Asking questions

Ask in plain English. The analyst picks the right tables, writes SQL against your warehouse, runs it, and replies in the thread. What comes back depends on the kind of question you asked.

Question types

Question type	Example	What you get	Tip
Specific metric	”What was revenue last month, by week?”	A written answer, a table, the SQL, a CSV, and a chart when it helps	Name the metric, the time range, and the breakdown
Raw data pull	”Pull order ID, order date, UTM source, and revenue for last week”	The rows as a table preview and a CSV, plus a short confirmation	Use “pull” or “export” and list the fields you want
Open-ended question	”How can we improve marketing performance?”	Deep Analysis: two analyses combined into one answer	Add context, such as a product launch or a budget change
Definition or schema	”What is LTV?” or “Which table has order data?”	A written explanation with links to the docs, no SQL	Ask “what is” for a definition, “which table” for schema
Data or attribution health	”How is my data health?”	A report on table freshness and UTM coverage	Run it before an analysis you plan to rely on
Follow-up	”Break that down by channel” or “Summarize this for leadership”	A new answer that builds on the thread	Stay in the same thread

Specific metric questions

This is the most common kind of question. One query runs, and you get an answer you can check against the SQL. The best questions name three things: the metric, the time range, and the breakdown. “Revenue last month by week” is a complete question. “How are sales?” is not.

Raw data pulls

Use pull or export language when you want the data itself rather than an interpretation. Name the fields, the time range, and any filters such as product, campaign, store, or channel. Data pulls always run as a single query. They skip Deep Analysis and never produce a chart. The reply confirms what was pulled and attaches the rows as a CSV. Asking for several dimensions does not make a question a data pull. “What is revenue by UTM source and medium?” asks what the data means, so it gets an analysis. “Pull order ID, UTM source, UTM medium, and revenue for last week” asks for rows. Data pulls return at most 100 rows. If you get exactly 100, assume the result was cut off and narrow the request.

Open-ended questions and Deep Analysis

Questions such as “Why are our metrics declining?” or “What should we focus on to grow?” have no single number as an answer. For these, the analyst runs Deep Analysis. Deep Analysis runs two analyses: your question, plus one strategic question the analyst writes to add a second angle. Each shows up in the thread with its own title. The final reply combines both into one answer with recommendations, and it carries the reasoning and feedback actions. Deep Analysis takes longer than a single query. If you want a fast, specific answer, phrase the question specifically. “What was our Meta ROAS last month?” answers faster than “How is Meta performing?” A single question never runs longer than 7 minutes.

Definitions and schema

Questions about what a metric means, how it is calculated, or where data lives are answered without running SQL. Definitions come from SourceMedium documentation. Schema questions are answered from your warehouse’s table metadata, and the reply includes Schema docs links to the table pages. “What is LTV?” gets a definition. “What was our LTV last month?” gets an analysis.

Data and attribution health

“How is my data health?” reports which tables are fresh, which are empty, and which domains are ready to analyze. “How is my attribution health?” audits UTM coverage on your acquisition orders. See Data and attribution health.

Follow-ups

In the same thread you can ask for a new breakdown, a summary, or a recommendation. Summaries and recommendations use what the thread already found and do not run new SQL. If you ask for one before any analysis has run, the analyst asks you to run an analysis first. If your request is ambiguous, it runs a fresh analysis instead. Thread rules, including when to mention the bot, are on Reading results.

Example prompts by domain

Copy one, then swap in the metric, time range, and breakdown you care about.

Orders and revenue

What was our total revenue last month?
Show me daily revenue for the past 2 weeks
How many orders did we have yesterday?
What's our average order value this quarter vs last quarter?
What was our revenue by channel last week?
How much did we discount last month?

Customers

How many new customers did we acquire last month?
What's the split between new and repeat customers this year, by month?
How many customers placed more than one order?
What's our repeat purchase rate?
What's the LTV of customers acquired in Q1?
Show me cohort retention by acquisition month

Marketing and ads

What was our ROAS by channel last month? Include spend and attributed revenue.
Show me ad spend by platform for the past 30 days
Which campaigns had the best ROAS last week?
What's our cost per acquisition by channel?
How much did we spend on Meta ads in January?

Email and SMS

Top 10 email campaigns by attributed revenue in the last 30 days
What's our email open rate trend?
Which SMS campaigns drove the most orders?
Which flows drove the most revenue last month?

Products, subscriptions, and returns

Top 10 products by units sold this month
What's our best-selling product by channel?
How many active subscriptions do we have, and how has that changed this year?
What was our refund rate last quarter?
Which products are returned most often?

Web analytics and attribution

What were our top landing pages by sessions last week?
What does the multi-touch attribution split look like by channel this quarter?
What percentage of orders are unattributed?

Media mix (MMM)

These work on Pro warehouses with a media mix measurement vendor connected (Fospha today). The vendor’s modelled results land in two tables, both in the catalog:

rpt_media_mix_channel_daily is daily by channel group and media platform (store, market, media source and currency are in the grain too), with spend, MMM-attributed revenue, last-click revenue and platform-reported revenue side by side. Use it for channel-level questions.
obt_media_mix_performance keeps every row at the level the vendor reported it (ad, ad group, campaign or channel), with the sales venue the revenue landed on. Use it for campaign questions.

Every MMM question works best in one shape: “According to the media mix model, [question], paid channels only, by channel group, for [window].” Paid means a channel group with spend greater than zero in the window. The window should be settled, such as a closed month, a quarter, the last 90 settled days or the last fully settled week. The analysis prompts below follow this shape, except the ones that ask by media platform, by campaign or about organic credit, which say so. Channel group names come from your table’s channel_group column, so fill in the bracketed names yourself. The platforms named are examples; for a platform that is not in your feed, expect a plain “not in the data”.

Do we have media mix model (MMM) outputs in SourceMedium? Tell me which tables hold them, which measurement vendor they come from, what grain each table is at, what date range they cover, and the last date that is fully settled.
According to the MMM, what share of attributed revenue did each media platform drive last month? Give me spend, MMM-attributed revenue, MMM ROAS and share of the total for each platform, and note which platforms carry organic rows.
According to the media mix model, which channel groups look like they are hitting diminishing returns over the last 26 settled weeks? Paid channels only, by channel group. Tell me how confident you are and why.
According to the media mix model, what would happen to attributed revenue if we cut spend on our TikTok channel groups by 30% next month? Paid channels only, by channel group, based on the last 90 settled days. Show your working.
According to the media mix model, which channel group should we cut first to save $100k a month on paid media, and why? Paid channels only, by channel group, based on the last 90 settled days.
According to the media mix model, which channel groups drove the most new-customer revenue over the last 90 settled days, and what was the new-customer CAC for each? Paid channels only, by channel group, spend divided by MMM new-customer conversions, and leave out any group the vendor does not split by new customer.
For September 2026, split MMM-attributed revenue into paid and organic using the vendor's channel groups. What share of attributed revenue was organic, and which organic groups are largest?
On [platform], how much revenue does the MMM credit to [paid channel group] versus [organic channel group] each month in 2026, and has the split shifted recently?
According to the media mix model, how does MMM-attributed revenue compare with last-click revenue for September 2026? Paid channels only, by channel group. Where does the model give more credit than last click, and where less?
How does September 2026 compare with September 2025 in MMM-attributed revenue and spend by media platform, paid rows only?
According to the media mix model, how is our ROAS looking? Use last month, paid channels only, by channel group.
According to the media mix model, what was our blended new-customer CAC across paid media in September 2026? Paid channels only, by channel group, spend divided by MMM new-customer conversions, and leave out any group the vendor does not split by new customer.
According to the media mix model, build me a weekly summary for the last fully settled week that I can paste in Slack: spend, attributed revenue and ROAS by paid channel group, paid channels only, with the week-over-week change against the week before.
According to the media mix model, which Meta campaigns worked best and worst in September 2026? Use the campaign-level table, rank by MMM ROAS recomputed from summed revenue and spend, and say how much weight the campaign split can bear.
According to the media mix model, where did September 2026's growth in attributed revenue come from versus August 2026? Show it by channel group, with paid and organic groups separated.
According to the media mix model, if I had to present one number for marketing efficiency to the board for Q3 2026, what would it be and why? Paid channels only, by channel group.
According to the media mix model, are we over-spending on brand search? Compare MMM ROAS with platform-reported ROAS for the [brand search channel group] and [brand shopping channel group], paid channels only, for Q3 2026.

Follow-ups can build on an answer in the same chat. After “According to the media mix model, which paid channel group had the highest attributed revenue in September 2026? Paid channels only, by channel group”, ask “Cut that channel group’s spend by 20% next month and show the expected effect on MMM-attributed revenue, first with average ROAS and then with a marginal estimate from the settled weekly data”, or after a Q3 ROAS table, “Which of those channel groups improved versus Q2 2026? Same groups, same method”. What to expect, and where the answers go wrong:

The feed carries the vendor’s modelled results: spend, MMM-attributed revenue, last-click and platform-reported revenue, and new-customer conversions. It carries no response curves, saturation points, headroom, adstock or confidence intervals. Channel contribution, ROAS by measurement lens, period comparisons and new-customer questions can be answered from the data. Saturation, marginal ROAS, what-if and “which channel to cut” answers are the analyst’s own estimates. Ask how an estimate was made, and treat it as a starting point for a spend test.
Grouping by media platform folds the zero-spend organic rows into the paid platform. An organic channel group can share a media_platform value with a paid group, so ROAS, a last-click comparison or a new-customer CAC by media platform includes credit the spend did not buy. Say “paid channels only, by channel group” in any question about return on spend or cost per new customer. A comparison table added to an answer about something else can still be by media platform, so read side tables with the same care.
For a headline or board number, ask: “According to the media mix model, what was blended MMM ROAS for [quarter], paid channels only, by channel group, as total attributed revenue divided by total spend?”
Email and SMS groups can carry their tool fees as spend, so they count as paid. Ask for them to be listed separately if you want media spend only.
A bare question such as “How is our ROAS looking?” can be answered from the order and ad tables rather than the MMM. Put “MMM” or “according to the media mix model” in the prompt when you want the modelled figures, and name the window.
The most recent days of the feed are not settled, and the feed ends before today; is_measurement_complete marks the rows that are. For a monthly number, ask for a month that has closed. If the answer falls back to a different window than you asked for, repeat “paid channels only, by channel group” for the fallback as well.
To check how fresh the feed is, ask for the vendor’s latest run across all rows (vendor_last_completed_at). An older month keeps the run time from its last rebuild, so one month’s value can look stale when the feed is current.
Year-over-year comparisons only cover the vendor’s history, which can be shorter than the platform’s. A channel can show up as “new” rather than grown. Ask for the first date the feed has for that channel before quoting a growth rate.
MMM-attributed revenue is modelled credit. It does not add up to order revenue, and the vendor’s channel groups are not the same as SourceMedium’s sm_channel.
The vendor may not split new and returning customers for every channel group. Ask for CAC as “spend divided by MMM new-customer conversions, leaving out groups the vendor does not split by new customer”. A blended CAC that keeps such a group’s spend in without its customers comes out too high.
Questions the feed cannot answer (a confidence interval, an adstock decay rate, a platform that is not in the feed) still deserve the same qualifiers, so that any point value or proxy quoted beside the “not in the data” answer is the paid, channel-group figure.
Campaign questions read obt_media_mix_performance. Ask for ROAS recomputed from summed revenue and spend rather than an average of per-row ratios, and treat a campaign ranking inside one platform as weaker evidence than the channel-level figures.

Customer support

These work when a help desk is connected (the integrations list shows which are supported). Tickets land in obt_customer_support_tickets, one row per ticket with status, channel, tags, assignee, resolution time and CSAT. The conversations themselves land in obt_customer_support_ticket_messages, one row per message with the sender, whether an agent wrote it, and the text. Both are in the catalog. Theme questions read the message bodies; counts, channels and tags come from the ticket table; first response time and agent workload need the messages.

Do we have customer support data in SourceMedium? Check the data catalog and tell me which tables hold support tickets and the individual support messages, what grain each is at, and which support platform the data comes from.
What are our customers complaining about the most? Look at the actual support conversations from the last 30 days and give me the top themes with rough ticket counts for each.
What are the top support ticket themes in the last 30 days based on ticket tags? Exclude spam and give me ticket counts per tag.
How many support tickets did we receive in the last 30 days, by communication channel? Exclude spam.
What is our average first response time on support tickets created in the last 30 days? Break it out by communication channel and explain how you computed it.
In the last 30 days, what share of support messages came from customers versus our agents, and which agents sent the most replies?
Which products generate the most support tickets? Use the last 30 days.

What to expect, and where the answers go wrong:

A theme question over the message bodies is the slowest thing on this page, because the analyst samples and classifies conversations in SQL. Expect ranges rather than exact counts, and ask which tickets could not be classified, since some channels carry no transcript.
Tags are the fast path to themes, but many tags are routing labels (channel, store, automation, priority) rather than topics. Ask for the two to be separated.
There is no first response time column. It has to be derived from the messages table: the first agent message the customer could see after the customer’s first message, leaving out tickets an agent started and internal notes. Ask for that method to be stated, and for a median next to the average, because a few slow replies pull the average up. Some help desks log both sides of a phone call at once, so those tickets read a first response time near zero.
Nothing on a ticket names a product. “Which products generate tickets” can only be answered by linking tickets to the customer’s recent orders, which covers the tickets whose customer has an order and counts every product in those orders. Treat the result as an estimate; ask your support team to fill a product field in the help desk if you need the real answer.
A help desk connected to one store covers that store only. Ask which store the tickets belong to.
If no help desk is connected, the support tables exist but are empty.
The Gorgias and Richpanel integrations fill the support tables today, and they differ in internal notes, phone calls, channel names and team assignment. The platform notes on the messages table and the tickets table say how.

The analyst also covers inventory and zero-party attribution, such as post-purchase survey responses.

Raw data pulls

Pull order ID, order date, customer email, UTM source, UTM medium, and revenue for last week
Export customers acquired in March with email, first order date, and lifetime revenue
Show line items for SKU ABC-123 from the past 30 days
Pull campaign-level spend, clicks, impressions, and campaign name for TikTok last month

Definitions and schema

What is LTV?
How do you calculate conversion rate?
Which table has order data?
What columns are in the orders table?
What's the difference between first-click and last-click?
What are the possible values for order_type?

Health checks

How is my data health?
Which tables are stale?
How is my attribution health?
Why is so much traffic direct/none?

Tips for better answers

State a time range

Say “last 30 days,” “in January,” or “Q4 2024.” If you leave it out, the query picks one, and you will see it in the SQL. “Recently” is not a time range.

Name the metric you mean

“Revenue” is clearer than “sales.” “Orders” is clearer than “transactions.” Use the terms from your dashboards. See Metrics.

Ask one thing at a time

“What was revenue and how did AOV change?” works better as two questions in the same thread.

Use top N for rankings

“Top 10 products by revenue” is specific. “Best products” is not.

Use pull or export language for rows

“Pull order ID, date, UTM source, and revenue for last week” returns rows and a CSV. “What are the trends in orders by UTM source?” returns an analysis.

Add context to open-ended questions

“How can we improve Q1 performance given the new product line?” gives Deep Analysis something to work with.

Check data health first when it matters

A “last 7 days” answer is wrong if the last 3 days have not synced. Ask “How is my data health?” before an analysis you plan to present.

Reading results

What each part of a reply means, thread rules, and how to validate a number.

Table schemas

Column-level documentation for every table the analyst can query.

Start here

SourceMedium Agent

SourceMedium Agent Skills

Your own agent

Reference

Question types

Specific metric questions

Raw data pulls

Open-ended questions and Deep Analysis

Definitions and schema

Data and attribution health

Follow-ups

Example prompts by domain

Orders and revenue

Customers

Marketing and ads

Email and SMS

Products, subscriptions, and returns

Web analytics and attribution

Media mix (MMM)

Customer support

Raw data pulls

Definitions and schema

Health checks

Tips for better answers

Reading results

Table schemas

​Question types

​Specific metric questions

​Raw data pulls

​Open-ended questions and Deep Analysis

​Definitions and schema

​Data and attribution health

​Follow-ups

​Example prompts by domain

​Orders and revenue

​Customers

​Marketing and ads

​Email and SMS

​Products, subscriptions, and returns

​Web analytics and attribution

​Media mix (MMM)

​Customer support

​Raw data pulls

​Definitions and schema

​Health checks

​Tips for better answers

​Related

Reading results

Table schemas

Question types

Specific metric questions

Raw data pulls

Open-ended questions and Deep Analysis

Definitions and schema

Data and attribution health

Follow-ups

Example prompts by domain

Orders and revenue

Customers

Marketing and ads

Email and SMS

Products, subscriptions, and returns

Web analytics and attribution

Media mix (MMM)

Customer support

Raw data pulls

Definitions and schema

Health checks

Tips for better answers

Related