Analytics and attribution

Without attribution, you are steering GTM by feel. You do not know whether the customer came because of an ad campaign, a cold email, content they found in a search, a podcast appearance, or a referral. You do not know which channel gives you the best customers. You do not know whether to increase spend on the channel that closed the deal or the channel that first touched the customer. You hire a “head of demand generation” and cannot tell if they are working. You run A/B tests on campaigns and cannot trace the results back to revenue.

Attribution is the answer to the question: which GTM activity drove this customer? It is the mechanism by which you connect dots between your investments (spend, effort, content, partnerships) and your outcome (revenue). Without it, you cannot optimize. With it—done right—you can move spend toward the channels and campaigns that actually move the needle, kill the ones that do not, and build a coherent GTM story that your board, your team, and your investors can believe.

Why attribution matters at different scales

Attribution matters differently depending on where you are. The diagnostic changes as you scale.

Early-stage (founder-led motion). You have one or two sales channels. Your co-founder is the salesperson. Most customers come inbound (referrals, word-of-mouth, a few cold emails that convert at 10%). You do not have a complex GTM—you have a customer who wanted the product and bought it. You track one metric: where did this customer come from? That is enough. Most early-stage attribution is simple: did this customer find us through a partner, an event, an article someone shared, or a cold email? A Slack message from the founder to the customer usually answers it.

But the moment you start running outbound campaigns, multiple content channels, or a referral program, simple source tracking breaks. You have customers who read three blog posts, attended a webinar, got an email, and then signed up. Which one drove the conversion? You need to know.

Series A+ (repeatable motion). Now you have multiple channels: maybe an SDR team running outbound, a content program, paid ads, strategic partnerships. Customers are touching multiple channels before they convert. Your CRM should track every touchpoint. You have enough volume that patterns emerge. You need to understand: which channels produce customers who stick around (low churn, high LTV)? Which channels have the lowest CAC? Which combination of touches most often leads to a close? You need multi-touch attribution, but it can live in your CRM. You do not need a data warehouse yet.

Scale stage (scalable, forecast-driven GTM). You have 5–10 active channels. Your motion is complex: maybe you run ABM alongside outbound, partner-led expansion alongside self-serve, multiple segments with different motions. You have thousands of touches per customer. Your CRM alone cannot answer the question. You need a unified data warehouse that can correlate spend, impression data, website behavior, email engagement, pipeline events, and closed revenue. You need to model attribution—not just track it—to understand incrementality and true ROI. You have attribution models, marketing mix modeling, and econometric testing to isolate channel impact.

The common founder mistake is to ignore attribution until Series A, then try to retrofit it. By then, your data is a mess. Your CRM has inconsistent tagging. Your website has no UTM parameters. Your email platform has no integration. You cannot trace 80% of your customers back to a source. You hire a Head of Marketing with a mandate to “fix” attribution, and they discover there is nothing to fix—no clean data to analyze. You lose six months to data cleanup.

The correct approach: measure attribution from day one, at whatever complexity matches your stage. Early-stage: a simple “source” field in your CRM. Series A: UTM parameters, multi-touch tracking, and source attribution built into your CRM workflow. Scale-stage: unified data pipeline, modeled attribution, and continuous measurement.

Single-touch attribution: why it is wrong (and why you will use it anyway)

Single-touch attribution assigns all credit for a conversion to one touchpoint. The most common variants are:

Last-click attribution. The last channel a customer touches before converting gets all the credit. Customer reads a blog post, gets an email one week later, clicks the email, and converts. All credit goes to email. Last-click is the default in most analytics tools (Google Analytics uses it by default). It is simple. It is wrong.

Last-click systematically overvalues remarketing and email, and undervalues awareness channels like content and ads. If content had not been there to provide context, the email would have been ignored. If ads had not built awareness, the email would have been spam. But content and ads get no credit.

First-click attribution. The first channel a customer touches gets all the credit. Customer sees an ad, reads three blog posts, gets an email, and converts. All credit goes to the ad. First-click overvalues awareness channels and undervalues conversion channels. If the ad had not been there, the customer would not have started the journey. But if email had not been there, they would not have converted.

Linear attribution. Every touch gets equal credit. Customer gets touched by four channels before converting; each gets 25% of the credit. Linear is less biased than single-touch, but it is still crude. The first touch is not as valuable as the touch that triggered the decision to evaluate.

The reason you will use single-touch attribution anyway: it is built into your tools. Google Analytics defaults to last-click. Your email platform reports on email-driven conversions (which is actually attributed-to-email, not caused-by-email). Your paid ads platform reports on click-based conversions. Your CRM reports on “sales cycle started from web,” which is a last-click proxy. You get a dashboard that says “30% of revenue from email, 20% from paid, 15% from partners,” and you feel like you understand. You do not. You understand which channels you are paying the most attention to measuring, not which channels actually drive revenue.

The founder mistake: over-optimizing based on single-touch attribution. You see that email has 40% of last-click conversions, so you invest heavily in email and cut content investment by half. Six months later, your email conversions crater because your inbound pipeline (fed by content) dried up. The content was not getting credit, but it was doing the work. Single-touch attribution made you invisible to what was actually happening.

Multi-touch attribution: the three models and when to use each

Multi-touch attribution distributes credit across the entire customer journey. There is no single correct way to do it. The choice of model depends on your GTM, your data quality, and your complexity. Three models cover most use cases:

Time-decay attribution. Recent touches get more credit than early touches. A customer’s journey has six touches: ad impression (month 1), blog post (week 2), email (week 4), webinar (week 6), another email (week 7), and sales call (week 8). Time-decay gives the sales call the most credit, then the webinar, then the email, and the ad almost none.

Time-decay works when your GTM is awareness → consideration → decision. Awareness channels (ads, content, events) start the journey. Decision channels (sales calls, product demos, proposals) end it. Time-decay reflects that reality. The closer to purchase, the more influence a touch has.

Time-decay does not work when your journey is highly nonlinear. If 50% of your customers are inbound (they already know you), they skip the awareness phase. If your motion includes partner channels, where a partner is doing most of the work early on, time-decay undervalues that partner effort.

Linear attribution. Every touch gets equal credit. A customer’s journey has six touches; each gets 16.7% of the credit. Linear works when you believe all touches are equally important. It is sometimes true. A customer who reads a blog, gets an email, and attends a webinar might really need all three; none alone would have driven the conversion. Linear is fairer than last-click, but it is often too generous to low-impact touches.

Algorithmic (data-driven) attribution. A model looks at all customer journeys and learns which touchpoints most often appear in conversions vs. non-conversions. If customers who converted had an average of 5.2 touchpoints and non-converters had 3.1, the algorithm learns that each touchpoint is worth incrementally more. If content-to-sales-touch sequences convert at 80% and ads-without-follow-up convert at 5%, the algorithm learns that content is high-value and orphaned ads are low-value. The model then assigns credit based on the touchpoint’s incremental impact on conversion probability.

Algorithmic attribution is the most accurate if your data is clean, but it is also a black box. You do not always understand why a touchpoint got credit. It requires volume—you need thousands of journeys to build a model—and you need clean data. Do not try algorithmic attribution until you have Series A+ volume and a unified data warehouse.

Building an attribution system: the three layers

An attribution system is not one thing. It is three layers: data, tracking, and analysis.

Layer 1: Data collection and hygiene. You cannot run attribution on bad data. This is where most companies fail. Invest in discipline.

• CRM fields. Your CRM must have a lead_source field (where did this person first come from?), a contact_method field (how are we reaching them?), and ideally a journey field (a log of all touches). Most CRMs have source fields; use them consistently. Do not have ten different people filling in lead source ten different ways.
• UTM parameters. Every link you own (emails, ads, content, social) should have UTM parameters: utm_source (where is this link?), utm_medium (email, paid, organic, referral), utm_campaign (which campaign?), and optionally utm_content (which specific creative?). If a link does not have UTMs, you cannot trace it back. UTMs cost nothing; discipline costs only time.
• Website and email tracking. Your website analytics should track: page visits, click events, form submissions, and the source of each visitor (via UTMs and referrer). Your email platform should track open, click, and conversion events tied to the contact. Your ads platform should track impressions and clicks.
• CRM integration. Your email platform, ads platform, and website analytics should integrate with your CRM so that touches are logged automatically, not manually. Manual data entry is where errors creep in.

The diagnostic: can you take a random customer from six months ago and trace their complete journey from first touch to purchase? If you cannot, your data is not clean enough for attribution.

Layer 2: Tagging and naming convention. Humans are the problem. Two people running campaigns use different naming conventions. One calls it “Q2_outbound_sdr,” the other “outbound_sdr_q2.” One tags partners as “partner,” the other as “partnership,” the other as “referral.” Your database becomes a mess of 50 variations of the same campaign. Rules prevent this.

Create a tagging schema and enforce it:

• Campaign naming: [YEAR][Q][MOTION]_[CHANNEL]_[DESCRIPTION]. Example: 2025Q1_outbound_sdr_cold_email. Everyone uses this.
• Channel taxonomy: Agree on twelve channels (e.g., outbound, inbound email, paid search, paid social, content, events, partnerships, referral, webinars, communities, sales-assisted, direct). Every lead maps to one primary channel.
• CRM validation: Set up CRM rules that flag entries that do not match your schema. Do not allow a lead source that is not on the approved list.

One startup I worked with had 300 variations of “partner” in their lead source field after two years of chaos. Cleaning it up took three months. They should have enforced the taxonomy on day one.

Layer 3: Analysis and modeling. Once data is clean, you can model. The simplest model for repeatable stage: cohort analysis.

Pick a time period (say, Q1 2025). Look at all customers who closed in Q1. For each customer, look at their entire journey (every touch, every channel). Count how many of each channel appeared. Create a distribution: “Q1 customers touched an average of 4.2 channels; email was in 85% of journeys, content in 70%, outbound in 60%, paid ads in 30%.”

Then correlate with LTV: “Customers whose journey included content have 20% higher LTV than customers without content. Customers whose journey included paid ads have 15% lower CAC.” This is not perfect causation, but it is better than guessing.

At scale stage, you move to econometric modeling: isolate the incremental impact of each channel on conversion by holding other factors constant. This requires a unified data warehouse and statistical rigor. But the principles are the same: trace the customer journey, weight touches based on their predictive power, and allocate credit accordingly.

Founder mistakes: the three most common

Mistake 1: Not measuring attribution until Series A. You run early-stage GTM without tracking. By the time you raise and need to report “which channels drive revenue,” your data is inconsistent. You cannot answer the question. You hire a data person to clean it up. They spend six months on that instead of building the attribution model you need.

The fix: Measure attribution from day one. Early-stage: just track source in a spreadsheet. Series A: formalize it in your CRM. It costs almost nothing. It saves you months later.

Mistake 2: Over-optimizing single metrics. You see that your SDR channel has the lowest CAC, so you double down on SDRs and cut partnership investment. Six months later, partnership customers have higher LTV and lower churn. You realized too late that CAC alone is not the goal; unit economics—CAC relative to LTV—is. But you already scaled the wrong channel.

The fix: Track CAC and LTV by channel, together. The channel with the lowest CAC might have the lowest LTV. The channel with the highest CAC might have a 10x LTV and be the most profitable. Optimize for payback period or CAC:LTV ratio, not CAC alone.

Mistake 3: Treating attribution as a data problem instead of a GTM problem. Many founders hire a data analyst to “fix” attribution. But the real problem is that the GTM is too complex to measure cleanly. You have ten channels, overlapping motions, partners doing undocumented work, and sales team managing customer relationships without logging them in the CRM.

The fix: Attribution is a GTM discipline, not a data discipline. Simplify your GTM first. Agree on how many channels you will actively run. Enforce data entry discipline. Make sales log customer interactions. Then the attribution layer becomes simple.

Attribution at different scales: the operating model

Founder-led stage (pre-Series A). You have one or two channels. You track: source in a spreadsheet. You know most customers personally and remember how you met them. This is good enough. Do not over-engineer. The moment you have two channels running simultaneously, add a source field to your CRM.

Repeatable stage (Series A – B). You have 3–5 active channels. You run attribution in your CRM. You use time-decay or simple linear. You track CAC and LTV by channel. You do not have a data warehouse; you have CRM reports. This is sufficient. You run monthly cohort analysis: for customers who closed in the last month, which channels did they touch? This tells you the shape of your customer journey.

Scalable stage (Series B+). You have 5+ channels. You have a unified data warehouse. You run algorithmic attribution. You do marketing mix modeling to isolate channel incrementality. You can say: “If we increase spend on partner channels by $100k, we expect revenue to increase by $X, with an ROI of Y%.” You have a data team running continuous optimization.

The rules of good attribution

Rule 1: Measure what you control. Do not measure the impact of a partner’s marketing campaign on your revenue. You do not control it, and the data is usually incomplete. Measure the impact of your own touches: your emails, your ads, your content, your events. If partners are driving revenue, measure the revenue by partner, but attribute credit based on your touches within those relationships.

Rule 2: Invest in source tracking before revenue tracking. Before you measure attribution, make sure you know the source of every customer. If 20% of your customers have an unknown source, your attribution model will be biased. A high-volume channel that brings in bad customers might look good because you cannot trace those customers to a source.

Rule 3: Multi-touch attribution is more complex and less robust than you think. There are too many possible models, too many ways to weight touches, and no universal correct answer. Use the simplest model that answers your question. For most Series A companies, time-decay or linear attribution answers the question better than last-click. For scale-stage, algorithmic is worth it only if you have thousands of journeys and clean data.

Rule 4: Do not trust any attribution model that does not account for touch order and time. If a customer gets 100 emails from you and then converts, that is not the same as converting after one email. A model that treats them the same is broken. Time-decay, linear, and algorithmic models all account for this. Last-click does not.

Rule 5: Attribution is for allocation, not celebration. Do not run attribution so your head of demand generation can claim credit for all revenue. Run it so you can move money toward the channels and campaigns that actually move the needle. The moment attribution becomes a political tool, it loses value. Use it to optimize, not to justify.

What comes next: from attribution to incrementality

Once you have basic multi-touch attribution working, the next frontier is incrementality testing. This is: is the channel actually driving revenue, or would customers have converted anyway?

A customer gets 100 impressions of your paid ad over a month, then converts. But would they have converted if they had seen 50 impressions? Would they have converted with zero impressions? You cannot know from observational data alone. You need to run an experiment: show the ad to some people (treatment group) and not to others (control group). Measure the difference in conversion rate. That difference is the incremental impact.

Incrementality testing is the right way to measure channel impact, but it is also expensive and requires volume. Most companies do not get there until they are scaling aggressively and have millions in ad spend.

For now, get the basics right: clean data, consistent tagging, multi-touch attribution, and honest CAC and LTV tracking by channel. That will get you 80% of the way there.

Next: C8: Agentic & AI-era GTM — where GTM tasks move from human execution to agentic automation, and how to evaluate agents for incrementality and causal impact (not correlation).