Six weeks, $3,000 in media, one SaaS product, and one person on your side to approve changes. I would spend the first two weeks fixing tracking and the demo-booking page, not running campaigns. At week six, success is a booked demo you can trace to ChatGPT Ads in your CRM. Clicks and somebody else’s case study do not count.

I learned the order from working with a SaaS client that changed its offer every quarter. When the offer changed, so did the meaning of a demo, and the earlier data stopped answering the question we thought we had asked. Make this a go/no-go test, including when the answer is no. Judge the manager by the evidence in your account at week six.

1. Set the constraint: $3,000 buys one answer

Cost against the constraint: $3,000 total media, spent in Weeks 3–6. Six weeks at $3,000 averages $500 a week, but that is not the spending schedule. The first two weeks get $0; the four live weeks average $750. That leaves enough time to fix measurement before the first paid click and enough budget to give a narrow test a chance.

ChatGPT ads in self-serve start at $25 a day in the US. Reported CPCs run $3 to $9, with software often $8 to $18. Use $10 a click as planning math, not a promise: $3,000 would buy roughly 300 visits. At a 3–5% booking rate, that would mean about 9–15 booked demos. Actual CPC and booking rate decide the result. The point of the estimate is to show how little room there is for five simultaneous experiments. This is a go/no-go budget, much like the $3,000–$5,000 tests SaaS teams run over four to six weeks.

Cut on day one: no CPM awareness test, no ten ad groups, no homepage traffic campaign to ‘learn.’ Buy clicks, test at most three prompt themes, and give each theme a relevant demo page. I used to tell clients broad learning would pay off later. With this budget, breadth mostly buys smaller samples. Depth buys a decision.

Stopwatch on dollar bills beside a calendar marked week six

2. Before Week 1: define a demo and its owner

Cost against the constraint: $0 media and four hours of your approver’s time before the six-week clock starts. What it buys is a definition sales and marketing can both use. Write it down: a booked demo is a calendar hold from the demo page with a business email and a company size above your minimum. A held demo means the person showed up. An opportunity means sales accepted it. Those are three different events, however neatly a dashboard stacks them.

The B2B demand-generation hierarchy puts meetings held above meetings booked, and both above leads. Use held demos as the primary week-six outcome, booked demos as the leading signal, and opportunities as evidence of quality. Do not ask a six-week test to prove closed revenue when the cited SaaS sales-cycle benchmark is 84 days; first-touch and pipeline-attributed Search ROAS can tell very different stories.

Create a ChatGPT Ads source value in the CRM before a form arrives, and make sure every demo form carries it through to the booking record. On Search, the familiar plumbing is a click ID stored with the form and mapped to the CRM’s lead and opportunity records. Use the tracking and offline-event options your ChatGPT ads account actually provides; do not make a platform upload the only place a demo exists. In your CRM, record form accepted, demo held, and opportunity created as separate events.

Cut: any reporting plan that ends at form fills. If sales cannot filter held demos by source, fix that before Week 1.

3. Weeks 1–2: fix tracking and the demo pages, then test both

Cost against the constraint: $0 media and about ten hours of development and marketing time. What it buys is a route from paid visit to CRM record. Set up the OpenAI Pixel and server-side Conversions API if those tools are available to your account. Where you use both, send matching event IDs to avoid counting one booking twice. Keep the CRM source and booking record as the week-six authority, whether or not an ad-platform event fires.

Then make a real test booking. Check the page, calendar, form, and CRM record together. If the booking does not appear with source ChatGPT Ads within an hour, do not launch. A platform conversion count will not rescue a lead your sales team cannot find.

Build the pages around buyer questions, not your navigation menu. Most SaaS teams can start from two page patterns:

  • The evaluator’s page: how the product works with the buyer’s stack.
  • The owner’s page: what it costs and how quickly the buyer can see value.

Give each planned theme its own version of the relevant page. Each version needs a headline that names the buyer’s situation, three bullets that answer it, a calendar embed above the fold, and no link sending the visitor back to the homepage. That is not an invitation to produce three elaborate websites. It is a way to stop asking a paid visitor to sort through four unrelated calls to action.

I have watched SaaS teams send paid clicks to a homepage with four CTAs and call the channel weak. The visitor was asked to choose the next step when the ad should have chosen it. Cut homepage traffic and cosmetic page experiments. End Week 2 only when a test visitor can book and appear in the CRM under the right source. No working booking path, no Week 3 ads.

4. Week 3: launch three prompt themes, not a keyword list

Cost against the constraint: up to $490 media at a $70 daily cap, plus three hours of management. What it buys is the first paid read on buyer intent. The unused $10 stays in the six-week budget; there is no reason to break the cap to make a spreadsheet round to $500.

ChatGPT ads have no keywords or demographic targets. Matching draws on conversation context, intent, ad and landing-page relevance, and the Context Hints you set at ad-group level. Hints guide delivery; they are not exact-match keywords or guaranteed placements. That is the mechanism Google advertisers tend to miss. Write the buyer’s moment, not a category label.

Launch one ad group for each of these three themes, with its corresponding demo-page version:

  1. A buyer comparing tools for a named workflow.
  2. A buyer stuck on an integration or migration question.
  3. A buyer asking about cost, implementation time, or a security review.

Use five to eight one-sentence hints per group. A hint might describe someone asking how to sync HubSpot demo outcomes back to ad spend. Start near the suggested $3–$5 max CPC for CPC buying, then watch whether the account can deliver visits within your cap. Ads appear below answers for logged-in adults on Free and Go plans, so do not plan around Search-sized volume.

Require approval before any increase to the $70 account-wide daily cap. Cut the fourth theme, the CPM test, and expanded hints until a theme has booked two demos. If the account cannot spend its cap after three days, check whether the hints match the pages before reaching for a higher bid. The first job is to find a buying moment, not to force delivery.

Paper-cut diagram of three narrow prompt paths leading to a demo calendar

5. Week 4: cut a weak theme and move its budget

Cost against the constraint: about $750 media and two hours of review. Approve a higher account cap of roughly $108 a day if Week 3 produced a working booking path and spendable traffic. What this week buys is the chance to stop paying for curiosity and concentrate on a theme that can book a meeting.

Pull the same three numbers for each theme: spend, visits, and CRM-booked demos. Check held demos as they come in. Ignore CTR when deciding which theme stays. Visit volume tells you whether the hints are finding conversations; booking rate tells you whether those conversations and the page fit each other. For context, the cited SaaS benchmarks put Search click-to-demo or trial at 3–5%, versus 0.64% for Display. They are a comparison point, not a target ChatGPT has already earned.

If a theme spends $250 with a booked-demo rate below 1%, pause it and move its allocation to the strongest theme. That is an early cut, not a statistical verdict on every buyer in the market. At this budget, you cannot keep three underperformers alive in the name of fairness. I have made that mistake. The extra clicks made the report longer, not the decision better.

Cut: new themes, bid increases made solely to force volume, and a CPM side quest. Keep spend with the group producing held demos rather than cheap visits. It is the same discipline behind moving B2B SaaS Google Ads decisions from leads toward pipeline signals. Enter Week 5 with the weakest theme paused, unless the CRM gives you a clear reason not to.

6. Week 5: freeze the offer while you spend another $750

Cost against the constraint: about $750 media and one decision meeting. With roughly $1,000 left for Week 6, this week buys comparability. Keep the approved cap near $108 a day, adjusting only as needed to stay inside the total budget.

This is where the urge to ‘improve the offer’ gets expensive. I have worked with a SaaS client that changed its offer every quarter. Once the offer changed, earlier bookings no longer measured the same proposition. In this test, the final two live weeks are too short to run a new offer and pretend it belongs in the same readout. Lock the offer, demo-page promise, and form fields. If sales wants a different headline, put it on the Week 7 list.

You may move budget between the surviving themes, tighten a hint that attracts stray conversations, or fix a broken calendar slot. Cut the new page, the pricing test, and the fourth theme meant to chase volume. Holding the conversion definition steady is what lets offline conversion records connect paid visits to pipeline. A clever new offer is not useful if it erases the question the first $2,000 asked.

7. Week 6: make the go/no-go call in your CRM

Cost against the constraint: the remaining media, about $1,010 if earlier weeks hit plan, and one readout with sales. Approve a cap of roughly $145 a day only if the remaining budget and booking evidence warrant it. Total media spend must not exceed $3,000. The larger final-week allocation goes to the surviving theme or themes, not a fresh experiment.

Filter the CRM to source ChatGPT Ads. Count booked demos, held demos, and sales-accepted opportunities separately. Divide actual spend by held demos for cost per held demo, then put that number beside your own non-branded Search result. The cited SaaS Search benchmark is about $5.34 a click with a 3–5% click-to-demo rate; it describes booked-demo economics before no-shows, not your cost per held demo. Compare like with like. If this test delivers eight held demos at $3,000, the cost is $375 per held demo. Check whether those meetings became opportunities, and keep AI-sourced distinct from AI-influenced pipeline. Do not let an attribution argument hide the meetings sales actually held.

Make the call against your Search number:

  • Kill if you have fewer than two held demos after at least 250 visits, or cost per held demo is three times Search with no opportunity created.
  • Hold if you have three to six held demos at 1.5–2 times Search cost and one theme is clearly best. Fix that theme’s page and hints, then retest with $1,500.
  • Scale if cost per held demo is within 1.3 times Search and the show rate matches Search. Move budget from your weakest Search ad group rather than asking for new money; keep approval on cap increases.

The decision follows the CRM, not the platform’s lead count. That is the point of bidding toward closed revenue rather than leads as pipeline matures. For this six-week test, held demos are the decision metric; closed revenue can inform what happens later. A no is a result. It is better to learn it for $3,000 than to keep buying clicks because the dashboard looks busy.

8. Demand a week-six readout, not somebody else’s case study

Cost against the constraint: $0 additional media and one manager conversation. Ask for five things:

  1. A CRM export filtered to ChatGPT Ads, with booked and held demos marked.
  2. Spend by theme and week, reconciled to the $3,000 ceiling.
  3. Booking-rate, show-rate, and cost-per-held-demo math.
  4. A dated list of what was paused, with reasons.
  5. Proof that test bookings reached the CRM, plus Pixel and Conversions API diagnostics where those tools were available.

If the readout opens with impressions, CTR, or another company’s logo slide, ask for your list of meetings. A manager should be able to name the winning theme and show what got cut to fund it. What the deck calls optimization, I call a list of decisions with dates.

When the constraint lifts, do not celebrate by buying more themes. Put the winner into execution that runs every day, not a review every Friday. That is why I point SaaS teams toward Search Ads run for you: groas runs Google, Bing, and ChatGPT ads for a $1,499 flat monthly fee, with bids and budgets worked around the clock, a landing page for every ad group, and a named manager who owns the result. Keep approval on cap increases. Keep held demos at the center of the decision. Let the next $3,000 chase the one theme your own CRM says can book them.