Helpling SG: Data Intelligence Progress

What we built

Every source lands in one governed BigQuery warehouse, isolated to Helpling's own data. Clavio, our analysis agent, sits on top and answers questions directly in Slack.

Google Search Console

1.48M rows, keyword and page level

Google Analytics 4

3.19M rows, sessions by channel and country

Google Ads

732K rows across 10 tables, down to search-term level

Customer Insights

85.7K rows, orders, GMV and marketing cost

Google Business Profile

4,752 reviews back to 2017

→

BigQuery warehouse

Single source of truth, refreshed on schedule, Helpling data isolated by design

→

Clavio agent

Writes and runs its own SQL, interprets results

Slack

Ask @Clavio anything, get answers with follow-ups, continue in threads

What is flowing today

Verified row counts and freshness, queried live from the warehouse this morning. Each source updates on its own cadence.

Source	Rows	History	Latest data	Status
Google Search Console	1,476,958	Jan 2025 onward	16 Jul 2026	Live
Google Analytics 4	3,194,386	Mar 2025 onward	01 Jul 2026	Live
Google Ads	732,046	Mar 2025 onward	18 Jul 2026	Live
Customer Insights	85,731	Jul 2024 onward	07 Jul 2026	Live
Google Business Profile reviews	4,752	Dec 2017 onward	21 Jul 2026	Live

Ask it anything, in Slack

No dashboards to learn. The team mentions @Clavio in the channel and the agent queries the warehouse, applies the right filters, and answers in seconds. It suggests follow-up questions and holds context inside a thread.

Adi

@Clavio what is our CAC by channel over the last 4 weeks?

Clavio

Here is CAC by channel for the trailing 4 weeks, from your Customer Insights data: SEM, Paid Social and Display broken out with orders, spend and cost per order, plus how each moved vs the prior period. Want this split by category, or compared to the same weeks last year?

Guardrails handled behind the scenes: the GA4 property is a global rollup carrying other markets and overlapping export batches, so exports are deduplicated and Singapore is isolated before any number is reported; reporting windows anchor to each source's real latest date; and a known gap in GMV reporting since early June is flagged rather than reported as real zeros.

What the data says

First cross-source analysis across all four datasets, run this morning. Every headline figure below was recomputed independently by a second analysis pass before it reached this page. Internal order and GMV figures are in EUR per Helpling reporting; Google Ads figures are in SGD.

12,176

Orders per month, Apr to Jun 2026 average

+24.9%

Order growth, H1 2026 vs H1 2025

€256K

Monthly 30-day GMV, April (last fully reported month)

€103.6K

Monthly marketing spend, Apr to Jun average

What is working

The business is growing: +24.9% orders year on year, with January the seasonal peak.
Organic customers are the best in the mix: €145 first-30-day GMV per new customer vs €87 from paid search, converting 2.2x better per click at roughly a tenth of the cost per order.
Referral (€15.6 per new customer) and CRM (+220% orders YoY) are quietly excellent.
Four paid engines deliver purchases at S$28-33: brand RLSA, both app campaigns, and reactivation. They get only 9.2% of budget.
AI search referrals went from near zero to ~46 orders/month since September, almost all new customers.
The repeat engine: each acquired customer is worth about 7.2 lifetime orders, and ~70% of orders arrive at effectively zero marketing cost.

What is not working

Paid CAC rose +79% in four months (€37.8 in March to €67.5 in June): orders fell 39% while spend rose 9%.
PMax was scaled 2.7x into deteriorating returns; its CAC tripled from €19 to €58.
The bidding signal is polluted: Google's algorithm is optimizing toward WhatsApp taps and app opens, not purchases, and sees only about half of real orders.
€104K went to channels costing over €150 per order in six months; S$15.8K of search spend in 12 weeks hit terms that never convert.
Non-brand organic clicks fell -34% YoY even as impressions grew 58% and rankings improved: a click-through collapse, not a visibility problem. SEO orders are flat while the business grows 25%.
Paid Social drove 54.8% of 2025 sessions but under 1% of tracked revenue.

How we make it work

Rebuild the Google Ads conversion signal: one deduplicated purchase action with real order value. Everything else in paid is downstream of this.
Rebalance spend: cap PMax at Q1 levels, cut the €150+ per order channels, scale the S$28-33 engines, pause unmeasurable YouTube brand.
Organic sprint on three arbitrage categories where demand, margin, and weak rankings meet: aircon, elderly care, move-out cleaning.
Rewrite titles and snippets on commercial pages: the site earns barely more clicks at position 1 than at position 5, which is abnormal and fixable.
Instrument the elderly care funnel: 51,910 paid clicks and 5,827 WhatsApp taps produced at most ~98 tracked orders in 12 weeks. Either bookings leak or they happen where we cannot see them.

The path to 5x

Scenario arithmetic from verified baselines, not a forecast. 5x GMV means moving from roughly €256K to €1.28M per month. Paid alone cannot buy it: at June's CAC, the extra acquisition orders needed would cost about €457K per month, 4.4x today's entire budget. The arithmetic that closes is 2.5x orders times 2x average order value, built from four levers.

Lever	Today, verified	The move	Modeled effect
Organic volume	12.4K organic clicks/mo; SEO at 534 orders/mo; 4,910 commercial keywords sitting at positions 4-15 with 163K impressions	Aircon, elderly, move-out category sprint plus click-through recovery	SEO to ~1,700 orders/mo at the current 46.4 orders per 1,000 clicks
Paid efficiency	€67.5 CAC in June vs €37.8 in March, same budget	Fix the conversion signal, cap PMax, reallocate to the S$28-33 engines	+837 orders/mo at constant spend if CAC returns to March
Order value mix	Cleaning is 69% of GMV at €17 per order; elderly care runs €311, deep cleaning €299, renovation €224, move-out €222	Grow the high-value categories organically where rankings are weakest today	Every 500 elderly orders/mo adds ~€155K GMV, worth ~1,800 cleaning orders
Repeat rate	7.2 lifetime orders per acquired customer; 86% of orders are repeat or untracked	Needs cohort data to decompose; the largest silent lever in the business	Unquantified until retention data lands in the warehouse

The category arbitrage in one line each: aircon is the largest commercial demand pool in the market (108K impressions per month) but our money page ranks around position 24, outranked by our own blog posts. Elderly care has the highest order value on the platform and almost no organic presence (position 18, 29 clicks a month) while paid carries it. Move-out cleaning is the number-two GMV category with core keywords already at position 3, one push from the top.

Honest caveats: GMV figures are first-30-day attributed values and understate lifetime revenue. The GMV feed has a reporting gap since 5 June that we have flagged for repair. Attribution between ad platforms and internal reporting never reconciles perfectly; where they disagree we show both.

Next steps

Priority	Action	Why now
1	Rebuild Google Ads conversion tracking: single deduplicated purchase action with order value, internal orders imported	Every bidding decision currently optimizes toward the wrong signal; this unlocks the paid CAC recovery
2	Paid rebalance: cap PMax, cut €150+ per order channels, scale app and reactivation, brand pause test	Roughly €17K/month of measured waste is redeployable immediately
3	Organic sprint: aircon money-page rebuild, elderly care cluster, move-out top-3 push, title and snippet rewrites	Highest-margin growth lever; demand is already there, position and click-through are the gap
4	Pipeline hardening: repair the GMV feed gap, tighten GA4 market isolation in the query layer	Keeps every future answer trustworthy
5	Clavio Phase 2: automated Monday morning performance narrative in Slack, alerts on CAC and click-through regressions	The team hears about problems in days, not quarters