You spend $4,000 on ChatGPT ads in March, get 22 demo bookings, and receive an agency screenshot celebrating a $182 cost per lead. By the first review, sales says only six demos have happened, two have become real opportunities, and none have closed.

Both reports can be true. Neither, by itself, tells you whether the spend worked.

Everything here follows from one fact: a booked demo, a qualified opportunity, and a closed deal are different events, not interchangeable definitions of a conversion. Until you connect them, cost per lead tells you what you paid for the first step. Nothing more.

I’ll build that connection using one running example: a B2B SaaS product selling a $24,000 annual contract. The point is not to make the report more complicated. It is to stop one tidy number from hiding the business underneath it.

A conversion means the counter moved

In an ad platform, a conversion is an action you choose to count. Someone clicks your ChatGPT ad, lands on your demo page, submits the form, and a piece of code fires. The platform records a conversion.

That record does not establish whether the email is real, whether the person attends, whether sales qualifies the opportunity, or whether finance ever sees $24,000. I used to tell clients a conversion meant interest. I was wrong. A conversion means the counter moved.

This explains how two agencies can report different costs from the same funnel without either inventing a number:

  • Count all 22 bookings: $4,000 divided by 22 is about $182 per booking.
  • Count the six held demos: the same $4,000 becomes about $667 per held demo.
  • Count the two opportunities: it becomes $2,000 per opportunity.

Same spend. Same customers. Different counting rules.

None of those ratios is inherently useless. The booking cost describes acquisition at the top of this chain. The opportunity cost describes how much of that acquisition survives sales qualification. Trouble starts when a report labels the first number “acquisition cost” and lets you assume it describes a customer.

For the running example, I would define a qualified lead as a held demo with a company over 20 employees and a named decision maker on the call. That is a counting rule sales and marketing can inspect. “Good lead” is an argument waiting to happen.

Before comparing any cost-per numbers, write down exactly what each denominator counts.

Connect the form submission to a customer record

Once the event has a definition, it needs an identity. Otherwise, March’s booking and August’s contract remain two unrelated entries in two unrelated systems.

Start at the landing page. When a visitor submits the demo form, website measurement should record one enquiry tied to that visit. The draft’s measurement model uses Pixel and Conversions API to record website events. The essential sequence is simple: the ad produces a visit, the page produces a form submission, and measurement records that submission.

If the event is missing, fires twice, or counts every reload, the conversion total is wrong before sales gets involved.

What the landing page cannot establish is what happens next. That belongs in the customer relationship management system, or CRM. Consider booking number 14:

  1. The buyer books a demo in March.
  2. The demo takes place on April 2.
  3. Sales creates a $24,000 opportunity on May 10.
  4. The buyer signs on August 18.

The website records the initial enquiry. The CRM records its progress. The link between those records is what connects August’s revenue to March’s acquisition activity.

Google’s offline conversion import model illustrates the mechanism: capture the click identifier, store it with the customer record, and match later stages back to the original click. That is a useful model for understanding the chain, not a promise that every ChatGPT Ads integration supports the same imports. Verify the available integration before promising that later stages will appear in the ad platform.

The connection can also break before an import is even relevant. Consent settings or ad blockers can interrupt measurement. A buyer may use a personal email on the form and a work email with sales. Different devices can complicate the trail.

That is why I want the platform, website, and CRM reconciled rather than treated as interchangeable sources. If Ads Manager shows 22 conversions and the CRM shows 17 real contacts, someone needs to explain the difference. It might be a firing error. It might be an identity problem. The discrepancy alone does not tell you which.

Fix the broken link before debating which channel deserves the credit.

Follow March’s buyers instead of comparing March’s totals

Now the records connect. The next problem is time.

At the first review, our March group has six held demos, two opportunities, and no wins. That does not establish failure. It establishes where those buyers have reached so far.

Long SaaS sales cycles leave buyers in calls, security review, or procurement well after the report for their acquisition month arrives. Dividing March spend by March closed revenue mixes two different populations: buyers acquired in March and buyers who happen to sign in March.

The latter may have arrived months earlier. The arithmetic works. The comparison does not.

Instead, I would give the March bookings their own row and keep following that same group. This is origin-month reporting: the group stays anchored to when its leads were created, even as its outcomes change.

Check it at 30, 90, and 150 days:

  • How many bookings have become held demos?
  • How many have become qualified opportunities?
  • How many have closed, remain open, or stopped progressing?
  • How much signed revenue belongs to the group?

When deals take 90 to 180 days, treating an early report as final can trigger the wrong budget cuts. A young group and a failed group can both show zero revenue. They are not the same thing.

By August, suppose the March group has produced two wins worth $48,000 against the original $4,000 spend. That is a later view of the same buyers, not a contradiction of the early report. More of their outcomes are now known.

Keep an immature group marked pending. Do not make “not closed yet” mean “will never close.”

Separate revenue attribution from proof of causation

Following a group through time answers which records connect. It does not settle why a customer bought.

Suppose buyer 14 clicks the ChatGPT ad on March 10, attends a webinar on April 2, then searches for your brand and returns through a retargeting click in May. The contract arrives in August.

Attribution assigns credit among those touches according to a rule and a time window. It does not tell you whether the buyer would have found you without the ChatGPT ad. That second question is incrementality, which requires a comparison group that did not see the ad.

This is the distinction I want kept visible: revenue associated with an ad is not automatically revenue caused by that ad. The difference between reported ROAS and incremental lift matters even when every tracking event fires correctly.

So I would describe the example’s $48,000 as observed revenue from the March-origin group. I would not quietly turn it into proof that ChatGPT ads created $48,000 that otherwise would not exist.

There is another limit: the CRM’s history can outlast the platform’s attribution or import window. A deal can be traceable in the customer record without appearing as a credited sale in Ads Manager.

For a new advertising surface, verify the current attribution settings and available windows before designing the report around them. Also distinguish the window used to assign credit from the rules governing which later events an integration can accept. Those are separate checks.

The Google example makes the practical problem clear: an offline import window can expire before a long-cycle deal closes. Where supported, an earlier qualified stage may fit inside that window even when the eventual contract does not. The CRM still needs to retain the late close against the original lead.

For buyer 14, the August contract should not disappear from March’s business results merely because the platform cannot credit or import it. Nor should the report pretend the platform recorded a sale that it did not.

Verify the window, choose an eligible stage where imports are supported, and keep the full sales history in the CRM.

Paper chain linking a demo form to a signed contract, with one broken link in the middle

Make the agency own the connection, not just the campaign

With the event, identity, timeline, and attribution limits established, an agency’s responsibilities become much easier to specify.

Running the campaign and proving its contribution are related but separate jobs. Execution covers the ads, budgets, and ongoing checks. Evidence shows which bookings become held demos, which become pipeline, and which close.

Most agencies sell the first job and grade themselves on the cheapest version of the second. A screenshot of form submissions is easier to produce than a reconciled view of sales outcomes. That does not make it an adequate substitute.

I want three things in writing before spend starts:

  1. The qualification rule. For this example: a held demo, a company over 20 employees, and a named decision maker on the call. Not a form fill. Not a no-show.
  2. The reconciliation owner. Someone is responsible for checking platform events against website records and CRM contacts, explaining discrepancies, and fixing tracking errors.
  3. The reporting treatment. Open deals remain pending; closed revenue stays attached to its origin group; unresolved outcomes remain visible.

At groas, continuous campaign execution and measurement accountability belong in the same operating model, with a named senior strategist, logged actions, and a flat monthly fee rather than a percentage of spend. The point is not to charge for more hours of inspecting a dashboard. It is to connect the work to qualified pipeline and closed revenue.

If the agency cannot show how 22 recorded conversions became identifiable customer records, I do not yet have measurement I can use. I have a screenshot.

For the follow-through from attendance to pipeline and closed-won, the next layer is what happens after a ChatGPT demo booking.

Assign ownership of the evidence before buying the traffic.

Keep observed, pending, and unresolved outcomes separate

The August report for our March group should not collapse everything back into one flattering ratio. I would ask for three visible buckets:

  • Observed revenue: two signed contracts totaling $48,000, against $4,000 in ad spend.
  • Pending pipeline: three opportunities still open at $24,000 each, with any sales-weighted value shown separately from signed revenue.
  • Other or unresolved outcomes: the remaining 17 bookings, including no-shows, bad fits, and lost contacts. Separate known dispositions from records whose outcome is still unknown.

That accounts for all 22 bookings. The two opportunities visible at the first review were an early snapshot; the August report captures how the group developed afterward.

An open opportunity is not a signed contract. A confirmed bad fit is not a mystery. A lost contact is not proof of either future revenue or permanent failure. Blending those statuses makes the report shorter by making it less useful.

Pending is not failure, and unknown is not zero. Keep the distinctions intact so the next decision rests on what is known, not on whichever total looks best.

Test the next pitch with the same model

Now suppose a new agency pitches 40 demos for $20,000 next quarter. That is $500 per promised demo. I would not accept or reject the offer on that ratio alone. I would work through the chain:

  1. What counts as a demo? If it means a form submission rather than a held call with a decision maker, the pitch describes the first event, not qualification.
  2. How does it connect to the CRM? Ask who preserves the customer record, reconciles discrepancies, and fixes tracking when the platform and website disagree.
  3. What happens when the sale takes longer than the platform window? Ask which earlier stage can be recorded or imported, where supported, and how the eventual contract stays attached to the original lead.
  4. How will the report separate outcomes? Signed revenue, pending pipeline, known losses, and unresolved records should not disappear into one cost-per-lead figure.

Those answers determine what the $500 actually buys and how I will judge it later. If the agency hesitates on all the connecting work, I do not have a measurement plan. I have media spend with a screenshot attached.

That is the standard I hold my own work to. Execution without evidence is activity. Evidence without execution leaves a clean report and a dirty account. groas is built to own both sides, with continuous campaign work and a named strategist accountable for the connection to business outcomes.

Start with the three events. Connect them before trusting the ratio. Then the demo booking becomes what it should have been from the beginning: the beginning of the evidence.