MEASUREMENT · Chapter 5

Attribute AI-sourced pipeline honestly, or stop calling it ROI.

The hardest part of AI lead generation is not building the workflow. It is proving it pays back. Most reports either credit AI for everything that happens after a tool is bought or credit it for nothing because the channel groups hide the traffic in Direct. This chapter is how to land somewhere useful.

The two numbers a founder actually needs

AI-originated pipeline. Deals whose first touch was an AI surface (ChatGPT, Perplexity, Claude, Gemini, Copilot, an AI-sourced cold reply). Divide by AI tooling spend in the period. That's the literal payback number: defensible, narrow, conservative.
Human-hours saved by AI assist. SDR enrichment, meeting notes, scoring, follow-up drafts. Value the hours at the marginal cost of replacement, not a fantasy rate. That's the productivity number: bigger, fuzzier, but the one that captures most of AI's actual value in lead gen today.

Report them separately. Combined into one "ROI of AI" figure, they obscure each other and the number won't survive a board question.

The classification rule, by scenario

Common patterns and how to tag them. Apply consistently in CRM custom fields, not just in reporting macros that get forgotten in three months.

Scenario	Classify as	Tagging rule
Lead arrives via ChatGPT citation, fills out form, books call	AI-originated	First-touch referrer is chatgpt.com/perplexity.ai/claude.ai. Tag at source via UTM or referrer rule in GA4.
Lead found via AI-built target list, contacted by human SDR	AI-assisted	AI sourced the account. A human did the conversion work. Tag the contact-source field, not the deal-source field.
AI-drafted email, human edit, lead replies, deal closes	AI-assisted	Credit the channel (email) for sourcing. Note AI assist in the deal record for ROI reporting on AI tooling, not in the standard attribution report.
Predictive scoring flagged a low-priority lead that converted	AI-assisted	The lead came from somewhere else. AI changed routing. Count the original source for attribution; track the scoring win separately.
Cold outbound from a fully automated sequence with no human in the loop	AI-originated	If it works at all, attribute to outbound-AI. In most B2B contexts it will not work. The deliverability damage is a hidden cost the attribution report should also surface.

Recover the AI traffic GA4 hides

GA4's default channel group does not know what chatgpt.com, perplexity.ai, or claude.ai are. AI referrals land in Direct unless you tell GA4 otherwise. The fix is a custom channel group with regex source rules, covered in detail in the companion piece on AI traffic measurement. Without it, every "we don't get AI traffic" claim is a tooling artifact, not a finding.

See also the measurement chapter of the AI Marketing Playbook for the broader analytics setup this layers on.

Best for

Teams already running CRM-source tagging consistently on inbound.
Companies whose finance lead asks for AI ROI quarterly and accepts a defensible answer over a flashy one.
Marketing leads who don't need AI to look like a hero to keep their job.

Fails when

CRM source fields are populated inconsistently or auto-filled to the same value.
Leadership wants 'AI ROI' as a single number, no caveats.
Attribution windows are too short to capture B2B deal cycles.

Verify before

Your GA4 custom channel group surfaces non-zero traffic from at least three AI hosts.
Your CRM separates contact-source from deal-source on every record.
Your AI-originated number excludes deals where AI was just the second touch.

Frequently asked

What's the difference between AI-originated and AI-assisted pipeline?

AI-originated pipeline is sourced by AI from end to end: a ChatGPT citation that drives a form fill, or an AI prospecting tool that built the target list. AI-assisted pipeline is sourced by humans but accelerated by AI: an SDR who used AI to enrich the account before the call, or a marketing team whose AI-drafted blog post drew the prospect in. Both are real. Reporting them as the same number is how AI lead-gen budgets get justified on inflated math.

How do I track AI-driven traffic in GA4?

Build a custom channel group with rules matching chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com, and the bing.com AI-panel parameters. GA4 puts most of this in Direct by default, which is why CFOs see no AI traffic and conclude AI isn't working. The channel group surfaces it. The Licheo ai-traffic skill covers the exact regex rules for the current crop of AI surfaces.

Can I attribute pipeline to a specific AI tool?

For inbound, yes. Tag the source via UTM, referrer, or a hidden field that records the path that produced the conversion. For outbound and nurture, attribute to the channel (email, outbound, content) and log the AI assist separately in the deal record. Trying to attribute pipeline directly to 'the AI scoring tool' or 'the AI email assistant' produces vanity metrics that don't survive a board question.

What's the right way to report AI lead-gen ROI?

Two numbers. (1) AI-originated pipeline divided by AI-tooling spend: the literal payback. (2) Human-hours saved by AI assist across the funnel, valued at the marginal cost of replacement (a junior SDR's loaded cost, not a fantasy hourly rate). Together they capture the originated and assisted value without double-counting.

How long until the attribution data is reliable?

Inbound attribution stabilizes in 30 to 60 days as the channel groups collect data. Outbound and AI-sourced pipeline attribution takes longer, 90 to 120 days, because deal cycles are longer and the sample size of closed-won deals is what makes the attribution defensible. Reporting AI ROI in the first month is mostly noise.

What CRM fields do I need to make AI lead-gen measurement possible at all?

A lead source field that records the actual channel, not a generic 'website' bucket. A field logging whether AI enrichment or scoring touched the deal. A timestamp for first contact and first meaningful reply, so speed-to-lead is measurable. Without these three fields captured consistently at the point of entry, no amount of downstream reporting can reconstruct what actually happened.

Is it worth building a custom attribution model, or is a simple channel report enough?

A simple channel report is enough for most 5 to 50 person B2B teams. Multi-touch attribution models add real value once a company runs several paid and organic channels simultaneously with a large enough deal volume to make the model statistically meaningful. Building one earlier just adds complexity to a decision a straightforward source-and-outcome report already answers well enough.

How often should the AI lead-gen dashboard actually be reviewed?

Weekly for the operational numbers (new leads, response time, pipeline created) and monthly for the ROI and attribution numbers, since those need enough data to be stable. Reviewing ROI weekly produces noisy, misleading swings; reviewing operational speed-to-lead only monthly means a broken workflow goes undetected for weeks before anyone notices the drop.

Official sources

← Back to the AI Lead Generation guide