Building a Second Acquisition Channel Where Every Lead Lives on WhatsApp
Standing up direct marketing from zero inside an established field-sales business — 2x MQL output in a month, 75% less ad spend, and a diagnosis that ended a recurring argument about why deals died.
Context
Taktis Indonesia is Broom’s direct selling arm in BPKB gadai — vehicle title lending, where a borrower refinances against the ownership document of their car or motorcycle. Taktis sits between borrowers and the multifinance partners who underwrite and disburse.
The commercial model shapes everything: revenue is commission-based and recognised only at GoLive — the moment a loan is approved and disbursed. Every stage before that is a pure cost centre. A lead that reaches the final step and dies costs exactly as much as one that never engaged.
The business grew on field sales. From launch through Q3 2025, Taktis ran entirely on its commercial team working regions directly — roughly doubling monthly GMV as headcount and regional coverage expanded. That team remains the majority contributor today.
In Q4 2025 we opened a second front. Direct selling and direct marketing — paid acquisition into WhatsApp, qualified by telesales, submitted to leasing partners. A different motion, a different funnel, and no existing infrastructure.
The rationale was margin, not volume. A field-sourced deal carries a commission to the commercial team; a lead sourced through direct marketing carries none. Same revenue, roughly double to triple the contribution.
And the longer play: once the channel produces surplus volume, it can supply leads to the field team at a reduced commission rate — lowering blended acquisition cost across the whole business. Direct marketing was never intended to replicate the commercial team. It was built to change Taktis’s cost structure and give it a second source of demand it fully owns.
That channel is what this case is about. I owned it end to end.
The Problem
Three structural constraints, all at once:
- Every lead lives on WhatsApp. Acquisition ran through Click-to-WhatsApp Meta Ads into a WhatsApp CRM. No traditional CRM, no clean event stream.
- Revenue sits at the end of a four-stage funnel, and the final stage is controlled by external leasing partners.
- No consolidated view between “lead arrived” and “loan disbursed.”
Ad spend was flowing. We could not tell a good lead from a bad one, and we could not say why deals died. Every fortnightly review cycled through the same three anecdotes — the leads are bad, the partners reject too much, customers cancel too easily — with nobody able to settle it.
A Shared Funnel, Then a Pipeline
I anchored the operation on a four-stage funnel that became the shared vocabulary across Growth, Sales and Finance:
- 1_Leads
- 2_Kirim Hitungan
- 3_MQL
- 4_GoLive
Kirim Hitungan is the estimation sent to the borrower; MQL is submission to a leasing partner; GoLive is disbursement.
Every record carries three orthogonal fields — funnel_status (active / drop-off / GoLive), sub_tag (reason detail), and drop_off_by (customer / Taktis / leasing). That turned why do deals die? from a debate into a filterable query.
The lead intelligence pipeline ingests three raw sources weekly — WhatsApp CRM tag exports, raw chat archives, and the telesales logbook — into a single tagged master covering 14,480 leads across 27 fields. Phone normalisation, tag-to-funnel mapping, chat content extraction (vehicle type, year, image detection), logbook cross-referencing, dedup.
One rule made it trustworthy: the human logbook is ground truth. Chat signals and CRM tags can be wrong; an agent updating a shared logbook cannot be. Where signals conflict, the logbook wins.
On top of it: monthly trend charts, interactive SVG funnel flowcharts, and multi-slide leadership decks generated in Python on a two-week cadence.
Standing Up Telesales
August 2025. The largest and most expensive leak was Kirim Hitungan to MQL — the point where a lead has to hand over identity documents, vehicle papers and financial detail. Bots, static replies and generic broadcasts weren’t closing it.
I built a three-agent function from zero: business case, headcount plan, hiring brief, SOP, MQL definitions, escalation paths to leasing partners, and the weekly review cadence. Day-to-day queue management, coaching and script iteration went to the Supervisor.
Two design decisions mattered more than the hiring:
Keep the scope narrow. The team owned exactly one conversion — Kirim Hitungan to MQL — for its first eight months. The temptation was to hand them broadcast, re-engagement and cross-sell as well. Resisting that is what made the conversion improvement visible and attributable.
Make the logbook the data layer. The telesales logbook I designed for operational tracking became the ground-truth layer of the entire pipeline. Operational instrumentation and analytical instrumentation were the same artifact.
Diagnosing Where Revenue Actually Died
Q1 2026. Top of funnel was solved. MQL volume was climbing. GoLive — the only stage that produces revenue — was flat.
So I ran a MECE issue tree on the MQL to GoLive drop-off, structured into four mutually exclusive, collectively exhaustive branches:
| Branch | Cause | Fixable? |
|---|---|---|
| A — Bad Credit | Fails partner underwriting threshold | Only upstream, via targeting |
| B — Customer Cancel | Walks away post-submission | Partly — response times, follow-up cadence |
| C — Eligibility Mismatch | Vehicle age, ownership, geography, documents | Yes — telesales qualification SOP |
| D — Partner Rejection | Non-credit decline: portfolio, policy, capacity | Partly — partner routing and diversification |
Every MQL-stage drop-off was classified into exactly one branch using fields already flowing through the pipeline. Validated against a 100-record sample for zero overlap and full coverage.
What it found:
- Branch A was the largest at ~40% — a credit-quality problem created by upstream targeting choices, not by execution
- Branch C was small. The recurring hypothesis that telesales was the bottleneck was wrong. The team was fine; the leads were the issue.
- Splitting the sample around a February 2026 targeting change that spiked motorcycle share quantified the gap: car leads produce GoLives at roughly 4x lower cost-per-GoLive than motorcycle leads
Output was a 14-action plan mapped to branches, each with a named owner, presented and adopted as the Q2 2026 execution plan.
Re-engagement followed directly from it. May 2026 produced 2,062 cold leads and 58 MQLs — meaning ~2,000 already-paid-for conversations had died, mostly at the bot stage. Rather than re-running acquisition against the same people, I designed a three-layer filter: regex scan of chat history for rejection language, exclude motorcycles on cost-per-GoLive economics, exclude vehicles older than 2016 to match partner eligibility bands. 821 clean records survived — a 40% keep rate, at zero incremental ad spend — handed to telesales as a warm queue. The filter became a reusable monthly module.
Results
All figures below are the direct marketing channel only, roughly nine months from standing start. Field sales remains the larger contributor to Taktis GMV.
MQL output doubled in a single month: 58 (May 2026) to 116 (June 2026), with no additional agent headcount — so cost per MQL roughly halved.
MQL conversion rate rose from 2.8% to 4.9% on a larger lead base. Quality and quantity moved together, which is the uncommon outcome.
Cold lead volume grew 65% across Q2 2026 — 1,422 in April to 2,354 in June.
Ad spend fell 75% across Mar – May 2026 — Rp35M to Rp17M to Rp8.8M — without sacrificing lead quality, driven by monthly key-visual creative testing tied directly to CPL by vehicle segment.
Volume up 65% while spend fell 75% means cost per lead improved several-fold over the quarter, driven by creative testing and vehicle-segment targeting rather than by loosening quality thresholds, since MQL rate rose over the same window.
First fully attributed GoLive closed the loop from ad click to disbursed loan through the instrumented funnel in June 2026.
30+ GoLives since launch, none of them carrying the commission a field-sourced deal would. Against monthly ad spend that had fallen to Rp8.8M by May 2026, the channel was contribution-positive inside its first year.
Exact GMV, annual financing amounts, commission rates and margin structures are withheld.
And a shared language. Funnel stages, sub-tags and drop-off ownership became the standard reporting vocabulary across Growth, Sales, Finance and Business — reviews now read branch sizes month over month instead of trading anecdotes.
My Role
Senior Growth Manager, reporting to Head of Growth. Eight direct reports: Growth Analyst · Product Manager · Finance Ops Associate · Telesales team (Supervisor + 2 agents) · Graphic Designer · Performance Marketing freelance.
Owned: full-funnel growth strategy, paid acquisition, lead intelligence pipeline design, telesales function stand-up, the MECE diagnosis, weekly leadership reporting, and cross-team alignment on funnel definitions and metric ownership.
Partnered on: leasing partner strategy, sales operations, unit economics and commission modelling with Finance, broadcast and re-engagement operations.
Learning
Infrastructure pays back later than expected, and larger.
For roughly six months, the visible output of the pipeline work was charts leadership already directionally knew. It looked like overhead.
The compounding value only arrived when I could ask “of the 821 May bot-only drop-offs, how many had a car and a clean credit signal?” and get an answer in seconds. At that point telesales prioritisation, re-engagement targeting and paid acquisition could all move on the same evidence instead of three separate opinions.
The correction I’d make: build the application layer earlier. A CSV-and-notebook stack got us to insight, but it should have become a proper product six months sooner.
The second lesson is about why the channel existed at all. Judged on volume against a two-year-old field operation, direct marketing looks small. Judged on contribution per transaction — no commission at all — it’s a structurally better unit, and one the business fully owns. A new channel should be measured against the economics it changes, not only the volume it adds.
Instrument the funnel before optimising it. You cannot fix a drop-off you cannot name.