"Customers used to complain about indexing every day. Now they don't mention it at all." Frank Hidalgo, PhD, CTO, StackAI (an Asana company)
Proof at a glance#
- Below 0.1% indexing failure rate, down from about 20%
- Daily complaints about indexing stopped after the new pipeline went live
- Parsing costs fell to roughly one-tenth of their earlier level through a broader redesign that combined Temporal with decoupled parsing, caching, and better retry logic
The customer#
StackAI, now part of Asana, gives regulated enterprises a no-code way to build AI workflows with governance and guardrails. Temporal coordinates the pipeline that turns files from SharePoint, Google Drive, and Dropbox into knowledge those workflows can search and use.
Asana is where work lives, and StackAI connects that work to the rest of the business. StackAI lets AI in Asana reach the other systems a company runs on, like Ironclad, DocuSign, Salesforce and ServiceNow, so it can check information and act on it across tools. For technical builders, StackAI also powers advanced agent workflows that pull data from knowledge sources (orchestrated by Temporal), run browser agents and protect sensitive data.
The challenge#
StackAI's indexing failure rate was about 20%, and customers complained about indexing every day.
The solution#
Temporal keeps indexing moving:
-
Resume at the failed step. A Temporal Workflow records each file's progress. If one Activity fails, Temporal retries that step without rerunning earlier stages that have already completed.
-
See exactly what went wrong. Engineers can use the Temporal UI to find the file, stage, and error behind a failure. Payloads are always encrypted, so customer data stays protected.
-
Scale different work independently. StackAI can assign different Activities to separate Worker pools and give each pool the memory and capacity it needs.
| StackAI's knowledge base indexing starts workflows on Temporal Cloud, which coordinates the Temporal Worker as it downloads, extracts, chunks, and indexes files, and retries failed work without rerunning completed steps. Knowledge base search queries the same vector database. |
|---|
The success of the indexing pipeline led StackAI to bring Temporal into the automation engine behind the node-based workflows its customers build.
The results#
-
Below 0.1% indexing failure rate, down from about 20%
-
Daily complaints about indexing stopped after the new pipeline went live
-
Parsing costs fell to roughly one-tenth of their earlier level through a broader redesign that combined Temporal with decoupled parsing, caching, and better retry logic
"Customers used to complain about indexing every day. Now they don't mention it at all."
Frank Hidalgo, PhD, CTO, StackAI (an Asana company)
The takeaways#
-
Retry the step that failed, not the whole pipeline. Temporal records each file's progress, so completed stages are never rerun.
-
Failures are easy to trace. The Temporal UI shows the file, stage, and error behind a failure, and payloads stay encrypted.
-
Temporal was part of a bigger redesign. The drop in parsing costs came from combining Temporal with decoupled parsing, caching, and better retry logic.
-
One use case opened the door. The success of indexing led StackAI to bring Temporal into the automation engine behind its customers' node-based workflows.
Check out the in-depth version of this story here.
