Intercom support-conversation volumes, the AI Tax Advisor / KI-Steuerberater launch impact
(1 May 2026), and engagement / conversion analysis for Belgium and Germany. Internal staff
(@accountable.eu) excluded.
New conversations created, Swan automation excluded. Stacked by country — Belgium and Germany shown as separate colors. The BE/DE split is estimated from the in-app conversation distribution (Intercom conversations carry no country field). Toggle granularity, year and country.
Split by country — Belgium and Germany shown separately. Each point = conversations ÷ active paid users (excl. free/trial/staff) that month; the headline KPI is Σ(conversations) ÷ Σ(active paid users) over the shown range. Dashed = linear trend per country. Responds to the country filter.
Split by country — Belgium and Germany shown separately. Denominator is unique users who submitted ≥1 tax submission that month (Amplitude uniques, not the submission-event count). Each point = conversations ÷ unique submitters; the headline KPI is Σ(conversations) ÷ Σ(unique submitters) over the shown range. Dashed = linear trend per country. Responds to the country filter.
Share on any given month — Intercom vs AI Tax Advisor
The two series above as a 100% split, from Jan 2025. Shows the AI Tax Advisor's rising share of monthly support touchpoints (Intercom conversations vs AITA active users).
Active sign-ups by sign-up year — support behavior since 2025
Each active sign-up (BE+DE, finished signup, all plans) placed in one bucket by what they did since Jan 2025: asked on Intercom only, asked the AI Tax Advisor only, both, or neither. 100% stacked by sign-up year — newer cohorts lean far more on the AI Tax Advisor. AITA = conversation tagged "Started from AITA"; Intercom = a direct (non-AITA) user-initiated conversation.
Source: Intercom conversations since 2025-01 (ex-Swan), joined per active sign-up (finished_sign_up = true)
Messenger & support volume
In-app messenger user messages — monthly
Solid lines: in-app AITA messages, conversation.created role=user (Intercom messenger before 1 May 2026, AITA after; vertical dashed line marks the switch). Amber dashed line: Intercom new conversations (ex-Swan), total, from Intercom. Responds to Year, Period, Country (the Intercom line is total, not country-split).
Intercom support conversations (ex-Swan) — 2025 vs 2026
Native Intercom volume, Swan banking automation excluded (tags 10631204 / 13016147). Responds to Year, Period, Channel, and Inbox team (switches to per-team windows when a team is picked).
Bars: conversations the AI routed to humans (tag 12284544), 2025 = 0 (tag is 2026-only). Amber line (right axis): share of ex-Swan Intercom conversations carrying the tag. Responds to Year, Period.
Source: Intercom tag 12284544
Conversations by support team
Team-assigned Intercom volume (~90% of total; rest unassigned). Responds to Year, Period, Country, Inbox team. Banking = Accountable Banking / Swan.
Share of each plan’s active users who created at least 1 AITA conversation (unique AITA users / active users). Intercom reach by plan is Phase 2. Responds to Plan family (highlights the group).
Source: Amplitude conversation.created uniques and _active uniques by gp:subscription_plan (May 2026)
Banking-Swan confounder (excluded from clean read)
Swan automation tags exploded ~17× YoY — mostly email, unrelated to AITA. Responds to Year, Period.
Source: Intercom tags 10631204 / 13016147
Hypothesis scorecard
#
Hypothesis
Verdict
Evidence
Reading & caveats
What this cannot conclude
Intercom volume is boundary-differenced total_count, not a per-message export — treat ±a few % as noise.
Team breakdown covers team-assigned conversations (~90% of total); the rest is unassigned.
Channel split is not available crossed with team or country; filters apply to the charts that carry the dimension.
Trial→paid conversion is unresolved (the revenue event is not the paid-conversion marker).
Paid-active level is plan-definition sensitive; the May dip is partly an unusually heavy April-2026 deadline artifact.
Paid is subscription_plan that contains "taxes" or "bookkeeping" (conservative floor). For May, settled (signed up May 1 to 12) is the comparable number; raw full month is censored.
Source: Intercom contacts/search. Confidence HIGH on trend, MEDIUM on absolute.
Conversion to paid by contact type
Matured cohort (signed up 1 Mar to 19 May 2026). Each user is bucketed by whether they used the AI Tax Advisor (AITA), reached a human support team (CS), both, or never asked a question. Bars show the paid conversion of each group. Correlation, not causation.
Year over year signups (clean seasonality control)
BE filing season runs March to June. Comparing the same month across years separates season from genuine change.
Paid uses the Intercom plan attribute (current state), a conservative floor. It can lag billing, so reconcile against Stripe or Qonversion once. Confidence MEDIUM on the absolute level.
Year over year shows SIGNUPS only (the clean seasonality control). Year over year conversion is not shown because subscription_plan is current state, so 2025 users who converted then later churned or downgraded would read as not paid.
The contact-type comparison uses a full conversation to contact join (BE+DE), synced separately from the live cohort counts. AITA detection is a floor (the tag plus the opening template), so the AITA and both groups may be understated. Confidence MEDIUM.
Correlation is not causation. Users who reach out may already be more engaged. The clean test is a randomized proactive Tax Coach nudge to half of new trials, a separate experiment, not something this dashboard can claim.
User base — signups, tax submissions & active paid users
Tax submissions — monthly
tax-submission.created. May is the normal post-deadline trough. Responds to Year, Period, Country.