An agency sent a lead forecast: how to tell whether to trust it

Check the links, not the bottom line

So you check the links, not the total.

The bottom line of a forecast is a product of its links, so an error in any link carries through. If two links are each overstated by 1.5 times, the result is overstated by 2.25 times. So you check the links, not the total.

No platform promises its forecast will come true. Google calls all Reach Planner values estimates that do not guarantee future performance; Meta says estimated audience size is not the number of accounts that will see your ads. An honest forecast is a calculation with named assumptions and a range, and an honest agreement is a test with checkpoints.

Disclosure. The publisher, WEBNOSTRA CAPITAL, is connected to WEBNOSTRA, a marketing services company that prepares forecasts like these for its clients. The same questions and rework apply to any vendor, including us, and to your in-house team. This article carries no link to a service.

The seven links of a forecast

Search forecasts usually start from clicks, social and display from impressions or reach. Reach counts people; impressions count every showing, so at a frequency of 3 each person saw the ad three times.

The last two links are often missing, yet they turn leads into money. Lead quality and conversion to sale are covered in separate articles.

The platform knows the auction, not your site or your sales team

Keyword Planner shows two kinds of numbers. Historical metrics default to a 12-month average and ignore bid, budget and seasonality; forecasts include them and are refreshed daily from the last 7-10 days. The top-of-page bid range is roughly the 20th and 80th percentile of past bids, a ready-made range for cost per click.

Performance Planner needs history: for Search, at least one conversion, 3 clicks and 10 impressions, spend in the last 17 days and no bid strategy change in 10 days.

In Ads Manager, Meta's bid estimate draws on the ad account's own history, so two accounts can see different numbers for the same targeting; actual reach depends on budget and performance.

Site conversion, qualified-lead share and lead-to-deal rate are either agency assumptions or your data. Without the advertiser's own conversion rate, Reach Planner uses a default based on inventory-wide medians.

Teaching example: how 120 leads become 38

Conditions of the teaching example. A company installs warehouse security cameras and advertises on Google Search. The agency forecast for month one: 40,000 impressions, CTR 5%, 2,000 clicks at $3, a $6,000 budget, 6% conversion, so 120 leads at $50. The numbers are made up; the node calculation is in the claim ledger.

Now add the missing links: sales qualify 60% of leads, and 20% of those become deals. The 38 leads give about 23 qualified leads, four to five deals and about $1,307 of ad spend per deal. The agency forecast with the same shares gives about 14 deals at $417. A more than threefold gap, yet the agency invented none of its numbers: each was simply taken from the wrong source.

This is ad spend per deal, not full acquisition cost, which adds team time, tools and the vendor's fee.

1. Seasonality. Impressions came from a 12-month average; the launch month runs at 80% of it. That gives 1,600 clicks, $4,800 spent and 96 leads: you cannot buy more clicks than there are searches.
2. Cost per click. Keyword Planner forecasts about $4, not $3. Now the budget is the limit: $6,000 buys 1,500 clicks, 90 leads at $66.67.
3. Conversion. The 6% came from another niche; the company's site converted 3% of paid search visits over three months. 45 leads at $133.33.
4. Learning. Assume leads per click are 30% lower in the first two of four weeks, our teaching assumption, not a platform figure. About 38 leads at $156.86.

Four traps that add up to a multiple-fold gap

Each trap looks minor alone; together they produce a multiple-fold gap.

Check the volume too. About 38 leads a month is roughly nine a week, well short of Meta's guide of about 50 results a week. Meta shows Learning limited when an ad set lacks results, so the agency would need a more frequent event or a slower, noisier read.

• A conversion rate from another niche or a market average: without your own history, the biggest unknown.
• No learning time. Google says Smart Bidding Learning status never lasts more than 7 days, but judge performance over 2 conversion cycles or 2 weeks, whichever is longer. Meta says an ad set usually exits learning after about 50 results in the week after its last significant edit, and costs more per result until then.
• Seasonality hidden in annual averages: compare the launch month with the keyword trend and last year in your account.

Decide on the cautious scenario, not the base case

If only the favourable one fits, you are paying for hope.

Ask for three scenarios, each link with its own value and source, and decide on the cautious one. If even that fits the cost per lead you can afford, the test is safe. If only the favourable one fits, you are paying for hope. A separate article shows how to derive it.

Even the favourable scenario yields fewer deals (about 10) than the agency forecast with the same shares (about 14): the forecast sits outside a reasonable range until sources support it. At two to ten deals a month, one deal moves spend per deal by tens of percent, so one month is weak ground for a verdict.

Seven questions to send the agency before launch

Send them in writing and ask for the answers in the same spreadsheet as the forecast. A well-prepared vendor will answer most straight away.

• Which tool and date produced impressions, CTR and cost per click? Attach the export.
• Is site conversion from our analytics or another project? Which period and traffic?
• What counts as a lead: form, call, message? How are duplicates and spam removed?
• What qualified-lead share and lead-to-deal rate do you assume, and why?
• How long will bidding learn, how many conversions a week does it need, and does the budget deliver them?
• What are the cautious and favourable scenarios, and which link moves the result most?
• What is outside the budget: taxes, agency fee, ad production, site changes?

Write the checkpoints down before launch, not after

Put the checkpoints in the contract or an email: metric, range edge and action if a number falls outside. Timing fits a sales cycle of about a month; use your own.

During learning, avoid needless structure or budget changes: Meta warns significant edits restart learning and advises against frequent budget changes, and Google judges Smart Bidding over 2 conversion cycles or 2 weeks.

1. Before launch: ad leads reach the CRM with their source, and analytics goals match what you count as a lead.
2. End of week one: impressions, clicks, cost per click and spend within the scenarios. Do not judge cost per lead yet.
3. Weeks two to three: cost per click and conversion against the range. A cost per lead above $237.50, the cautious edge, means examining the link that moved, not changing every setting.
4. End of month: qualified-lead share as judged by sales, the first look at quality.
5. A full sales cycle after the first leads: deals and ad spend per deal against what you can afford. Continue, scale or stop.

What this rework will not show

The rework finds weak assumptions; it does not out-predict the platform. It treats links as independent averages, though they are linked, and the learning adjustment is an editorial assumption no platform publishes.

Learning times and thresholds come from Google and Meta help pages as of the check date and can change. We give no industry norms for CTR, conversion or cost per lead. The calculation does not guarantee sales.

Key decisions

1. accept a forecast only when every link has a source and a period, and take site conversion rate and qualified-lead share from your own history
2. ask for a range of three scenarios and compare the cautious one, not the base case, with the cost per lead you can afford
3. agree checkpoints before launch: what you check after learning, after the first month and after a full sales cycle

What to discuss with your team

• Every link in the forecast has a source, an export date and a period
• Site conversion rate comes from your own analytics or is shown as a wide range, and learning time and launch-month seasonality are accounted for
• The forecast includes qualified-lead share and lead-to-deal rate
• The cautious scenario has been compared with the cost per lead you can afford
• Checkpoints, range edges and actions are written down before launch

Sources

1. Google Ads Help: About Keyword Planner forecasts (forecasts vs historical metrics, bid ranges)support.google.com · checked 31/10/2026
2. Google Ads Help: Use Keyword Planner (forecast refresh)support.google.com · checked 31/10/2026
3. Google Ads Help: About Performance Planner (forecast data and eligibility)support.google.com · checked 31/10/2026
4. Google Ads Help: About Smart Bidding learningsupport.google.com · checked 31/10/2026
5. Google Ads Help: About Reach Planner (estimates, no guarantee)support.google.com · checked 31/10/2026
6. Google Ads Help: About forecasts in Reach Planner (default conversion rate, data window)support.google.com · checked 31/10/2026
7. Yandex Direct Help: Budget forecastyandex.ru · checked 31/10/2026
8. Yandex Advertising: Budget forecast in Yandex Direct, guide in Russian (updated 19 Feb 2025)b2b.yandex.ru · checked 31/10/2026
9. Yandex Direct Help: Campaign has few conversions (goal frequency and learning time)yandex.ru · checked 31/10/2026
10. Meta Business Help Center: About estimated audience sizefacebook.com · checked 31/10/2026
11. Meta Business Help Center: About the learning phasefacebook.com · checked 31/10/2026
12. Meta Marketing API: Estimated daily results, developer documentationdevelopers.facebook.com · checked 31/10/2026