A typical Deep Research run is 5–10 minutes. If yours is taking 20+ minutes, this page walks through the most common reasons and what to do.
1. Document indexing is still in flight
If you uploaded a document right before starting Deep Research, agents that depend on the documents (AI Summary, Document Q&A, sometimes Units) will wait for indexing to complete before starting. Indexing a 200-page PDF can take 1–2 minutes on its own.
Check: the Documents section card has a "Processing" pill on any document.
What to do: wait for indexing to complete, then look at the section progress. Other sections (Taxes, Comps, POIs) should be running in parallel and won't be blocked.
2. Large rent roll
A rent roll with 500+ units takes longer to extract and validate than a 50-unit one. Bigger isn't bad — it's just real work.
Check: the property's expected unit count from Basic Info or the OM.
What to do: wait. If the rent roll is genuinely huge (1000+ units), expect 3–5 minutes for the Units section to settle.
3. Tertiary-market data thinness
Properties in tertiary markets sometimes trigger more retries on listing platforms (which return slower or empty responses) and on the v2 Units pipeline. The agentic crawler retries when it hits a rate-limit or a thin source.
Check: how many of the property's section cards are stuck on "Running" past 3 minutes.
What to do: for properties you'll research repeatedly in the same submarket, the Redis cache fills in over time and subsequent runs are faster. If a particular section keeps timing out, report the property URL so we can tune the agent.
4. Org-wide queue
If many properties are running concurrently in your org, later ones queue up.
Check: any section card showing "Queued (position N)".
What to do: wait — runs proceed in order. If your team is consistently saturating capacity, talk to sales about Enterprise concurrency or a Private Server.
5. Listing platform rate limit
The comps and units agents occasionally hit listing-platform rate limits. We back off and retry. A backoff cycle adds 30–90 seconds.
Check: the section log shows "Retrying (rate limit)".
What to do: wait for the retry, or cancel and re-run later when the platform is less hot.
6. v2 anomaly detection re-running
The v2 Units pipeline includes anomaly detection — when the agentic crawler returns a suspicious unit count, it re-attempts the discovery rather than committing the bad number. That's slower but produces better data.
Check: the Units status pill shows "Verifying".
What to do: wait. If the verification keeps failing, switch to v1 (Stable) for a more deterministic run, then report the v2 issue.
7. Backend incident
If our agent platform is degraded, all properties in your org will run slowly. We surface this on the Status page.
Check: /status — incidents are posted there.
What to do: wait for the resolution; we'll auto-retry runs once the platform is healthy again.
When to give up and re-run
If a run is past 15 minutes with no progress on any section, it's worth canceling and re-running:
- Cancel all sections from the property header.
- Wait 30 seconds.
- Re-start Deep Research.
If the second run also stalls, contact support with the property URL.