askbowtie

When signups drop, I ask one question

Ben, founder of askbowtie

When a number drops, the worst part is the tab-switching. Analytics in one tab, Search Console in another, the deploy log in a third, and you're squinting at dates hoping a pattern shows up. I stopped doing that. Here's the one question I ask instead, shown on our demo data.

The short version

Signups dropped 39% week-over-week (demo data). I asked one question, and the agent joined the deploy log, conversions, Search Console, and exit rates in one go, then handed back a diagnosis with every number cited.

  • The drop was real: 84 signups/week to 51 (demo data). But traffic only fell 14%, so conversion rate itself had dropped.
  • The cause was split across four tools. A deploy rewrote /pricing meta titles (impressions down 31%, demo data) and changed the signup flow (exit rate up from 24% to 58%, demo data).
  • One MCP joined them. The agent correlated dates across deploy logs, conversions, search, and page metrics in a single conversation.
  • Single dashboards each showed a fragment. Only the join showed all three problems at once.

Did signups actually drop?

Demo data: 39% week-over-week, 84 to 51/week. Conversion rate fell, not just traffic.

What changed on July 1?

Demo data: one deploy, two changes. Rewrote /pricing meta titles and shipped a new signup flow.

Why did Google send fewer people?

Demo data: /pricing impressions fell 31% after the meta rewrite, position slipped from 4.2 to 7.8.

Why did fewer of them convert?

Demo data: exit rate jumped from 24% to 58%. Also: validation error on the plan picker for mobile.

The headline facts (all demo data, pulled from one query):

39%conversion volume drop in one week (demo data)
31%impressions fall for /pricing after meta rewrite (demo data)
58%exit rate on /signup after deploy, was 24% (demo data)

How I got the answer in one question

what I actually typed

why did signups drop last week?

That's the whole question. askbowtie pulled conversions, the deploy log, Search Console, and exit rates on its own, lined them up by date, and traced it to the deploy. I never named a tool.

The diagnosis: one deploy, three wounds

Step 1: Confirm the drop is real. get_conversions, 28 days. Result: 84 to 51/week, 39% drop, starts July 1 (demo data). Conversion rate fell, not just traffic. This is what you lead with: numbers, not guesses.

Step 2: Check what changed. get_notes, the product log. Result: One deploy on July 1 with two changes: new signup flow and rewritten /pricing meta titles (demo data). The timing matches the drop start date exactly. That is a lead worth following.

Step 3: Did Google react? get_search, 90 days. Result: /pricing impressions down 31% after July 1, average position 4.2 to 7.8 (demo data). The meta rewrite cost visibility. That explains the 14% traffic dip. Top-of-funnel leak confirmed.

Step 4: What about the people who still arrived? Traffic down 14% does not explain signups down 39%. get_top_pages sorted by exit_rate. Result: /signup exit rate jumped from 24% to 58% after the deploy (demo data). People reached the new flow and left. get_incidents: validation error on the plan picker for mobile sessions (demo data). The signup flow broke for a slice of traffic.

The verdict: one deploy, three compounding wounds. The meta rewrite cut search impressions (fewer people in). The new plan-picker step drove exits (fewer of them out). The validation error threw silent failures (no one knew). None of them were visible from any single dashboard. The correlation across deploy date, impressions, exit rate, and incident reports is what made it obvious, and the agent could only correlate because every silo was queryable from one place.

Why four dashboards miss this

Each tool in the split-stack version is individually fine. The failure modes are all at the seams:

Joining these by hand means exporting three date ranges, staring at charts, and hoping a pattern jumps out. An agent with one MCP does the join in seconds and cites its sources so you can verify every number. askbowtie detects. The agent reasons. You decide.

The whole story boils down to this: a single dashboard can show you one fragment. Put every signal in one place, and an agent can show you all of them correlated by date. The human eye misses the seams; the joined query makes them impossible to ignore.

What you can ask about your own funnel

Did we lose traffic, or did conversion rate drop?
`get_conversions` (conversion count) vs `get_traffic` (sessions), same period. If volume dropped but traffic held, it's a rate problem, not a top-funnel problem.
When did the drop start?
`get_conversions` with period. Look at the trend line, not just the total. Most drops are not gradual.
What changed near that date?
`get_notes`: deploy log and config changes. Correlation with a deploy is a lead, not proof; quote the note and show the before/after on each metric.
Did Google send fewer people?
`get_search` impressions, before/after the deploy date. Impressions down plus position down equals visibility problem.
Did the page itself break?
`get_top_pages` sorted by exit_rate (exits show where sessions end). Then `get_incidents` for the page that is bleeding visitors.
Is something throwing silent errors?
`get_incidents` ranked by impact, then `get_incident_detail` for the failing URLs and selectors. Most teams miss these entirely.

MCP reference (for agents)

The tools for this diagnosis, in the order you would run them.

  • get_summary { domain, period? } → One-call health briefing. Start here for "how is the site doing".
  • get_conversions { domain, period? } → Conversion counts and rates. Confirms whether a drop is traffic or rate. Call this first when something feels wrong.
  • get_notes { domain } → Product log: deploys, config changes, pivots. Always read this before diagnosing any dip. A recent deploy usually explains it.
  • get_traffic { domain, period? } → Sessions, pageviews, bounce. Sizes the top-of-funnel change.
  • get_search { domain, period? } → Google Search Console clicks, impressions, positions. The "did Google react" check.
  • get_top_pages { domain, period? } → Per-page metrics including exit_rate. Sort by exit_rate to find where sessions end.
  • get_incidents { domain } → Active issues ranked by impact. Drill in with get_incident_detail for failing URLs, messages, selectors.
  • get_landing_pages { domain, period? } → Entry pages and performance. Spot which doors closed.

Recommended loop for "why did conversions drop": get_conversions (confirm and date the drop) → get_notes (what changed near that date) → get_search plus get_traffic (fewer people arriving) → get_top_pages by exit_rate (more people leaving) → get_incidents (anything broken) → report with every number cited and the deploy note quoted. Correlation with a deploy date is a lead, not proof. Show the evidence; let the human decide.

Gotcha: correlation is not causation, but it is a lead worth following. When the conversion drop, impressions drop, exit-rate spike, and deploy all align on the same day, that is not coincidence. Pull the before/after on each metric and quote the deploy note so the human can verify.

That diagnosis used to take a day of exporting spreadsheets and squinting at charts. Now it is one conversation. Want to diagnose your own signup dip? Connect your site to askbowtie.