Results Don’t Match Expectations
The numbers seem wrong
Use the same checks your SourceMedium team uses before treating the result as a pipeline issue.Check the date range in the SQL
Review the generated SQL query. The AI Analyst may have interpreted your time range differently than you intended. “Last week” could mean the previous 7 days or the previous calendar week.
Verify the metric definition
Confirm the AI used the metric you expected. “Revenue” might be gross, net, or total depending on context. Check the SQL to see exactly what was summed, and compare against Metric Definitions and Revenue Fields.
Check Data Health
Ask “How is my data health?” — if the relevant tables are stale, recent data may be missing from results.
Check filters and segmentation
Confirm the query used the same filters as the dashboard or external report: channel, store, order validity, customer type, subscription status, product, SKU, and date basis.
Compare complete days only
Avoid current-day data when reconciling. Use a completed date range that should be fresh based on Data Freshness.
How to investigate a discrepancy
When SourceMedium does not match Shopify, an ad platform, Amazon, Recharge, or another external report, start with Why would external reports not match the SourceMedium dashboard?, then identify the exact comparison being made.| Check | Why it matters |
|---|---|
| Date field | Order date, refund date, click date, conversion date, and platform reporting date can produce different totals. |
| Metric | Gross revenue, net revenue, total revenue, discounts, refunds, and platform-reported revenue are different measures. |
| Filters | Channel, store, order validity, customer type, subscription flags, product filters, and dashboard defaults can change the denominator. |
| Freshness | SourceMedium’s SLA is complete data through the previous day; some platforms lag longer. |
| Attribution logic | SourceMedium attribution and platform-reported attribution answer different questions. |
| Source system behavior | Some source systems have their own reporting windows, API lags, and attribution logic. |
Documentation can explain definitions, expected freshness, and the checks above. If the discrepancy remains after those checks, it usually requires account-specific investigation by SourceMedium support.
High (direct) / (none) in results
This indicates an attribution issue, not a query issue. The AI Analyst is correctly reflecting what’s in your data.Attribution Health
Learn how to diagnose and improve attribution coverage.
The AI Misunderstood My Question
It answered a different question
Rephrase with more specificity:| Instead of | Try |
|---|---|
| ”How are we doing?" | "What was our revenue last 30 days vs. the previous 30 days?" |
| "Show me sales" | "Show me daily order count for the past 2 weeks" |
| "Best products" | "Top 10 products by revenue this month” |
It used the wrong table
If the AI queried the wrong data source, clarify the domain:| Instead of | Try |
|---|---|
| ”Show me performance" | "Show me ad performance by platform" |
| "Customer data" | "Show me new customer acquisition by week” |
No Results Returned
”No data found” or empty results
- Check the date range — You may be querying a period with no activity
- Check filters — The AI may have applied filters that excluded all rows
- Check Data Health — The table may be empty or not yet populated for your account
Query timed out
Complex queries against large date ranges may time out. Try:- Narrowing the date range
- Asking for aggregated data instead of row-level details
- Breaking the question into smaller parts
Chart Not Generated
The AI Analyst generates charts when visualization adds value. Charts may not appear when:- The result is a single number (e.g., “total revenue was $142,350”)
- You asked for a raw data pull or export
- The data structure doesn’t lend itself to visualization
- There was an error generating the chart
Raw data pulls are optimized for CSV output, not charts. Ask a follow-up like “analyze this by channel” if you want interpretation and a visualization.
Access or Installation Denied
”This workspace is not authorized”
Your Slack workspace or installing user’s email may not match the tenant configured for your organization. Contact your SourceMedium team to verify the Slack app configuration.”Installation denied” or “access denied”
The AI Analyst only works for approved Slack workspaces and users. If the user should have access, ask your SourceMedium team to update your organization’s AI Analyst access.Slow Responses
Response time depends on:- Query complexity — Aggregations across large date ranges take longer
- Table size — Some tables have more data than others
- Current load — Occasional delays during peak usage
My question says it is queued
The AI Analyst processes one analysis at a time within a Slack thread so follow-up context stays consistent. A thread can have one active analysis and one queued question. If another question is already queued in that thread, wait for the current analysis to finish before asking again.I hit the concurrent analysis limit
Each Slack user can have up to 2 running analyses at the same time. If you see a message saying you already have 2 analyses running, wait for one to finish before starting another.The AI Said It Can’t Help
The AI Analyst is designed for e-commerce analytics questions. It will decline requests that:- Ask about data outside your BigQuery warehouse
- Request actions (modifying data, sending emails, etc.)
- Are unrelated to analytics (general knowledge questions)
Common Questions
Can I ask follow-up questions?
Can I ask follow-up questions?
Yes. Follow-up questions are supported in Slack threads.
- In DM, you can ask follow-ups naturally.
- In channels, mention the bot with
@SourceMediumfor follow-up analysis (for example,@SourceMedium what were sales yesterday?).
Why can’t I switch stores in the same thread?
Why can’t I switch stores in the same thread?
In multi-store tenants, once a store is confirmed, that thread stays on that store. This prevents mixed-store analysis in the same thread. Start a new thread to switch stores.You can view your available stores in Data Availability Report. The list is shown by
sm_store_id (see the Glossary).Why did my typed reply confirm the store?
Why did my typed reply confirm the store?
If a thread is waiting for store confirmation, any reply in the thread (for example, “yes” or typing a follow-up question) is treated as confirmation so analysis can continue. After confirmation, the question you already asked runs automatically.
Why didn’t the bot answer my channel thread reply without @mention?
Why didn’t the bot answer my channel thread reply without @mention?
In channels and private channel threads, follow-up analysis requires mentioning the bot with
@SourceMedium. If you reply without that mention, the bot won’t run a new analysis turn.Why did it ask me to run analysis first before summarizing?
Why did it ask me to run analysis first before summarizing?
Summary and recommendation prompts depend on prior analysis in the same thread. If that thread does not have enough prior analysis context, the AI asks you to run analysis first.
Why did it miss an older detail from this thread?
Why did it miss an older detail from this thread?
Continuity in a thread is condensed so follow-ups stay fast and reliable. In long threads, some older details may be compressed. If an older detail matters, restate it directly in your next prompt.
How do I reset context?
How do I reset context?
Start a new thread to reset context.
How often does data refresh?
How often does data refresh?
Most tables refresh daily. Our SLA guarantees fresh data through yesterday based on your reporting timezone. Today’s data is not guaranteed. See Data Health for details.
Can I export the results?
Can I export the results?
Yes. Data responses include downloadable files such as
query.sql, results.csv, and the chart image if one was generated. Raw data pulls attach results.csv automatically.Can multiple people use it at once?
Can multiple people use it at once?
Yes, approved users can ask questions simultaneously. Each user is capped at 2 running analyses, and each thread allows one active analysis plus one queued follow-up question.
Why do I see different results than my teammate?
Why do I see different results than my teammate?
You shouldn’t — everyone queries the same underlying data. If results differ, check:
- Are you asking about the same time range?
- Are you using the same metric definition?
- Did the data refresh between queries?
Multi-store thread scoping
Learn selector prompts, confirmation, lock rules, and channel mention requirements.
Thread continuity and follow-ups
See what the AI remembers in the same thread and how to reset context.
Getting More Help
If you’re stuck:- Check the SQL — The generated query often reveals how the AI interpreted your question
- Try a simpler question — Start basic, then add complexity
- Ask about schema — “Which table has order data?” can help you understand the data structure
- Contact support — Reach out to your SourceMedium team for persistent issues