Private beta · Limited onboarding. Request access to join
Intently AI
Ask runs on the same first-party capture, attribution, and ad spend as your reports — so answers name the model, window, and ROAS instead of inventing credit.
Ask Intently AI
Reading reports…
“Which campaigns have ROAS under 0.5x on last-touch?”
Architecture
Capture, revenue, and spend feed Attribution before you ask. Attribution Ask is grounded in those reports (same models and windows you inspect) while behavioral Ask runs on the same first-party events. Tools that only see GA4 pageviews or ads UI metrics rarely match that stack.
01 · Capture
Page views, forms, scroll, ecommerce, and custom events land server-side from your site. That is the same signal multi-touch attribution and behavioral Ask run on.
02 · Revenue
Purchases and HubSpot/Salesforce closed-won attach to linked journeys, so attributed revenue and pipeline credit use the same touch models you inspect.
03 · Ad spend
Synced or CSV spend lands in Attribution so campaign ROAS is computed on your models. From the same surface you can pause a campaign or apply +20% daily budget on those networks — confirm required.
04 · AI layer
Attribution Ask is grounded in those reports: model, lookback, channel totals, delivery rates, and ROAS recommendations. It does not invent credit splits. Behavioral Ask covers forms, optional visitor engagement scores (Ask depth), and conversion paths on the same capture data.
Why it's rare
Behavior, attributed revenue, and ad spend already live in Attribution before you ask, so AI reads finished reports rather than assembling them at ask time.
Attribution answers name the model and lookback window, and quote channel revenue, delivery rates, and campaign ROAS from reports you can open yourself.
Failed destination sends get a delivery diagnosis path. Attribution Ask also sees verified delivery rates and match quality, so answers reflect what actually reached each platform.
On-site actions
Experience is the investigation layer on the same first-party capture: prioritized tests to try, readiness for the signal you need, optional visitor engagement scores when Ask-depth Engagement Tracking is on, converter vs non-converter cohort gaps, and Ask on that evidence. Associations to try, not guaranteed lifts.
Experience
On-site investigation for attributed traffic
Tests to try, not guaranteed lifts · Meta paid · last 30d
What to try next
2 rankedReduce abandon on “Company size”
highIn Meta paid traffic, non-converters abandon this field more often than converters — a form friction association to test, not proof of cause.
Non-converter abandon 50%·Converter abandon 12%·Gap +38pp
Get more of this cohort onto /pricing
highConverters from branded search hit /pricing more often than non-converters. Test routing or creative that lands more of this traffic there.
Converter rate 82%·Non-converter rate 41%·Gap +41pp
Engagement on conversions
Evidence · credited traffic
Low
18
Mid
41
High
52
No score
9
Investigation readiness
Core capture · Ask depth
Core
CI · Forms · Sections
Depth
Scroll · Clicks
Cohort gaps
Converters had · non-converters often didn't
| Experience | Type | Conv. | Non-conv. | Gap |
|---|---|---|---|---|
| /pricing | page | 82% | 41% | +41pp |
| Company size | form | 12% | 50% | +38pp |
| ROI calculator | section | 55% | 22% | +33pp |
“Which pages do converting Meta sessions hit that non-converters miss?”
New · MCP
A read-only MCP server exposes the same attribution and Ask builders your dashboard uses. Connect with your account API key (adt_…) — no separate MCP key. Tools stay scoped to your account.
Cursor MCP config
https://getintently.io/api/mcp{
"mcpServers": {
"intently": {
"url": "https://getintently.io/api/mcp",
"headers": {
"Authorization": "Bearer adt_YOUR_API_KEY"
}
}
}
}Claude Code needs "type": "http" in its MCP config — see the setup guide. Full Claude Code steps
Interactive demo
Pick a prompt. The demo streams an answer with citations and which data layers were used. Numbers are illustrative. In-product Ask cites your live reports.
Ask Intently AI
Demo · illustrative data
Try a question
“Which campaigns have ROAS under 0.5x on last-touch?”
FAQ
Grounding, citations, MCP, and what we do not claim yet. Book a demo if your stack needs a walkthrough.
No. Ask runs against the same attribution and analytics builders your dashboard uses — models, lookback windows, channel revenue, campaign ROAS, delivery rates, and CRM closed-won when connected. It is not free-text retrieval over a dump of events.
Attribution answers cite the model, lookback window, and the report numbers behind the claim. You can open Attribution and reconcile the same figures. Ask does not invent credit percentages or ML-assigned credit splits.
Spend and ROAS (which campaigns are under target), multi-touch model disagreement, channel and closed-won credit when CRM is linked, delivery and match quality, plus behavioral questions on forms, content engagement, and converter vs non-converter paths on the same first-party capture.
MCP is the same read-only attribution and Ask path exposed to Cursor, Claude Code, and Claude Desktop over Streamable HTTP. You authenticate with your account API key (adt_…). There is no separate MCP key, and v1 does not include write tools such as pausing ads.
Yes on Meta, Google, and Microsoft when spend sync has stored campaign ids. From Attribution, pause/scale suggestions can Apply pause or Apply +20% daily budget after you confirm. That write-back is in-product — not via MCP. Shared or non-daily budgets may skip scale with a clear message; CSV-only networks do not get write-back today.
Not yet. You get cited answers in-product and via MCP today. Saving Ask answers as widgets or email digests is on the roadmap — we do not claim it until it ships.
Ask and MCP access follow your Intently plan and entitlements. On a demo we map whether Attribution Ask, behavioral Ask, and MCP fit the stack you already run — capture, destinations, spend, and CRM.
Ask is only as useful as the pipeline behind it. Connect first-party capture, destinations, and (usually) Stripe or CRM and ad spend first. Until those reports exist, there is nothing solid to cite — which is why we walk the stack on a demo rather than promising answers out of an empty account.
We’ll walk through Attribution Ask on your stack: models, ROAS, pause/boost, CRM, and delivery proof in one place.