Evidence

Ten ordinary questions.
Ten quiet forks.

The failure mode of fast AI analytics isn’t a broken query — it’s a fluent, confident answer to a question the tool quietly misread. Here are ten questions a business user actually types, the ways each one forks into a different number, and the interpretation Spotonix makes visible before it runs.

See it on your dataGet demo access

How to read a trace

Each question below looks unambiguous to a person and forks into two or three different numbers in SQL. For each, we show the readable plan Spotonix surfaces — the Segments, Calculations, scope, and time window it selected — and, when the fork is material, the clarifying question it raises instead of guessing. The visible interpretation plan and the clarifying gate are live today Live.

01

How many active users did we have last month?

SaaS · engagement

Why it forks

  • “Active” = signed in vs performed a key action vs held a billable seat — often 2–3× apart.
  • “Last month” = previous calendar month vs a trailing 30 days.
  • Count distinct users vs distinct accounts.

The plan Spotonix surfaces

Segments
Users with ≥1 key action, excluding internal/test accounts
Calculations
Distinct user count
Scope / filters
Production tenants only; exclude internal org IDs
Time window
Previous calendar month, single value
It asks

“By “active” do you mean users who performed a key action, or anyone who signed in? The two differ by about 2× on your data.”

Silent failure mode: Returns signed-in users (last_login) and labels them “active” — counting people who never engaged.

02

What's our revenue this quarter?

Finance · revenue

Why it forks

  • “Revenue” = bookings vs recognized (ASC 606) vs billed vs cash collected.
  • “This quarter” = fiscal vs calendar.
  • Gross vs net of refunds and credits.

The plan Spotonix surfaces

Segments
All customers (no cohort filter)
Calculations
Recognized revenue, net of refunds
Scope / filters
Exclude internal test invoices; all product lines
Time window
Fiscal Q3 FY26, single value
It asks

“Do you want recognized revenue or bookings? Your board deck uses recognized; a pipeline review usually means bookings.”

Silent failure mode: Sums an amount column that turns out to be bookings, over the calendar quarter, against a recognized-revenue target.

03

Show me our top customers.

GTM · accounts

Why it forks

  • “Top” by revenue vs usage vs seat count vs lifetime value.
  • Window = current period vs lifetime.
  • Cut = top N vs top X%.

The plan Spotonix surfaces

Segments
All active accounts, parent roll-up applied
Calculations
Trailing-12-month recognized revenue, ranked
Scope / filters
Exclude churned accounts
Time window
Trailing 12 months; top 20
It asks

“Rank by revenue or by usage? And over the last 12 months or all time? Those pick different accounts.”

Silent failure mode: Ranks by raw SUM(amount) all-time with no parent roll-up, so a big account split across child records never appears.

04

What's our churn rate?

SaaS · retention

Why it forks

  • Logo churn vs gross revenue churn vs net revenue churn (expansion offsets).
  • Unit = customer vs seat vs dollar.
  • Denominator = customers at start vs average; monthly vs annualized.

The plan Spotonix surfaces

Segments
Customers active at the start of the window
Calculations
Gross revenue churn = churned MRR ÷ starting MRR
Scope / filters
Exclude free/trial; contraction counted, expansion excluded
Time window
Trailing 3 months, monthly (annualized note attached)
It asks

“Gross or net revenue churn? Net nets out expansion and can read very differently for you this quarter.”

Silent failure mode: Computes logo churn (customer count) and reports it where the reader expected revenue churn.

05

Which campaigns are converting best?

AdTech · performance

Why it forks

  • “Conversion” = click→signup vs signup→paid vs impression→click.
  • Attribution = first touch vs last touch vs position-based; 7-day vs 30-day window.
  • Spend-weighted vs raw rate.

The plan Spotonix surfaces

Segments
Campaigns with ≥1,000 impressions (stability floor)
Calculations
Paid-conversion rate = paid signups ÷ clicks, ranked
Scope / filters
Last-touch, 30-day; exclude branded search
Time window
Last 30 days, per campaign
It asks

“Convert to signup or to paid? And on a last-touch, 30-day window? Those reorder your top campaigns.”

Silent failure mode: Uses click→signup with same-session attribution, flattering campaigns that never produce paid customers.

06

How are our regular customers doing?

Commerce · cohorts

Why it forks

  • “Regular” = purchase frequency vs recency vs tenure vs cumulative spend.
  • “Doing” = revenue vs orders vs retention vs margin.
  • Comparison = vs last period vs vs all customers.

The plan Spotonix surfaces

Segments
“Habitual Buyers” — ≥3 purchases / trailing 90 days (approved Segment)
Calculations
Revenue and order-count QoQ change
Scope / filters
Exclude one-time promotional orders
Time window
This quarter vs prior quarter
It asks

“Your team has an approved Habitual Buyers Segment (≥3 orders / 90 days). Use that, or a different definition of “regular”?”

Silent failure mode: Invents an ad-hoc “≥2 orders ever” threshold, silently redefining a cohort the company already agreed on.

07

What's our NRR?

SaaS · expansion

Why it forks

  • Which start-of-window base cohort; new logos excluded.
  • Window = trailing 12 months vs quarterly annualized.
  • Currency at fixed budget FX vs spot.

The plan Spotonix surfaces

Segments
Customers active 12 months ago (NRR base cohort)
Calculations
(starting MRR + expansion − contraction − churn) ÷ starting MRR
Scope / filters
Exclude new logos acquired inside the window; budget FX
Time window
Trailing 12 months, single value
It asks

“NRR on the trailing-12-month base cohort, new logos excluded, at budget FX — confirm before I run it?”

Silent failure mode: Includes new-logo revenue in the numerator, inflating NRR by counting growth retention is meant to exclude.

08

Are we growing?

Exec · headline

Why it forks

  • Metric = ARR vs recognized revenue vs active users vs logos.
  • Basis = period-over-period vs year-over-year.
  • New vs net (does churn count against it?).

The plan Spotonix surfaces

Segments
All customers
Calculations
Net ARR growth YoY = (ending ARR − ARR 12mo ago) ÷ ARR 12mo ago
Scope / filters
ARR from active subscriptions; exclude one-time services
Time window
Now vs 12 months ago, single value
It asks

“Growth in ARR or users? And year-over-year or vs last quarter? Those can tell opposite stories right now.”

Silent failure mode: Charts raw signups over time — up and to the right — while net ARR is flat.

09

What's the average deal size?

Sales · pipeline

Why it forks

  • Center = mean vs median (a few large deals skew the mean).
  • Population = new business vs new + expansion vs all closed-won.
  • Value = ACV vs TCV; closed vs created this period.

The plan Spotonix surfaces

Segments
Closed-won opportunities, new business only
Calculations
Median ACV (distribution is right-skewed)
Scope / filters
Exclude renewals, expansions, and $0 internal deals
Time window
Deals closed this quarter
It asks

“Mean or median deal size? You have three deals over 5× the typical size this quarter, so the mean runs high.”

Silent failure mode: Returns AVG(amount) over all closed-won including renewals, blending new and expansion into one misleading number.

10

How many transactions were flagged as suspicious last week?

FinTech · risk

Why it forks

  • “Flagged” = by which rule/model vs above which score vs manual vs auto.
  • “Suspicious” = flagged vs flagged-and-confirmed.
  • “Last week” = ISO calendar week vs trailing 7 days; include reversed?

The plan Spotonix surfaces

Segments
Transactions with risk score ≥ approved review threshold
Calculations
Distinct transaction count
Scope / filters
Exclude reversed/voided; all rails; auto + manual flags
Time window
Previous ISO week, single value
It asks

“Flagged (entered the review queue) or confirmed suspicious? And last calendar week or the last 7 days? The counts differ materially for audit.”

Silent failure mode: Counts every row above a hard-coded score, including reversed transactions — a number no auditor can reconcile.

What these traces do and don’t claim

These are illustrative of the mechanism, not customer data, and the definitions shown are examples. Spotonix does not claim a single “correct” reading of any question — it makes the reading visible and yours to confirm. Approved definitions are persisted and reused for later questions Live — this compounds from what your team approves, not automatically from every query. See how it works and Context Graph for the detail.

Bring one question your company argues about

Watch Spotonix read it back before it runs.

Use a real recurring question. Inspect the plan, the clarifying gate, and the query bindings before anything executes.

See it on your dataGet demo access