Customer acquisition cost is one of the most commonly cited metrics in SaaS growth. Teams optimize for it. Finance tracks it by channel. Boards ask about it every quarter. It's also one of the most systematically misleading inputs to budget allocation decisions — because it captures the cost of getting a customer in the door and says nothing about what that customer does in the subsequent 12 months.
The attribution model problem in B2B SaaS runs deeper than just CAC, though. Last-touch attribution — where the final trackable marketing touchpoint before conversion gets 100% credit for the customer — is the default model in most CRM setups and marketing analytics stacks. It's fast to implement, easy to report, and consistently wrong about which channels are actually driving growth.
Why last-touch attribution fails B2B SaaS specifically
In consumer e-commerce, last-touch attribution is imprecise but often defensible — purchase cycles are short, touchpoints are relatively few, and the conversion event is clear. In B2B SaaS with deals above $500 ACV, the purchase cycle routinely runs 30–120 days with 6–15 distinct touchpoints across organic search, paid channels, retargeting, sales outreach, review sites, and peer referral. Giving 100% credit to the touchpoint that happened to be the last one before the rep sent the contract is not attribution — it's a lottery.
The practical consequence: your paid search campaign that delivers demo requests right before deal close gets credited for deals that your organic content blog post warmed up six weeks earlier. Your SDR outbound sequence that initiated first contact gets zero credit. Your G2 review profile that the prospect used to validate their evaluation gets zero credit. The content team that produced the three-article series that moved the prospect from awareness to evaluation gets nothing. So the content budget gets cut and the paid search budget grows — not because the data says that's the right call, but because last-touch makes it look that way.
The attribution model landscape: what actually exists
Marketing attribution has several established models beyond last-touch, each with different assumptions and different failure modes.
First-touch: Credits the first known touchpoint with 100% of the deal value. Better than last-touch for measuring awareness channel performance, but equally wrong about the multi-touchpoint reality. It ignores everything that happened between first contact and conversion, including the CSM demo, the case study, and the comparison page visit.
Linear: Distributes credit equally across all tracked touchpoints. Directionally more accurate than single-touch models, but treats a 2-second ad impression the same as a 40-minute product demo. Blends out the significance of the most influential touchpoints.
Time-decay: Gives more credit to touchpoints closer to conversion. This captures the intuition that late-stage touchpoints (demos, pricing page visits, competitor comparisons) are more directly related to the purchase decision. The limitation is that it systematically undervalues the awareness channels that initiated the consideration cycle, which is exactly the attribution problem content teams are trying to solve.
Position-based (U-shaped): Typically splits 40% credit to first touch, 40% to last touch, and 20% distributed across middle touches. A reasonable heuristic that acknowledges both the initiation event and the close event. Still arbitrary in how it weights the middle.
Self-reported attribution: Asking customers at onboarding or in a post-purchase survey "how did you first hear about us?" This is the dark funnel answer that no model can capture systematically — word of mouth, conference conversations, Slack community mentions, analyst recommendations. Self-reported attribution is inconsistent and incomplete but often captures signals that your tracking stack misses entirely.
We're not saying multi-touch attribution solves the attribution problem completely — no model does. What we're saying is that last-touch, as a single-model default, produces systematically biased budget decisions that deprioritize awareness and consideration-stage channels in favor of conversion-stage channels.
The retention dimension that attribution models all miss
Every attribution model described above measures the same thing: which channels produce conversions. None of them measure which channels produce customers who stay.
A channel that produces customers at $800 CAC with 85% 12-month MRR retention is more valuable than a channel producing customers at $600 CAC with 52% 12-month retention. The math is obvious when you state it that way. The problem is that most SaaS teams run their channel evaluation entirely on cost-to-convert and have no systematic way to connect acquisition source to downstream retention behavior.
The right analysis is a channel retention matrix: for each acquisition channel, a cohort retention curve at 90 days, 180 days, and 12 months. The output is a straightforward table — channel in rows, retention period in columns — but the insight density is substantially higher than a CAC comparison.
The patterns that consistently emerge across growing SaaS companies:
Paid search cohorts tend to arrive with high intent but low fit. Customers who searched for a solution to a specific problem, converted, and churned once the immediate need was addressed. 12-month MRR retention in this channel commonly runs 55–70% — strong early, steep drop in M3–M6.
Organic content cohorts show longer consideration cycles and higher product fit at onboarding. Customers who consumed educational content before converting understood the product better and arrived with lower initial churn risk. Retention in this channel tends to run 15–25 percentage points higher than paid search at 12 months — which offsets the higher apparent CAC when you compute true LTV.
Referral and partner cohorts have low CAC but bimodal retention — high-performing clusters from strong-fit referrers alongside high-churn clusters from low-fit referrers. Channel-level averages mask this variance. Breaking referral cohorts by referrer type (customer referral vs. partner referral vs. integration marketplace) is often more revealing than the channel aggregate.
Outbound/sales-led cohorts have the highest CAC and often the strongest 12-month retention in complex B2B SaaS — sales-qualified leads with proper fit documentation, internal champions, and defined use cases arrive with lower churn risk. CAC-only comparison makes outbound look expensive. Retention-adjusted LTV often makes it look like the most capital-efficient channel you have.
The CAC payback period miscalculation
The standard CAC payback formula is: CAC ÷ (monthly MRR × gross margin). The number of months to recover acquisition cost from gross profit. A 12–18 month payback period is commonly considered acceptable for SaaS.
The formula has a structural flaw: it uses initial MRR at acquisition, assuming the customer is still active through the entire payback window. For a channel with 85% 12-month retention, this assumption is close enough. For a channel with 55% 12-month retention, the assumption is materially wrong — 40% of that cohort churned before month 12, and their churn happened throughout the period, not at the end, so the actual revenue recovered is substantially less than the formula implies.
A retention-adjusted payback calculation — where expected monthly revenue is weighted by cohort survival probability at each month — produces different numbers per channel. For a 55% retention channel, the retention-adjusted payback period runs 40–60% longer than the naive calculation. A channel that looks like a 10-month payback on the standard formula may be a 15–16 month payback in reality. If your board is using payback period as a growth efficiency signal (which most boards are), this gap matters.
Building the attribution model that actually works for your stage
The structural fix is joining acquisition data to retention data at the customer level. Your CRM or marketing attribution tool records which channel each customer came from. Your billing system records that customer's MRR trajectory. The join field is a shared customer identifier. The analysis is a 12-month cohort retention curve broken out by acquisition channel.
Building this join is a one-time data engineering task — typically a few hours of work if your customer IDs are consistent across systems, longer if they aren't (which is why this analysis is often deferred until someone builds a clean customer-level table in the data warehouse). Once it exists, running the analysis quarterly takes minutes.
Beyond allocation, the channel retention matrix informs the CAC ceiling for each channel. A channel with 12-month retention 20 percentage points above your median doesn't just produce better customers — over a 24-month customer lifetime, the compounding effect of higher retention on total revenue per acquisition can produce 2–3× the LTV difference suggested by the headline retention delta. The ceiling on what you can pay to acquire a customer from that source is proportionally higher.
The self-reported attribution problem and the dark funnel
Every multi-touch model shares one fundamental limitation: it only credits touchpoints your tracking stack can see. In B2B SaaS, a significant portion of the purchase journey happens in places that are invisible to any attribution tool — Slack communities where your product gets discussed, LinkedIn posts shared among peers, analyst recommendations, word-of-mouth mentions at industry events, and podcast appearances that reach someone at exactly the right moment in their evaluation cycle.
Self-reported attribution — asking customers directly "how did you first hear about us?" or "what primarily led you to purchase?" — captures signals that your model cannot. It's collected inconsistently and remembered imperfectly, but it fills in the dark funnel dimension that multi-touch models structurally miss. The most useful deployment is as a supplementary signal layer: use the multi-touch model for channel-level budget allocation, use self-reported attribution to identify categories of influence that aren't appearing in your tracking (if 30% of customers say "a colleague recommended it" but your CRM shows no referral source, you have an untracked channel worth instrumenting).
The combination of a retention-adjusted multi-touch model plus systematic self-reported attribution collection gets you closer to the full picture than either alone. It won't be perfect — attribution is fundamentally an estimation problem, not a solved one. But it's substantially more accurate than last-touch, and the decisions it informs (where to invest the next $100K in acquisition budget) are better for it.
Last-touch will keep your paid search team looking efficient right up until the quarter when NRR softens and nobody can explain why. The retention matrix is the answer to that question, built before you need it.