finished?
Separate a working explanation from an established finding. You can describe the confirmed symptom, actions, and remaining questions while the cause is being investigated. Add the reviewed explanation later, and correct any earlier material claim that turns out to be wrong.
A note from Uriel Bitton
I share practical lessons from building Buildside and learning alongside SaaS founders.
Keep building with us
Get practical founder lessons and product updates in your inbox.
Keep reading
How to Correct a Build-in-Public Update
Fix the original claim, reach readers who saw it, and repair any conclusions that relied on it. Includes a worked SaaS activation correction.
How to Communicate a SaaS Outage While Building in Public
Write useful public SaaS outage updates, keep customer information consistent across channels, and follow recovery with a factual founder update and visible corrective work.