Paid Acquisition — The Decisions The System Must Get Right
Created: 2026-07-17 · Owner: Marita (Marketing Ops) · Method: design from real decisions first, then curate canon, then build.
Status: Foundation pass. This file frames everything else in PaidAcquisition/. Start here.
Principle: we are not building "a paid-ads AI." We are building a system that reliably answers a fixed set of money decisions, grounded in real ad data + an authoritative canon, with every answer cited and verifiable. Decisions come first; the RAG and the agent exist only to serve them.
0. Why this exists (the biting problem)
MultiSensor AI (industrial condition-monitoring; B2B, long sales cycle) has spent ~$41k YTD in paid media (LinkedIn, Google, Reddit) and can tie almost none of it to a contact, opportunity, or deal. Three independent systems agree the money is real but the measurement is broken:
- HubSpot CRM (portal 20335613): of 8,911 new contacts YTD, only 4 carry any paid attribution; 8,634 are tagged "Offline"; no "Paid Social" source bucket exists. (Source:
Lead Source Attribution Cleanup Tracker June 26 2026.xlsx, MSAI SharePoint.) - GA4: all paid channels combined drove ~663 sessions, and only 26 "Paid Social" sessions all year against ~$15k of LinkedIn spend; ~0 demo/contact events attributed to paid. (Same tracker.)
- HubSpot Ads screen: LinkedIn shows $14,985 spend / 65K impressions / 611 clicks / 0 contacts / 0 deals / ROI blank. (Same tracker.)
- Current read (mid-July Windsor pull, per task brief): LinkedIn (
$15.4k) and Reddit ($6.2k) returned zero platform conversions; Google returned 7 GA4 form-fills that never reached HubSpot pipeline.
The company has a ~$40k/month authorized ad budget (Katie Stine, 2026-07-09 ads-planning call) but is deliberately underspending because leadership "can't just say let's spend $40,000 when we haven't seen any conversions yet." So the real, blocking business decision is: fix measurement, then deploy the authorized budget with confidence. VP Marketing Alecia O'Brien has frozen channel-level ROI reporting and the "kill LinkedIn" decision until the data is trustworthy (Tracker, Phase 0).
1. The decision list (what the agent must get right)
Each decision states the question, why it bites, the inputs, and what a good answer looks like. These are the eval targets for the agent (see RAG-AND-AGENT-DESIGN.md).
D1 — Is conversion tracking broken, and exactly where in the chain?
- Question: Where does the click-to-CRM chain break, per platform (LinkedIn, Google, Reddit)? Tag firing? Conversion config? UTM capture? Source mapping? GA4 classification? Ad-account-to-CRM sync? Landing-page bounce before the tag fires?
- Why it bites: Every downstream decision (budget, channel, benchmark) is unsafe until this is answered. You cannot judge a channel you cannot measure.
- Inputs: ad-platform conversion settings; UTMs on live final URLs; HubSpot Original Source config + workflows; GA4 channel grouping; HubSpot Ads connection; landing-page speed / bounce; the three-way reconciliation (clicks vs sessions vs contacts).
- Good answer: a per-platform, per-stage diagnosis naming the specific broken component and the specific fix, ordered by blocking dependency, with the acceptance test (one live test conversion lands in HubSpot with correct source + campaign + UTMs).
D2 — Why did LinkedIn ($15.4k) and Reddit ($6.2k) return ZERO platform conversions?
- Question: Is zero tracked conversions the same as zero actual conversions? What specifically produces the zero?
- Why it bites: A "zero" reading is being used (or was about to be used) to justify killing LinkedIn ABX — a decision on broken data. Must separate "the channel failed" from "we never measured it."
- Inputs: whether platform conversions are set as primary vs secondary/observational; presence of Conversions API (CAPI) / server-side postback; Insight Tag / Reddit Pixel install + event mapping; UTM stripping via redirects (the WTWH redirect question, 2026-07-09 ads meeting); mobile landing-page load 2.5–2.7s → 46% bounce before GA4 records a session (2026-06-25 ads meeting) — the click→session collapse (hundreds of clicks → 86 sessions).
- Good answer: an explicit statement that the zero is a measurement artifact (secondary/observational conversions + missing CAPI + UTM/redirect loss + pre-tag bounce), NOT proof the channels don't convert, with the evidence for each cause and the fix per platform.
D3 — Why does Google return GA4 conversions (7 form-fills) that never reach HubSpot pipeline?
- Question: Google is the one channel showing some tracked conversions in GA4, yet those form-fills don't appear as HubSpot contacts/opportunities. Where is the GA4→HubSpot handoff lost?
- Why it bites: Even the "working" channel leaks. If GA4 sees 7 fills and HubSpot sees ~0 paid, the source-mapping / form-to-contact / lifecycle plumbing is dropping known conversions.
- Inputs: GCLID auto-tagging on/off; whether Google's UTMs survive to the HubSpot form submission; HubSpot Original Source drilldown mapping (paid landing as Offline/Direct); whether MQL/SQL lifecycle workflows re-stamp/clobber source; form → contact → deal association.
- Good answer: the specific point of loss (most likely source-mapping + a lifecycle workflow overwriting Original Source, and/or UTMs not persisted to contact properties on form submit) and the fix, validated by tracing one of the 7 fills end-to-end.
D4 — How do we fix ad-click → UTM → HubSpot attribution (the plumbing)?
- Question: What is the correct, durable target-state architecture so every future paid click resolves to the right source, campaign, and deal — and survives lifecycle changes and redirects?
- Why it bites: This is Alecia's explicit, board-level Q3 ask and the single prerequisite before scaling spend.
- Inputs: a locked UTM standard; HubSpot source map (Paid Social / Paid Search buckets); tracking-URL registration; a lead-source stamping workflow writing a separate clean property (never overwrite native Original Source); a drift-alert workflow; multi-touch attribution model; and — the durable layer — server-side event delivery (CAPI / server-side GTM) so tracking survives ad blockers, ITP, and consent loss.
- Good answer: the phased plumbing plan (tag at source → HubSpot source map → conversions & reconcile → attribution & restart) with owners (Marita verifies agency-owned rows; does not execute them), plus the server-side upgrade as the resilience layer.
D5 — Where should the paid budget actually go? (allocation)
- Question: Given a ~$40k/month authorized budget and a broken-then-fixed measurement system, how should MSAI split spend across Google (search/competitor/awareness), LinkedIn (ABX/competitor), Reddit, and any new channel — by funnel stage and by expected cost-per-opportunity?
- Why it bites: This is the actual money question. It must NOT be answered on platform-conversion counts (which are broken); it must be answered on pipeline / cost-per-opportunity, full-funnel logic, and B2B channel economics.
- Inputs: verified spend + CPC by channel (Google ~$6 CPC, LinkedIn blended ~$20+ / ABX ~$94 CPC, Reddit ~$1.30 CPC — Log 2026 + tracker); competitor-search performance (72% top-of-page, kept live); the full-funnel plan (awareness → competitor comparison → book-a-demo retargeting); B2B benchmarks from the canon; MSAI's long cycle + high ACV ($550K Walmart opp partly LinkedIn-influenced); the ELT decision to shift advertising 35%→47% of budget and PPC toward brand awareness/clicks-to-site.
- Good answer: a staged allocation (protect what demonstrably works, fix tracking before scaling, test Reddit at controlled spend, hold ABX decision until measured) expressed as ranges tied to expected cost-per-opportunity, explicitly gated on D1–D4 being fixed, and grounded in the canon's B2B allocation guidance — draft, for human approval.
D6 — When is a channel decision (e.g., "kill LinkedIn ABX") safe to make?
- Question: What evidence threshold must be met before pausing/scaling a channel?
- Why it bites: Prevents repeating the pattern of acting on broken data; encodes Alecia's Phase-0 freeze into a reusable rule.
- Good answer: a decision gate — no channel is killed/scaled until (a) tracking validated by a live test conversion, (b) ≥N tracked conversions or a defined spend/time floor, (c) cost-per-opportunity computed on pipeline not platform counts.
D7 — What benchmark is this channel being held to, and is it fair for B2B industrial?
- Question: For each channel/metric, what is the current credible B2B benchmark (CPC, CPL, CPO, CTR), and does MSAI's number beat/miss it once measured correctly?
- Why it bites: "Expensive" and "cheap" are meaningless without a benchmark and without correct measurement; LinkedIn's high CPC may be justified by ACV.
- Good answer: each MSAI figure placed against a cited, dated benchmark, with the industrial/high-ACV context applied.
2. Decision → knowledge-pack / canon mapping
| Decision | Primary knowledge-pack doc | Canon sub-area |
|---|---|---|
| D1 tracking broken where | KB/01-conversion-tracking-server-side.md |
Measurement/attribution |
| D2 LinkedIn/Reddit zero | KB/01-conversion-tracking-server-side.md |
Platforms + measurement |
| D3 Google fills lost | KB/02-utm-to-crm-attribution.md |
Measurement/attribution + HubSpot |
| D4 fix plumbing | KB/02-utm-to-crm-attribution.md |
UTM/CRM + server-side |
| D5 budget allocation | KB/03-b2b-budget-allocation.md |
Strategy/budget/bidding |
| D6 decision gate | KB/03-b2b-budget-allocation.md + KB/04-measurement-benchmarks.md |
Strategy + unit economics |
| D7 benchmarks | KB/04-measurement-benchmarks.md |
Benchmarks |
3. Weighting
MSAI is B2B industrial/SaaS, long cycle, high ACV → weight B2B sources (LinkedIn ABM, pipeline/CPO over platform conversions, MER/CAC on revenue). But the same system will serve ecommerce clients (TFC's Cali Extrax, TGH Houston) → canon also carries ecommerce measurement (Meta CAPI, ROAS, catalog) noted as secondary. See 01-CANON.md.