Paid Media Budget Allocation Playbook

Split a paid budget so that the next customer costs about the same on every channel you fund, and no more than your CAC target. That point, not the cheapest channel, is where the same money buys the most customers.

Getting there takes a CAC target worked back from lifetime value, a cost per customer for each channel from your own click, lead and close rates, a curve for how fast each channel gets dearer, and enough budget in each for its bidding to learn. The free Paid Media Budget Allocator does the arithmetic; this playbook explains it, so you can check the answer and defend it. It starts once you know which channels might work: choosing and testing them is in the Demand Gen & Performance Marketing Playbook, and building the campaigns in the Digital Advertising Playbook.

In my view, the most common mistake in paid media is treating the budget as the input and the results as whatever falls out. Reverse it: the input is your customer acquisition cost (CAC) target, and lifetime value (LTV) sets it.

The widely used rule is an LTV:CAC ratio of at least 3:1: every dollar spent winning a customer should come back at least three times in gross-margin lifetime value. David Skok, who popularized it, puts the best SaaS businesses above 3 and some as high as 7 or 8^[1]. Work backwards from there: a customer worth $15,000 in gross-margin LTV allows a CAC of up to $5,000.

Two cautions. First, that $5,000 is fully loaded: it has to cover sales salaries, tools, content and overhead as well as ads. Subtract the rest, and what is left is your paid-media CAC target, the ad spend you can afford per customer, which is much smaller. Second, LTV:CAC is a ceiling check, not a growth strategy.

Paid channels are not interchangeable buckets you pour budget into. Each captures a different part of demand, at a different price, with a different lead quality. Google, Meta and Amazon are forecast to take 62.3% of the world’s digital ad spending in 2026, with Meta passing Google for the first time^[2], but the right mix for you almost never matches the world’s. What each channel is actually good at:

Search captures demand that already exists. Someone typing “contract management software” is in the market now, and Google is the biggest way to be in front of them. Across 13,474 US search campaigns from April 2025 to March 2026, WordStream’s median cost per lead was $66.69, and $93.69 for business services^[3]. Microsoft Advertising runs the same model on Bing, imports campaigns directly from Google Ads, and says its usually lower competition can mean lower costs^[4],[5]. Search converts well but caps out at the number of people searching.

LinkedIn reaches a named buyer by job title, company and seniority: demand you create rather than harvest. Its clicks are dear, €5.98 among Dreamdata’s customers against €1.60 on Meta, but the deals are larger. Dreamdata’s 2026 benchmarks, built on its customers’ B2B buyer journeys, report a return on ad spend of 121% for LinkedIn, against 67% for Google Search and 51% for Meta, and say LinkedIn takes 41% of total B2B ad budgets^[6].

Meta and TikTok create demand at scale. Meta (Facebook and Instagram) is cheap per impression and unmatched for volume, but intent is lower, so leads usually need nurturing. TikTok reaches younger, consumer-leaning audiences. Both are strongest for awareness and retargeting.

YouTube builds recognition and explains a product visually, so it suits anything that benefits from a demonstration: software, courses, considered purchases. It is often run as the brand layer for buyers not yet in the market.

Amazon meets buyers at the point of purchase. For physical products sold there, it is high intent and closes the attribution loop: the sponsored click and the purchase are measured on one platform. For anything else it is irrelevant.

Directories and communities are niche and high-context. Capterra and G2 reach software buyers who are comparing products; the common questions at the end cover what they cost. Reddit reaches technical audiences in their working context, Quora catches research questions and Pinterest drives discovery for consumer brands. None scale like the majors.

You should not run all of these. Start with the handful of channels where your buyer researches and decides, and that set differs by business model: Figure 2 shows the default lineup the allocator opens with for four of them.

The principle underneath: match the channel to where the buying decision happens. A high-value B2B product belongs where you can target a named title and explain value (LinkedIn, search, video); a $40 consumer product belongs where attention and purchase intent are cheap and immediate (Meta, TikTok, Amazon, Google Shopping). Running LinkedIn for a $40 product, or Amazon for enterprise software, wastes money however well you allocate it.

Benchmarks are for calibration when you start cold, such as the WordStream medians in Step 2. Once you have your own data, it wins: your offer, audience and creative move these numbers more than any industry average. For each channel you plan to run, pull four inputs:

• Cost per click (CPC): from the ad platform, or a benchmark to start. For a pay-per-lead channel, use the cost per lead instead.
• Click-to-lead rate: the share of clicks that become a lead, such as a form fill, demo request or sign-up, or for ecommerce an add-to-cart.
• Lead-to-customer rate: the share of those leads that become paying customers. This is where channel quality shows.
• Value: average contract or order value, gross margin and gross-margin lifetime value (Step 1).

Comparing cost per lead misleads, because cheaper leads often close worse. The honest unit is cost per customer, and it takes two divisions:

Cost per lead = CPC ÷ click-to-lead rate. Cost per customer (base CAC) = cost per lead ÷ lead-to-customer rate. For a pay-per-lead channel, skip the first division.

Repeat for every channel and you have a base CAC for each: the cost per customer at a modest level of spend. That ranking is the backbone of the allocation, and Step 4 is why it is only the start.

How to estimate what a keyword will cost before you spend

Step 3 needs a cost per click (CPC) for every channel, and before you have run a single ad you do not have one. AAJ does not publish its own CPC averages. What you can do is estimate your own CPC in four moves: read the platform’s bid range, work out the most you can afford per click, buy a small test to get the real number, and read who you were bidding against. The method is written for search, where you bid on keywords. LinkedIn and Meta price clicks differently, and the last part of this section covers them.

Why do CPCs for AI keywords vary so much? On Google, the CPC for any keyword, including an AI keyword, is not a list price. Every search runs its own auction, and Google runs separate auctions for each ad location on the page^[8]. Your position comes from Ad Rank, which Google builds from six factors: your bid, the quality of your ad and landing page, your Ad Rank thresholds, how competitive the auction is, the context of the search (the words, location, device and time) and the expected impact of your ad assets^[8],[9]. What you pay depends on the competition below you: Google says you only pay “what’s minimally required to clear the Ad Rank thresholds” and beat the Ad Rank of the competitor directly beneath you, so actual CPC is often less than your maximum bid^[10]. Microsoft Advertising also prices clicks by auction, and says “the ad auction is dynamic and continuous, so CPC changes constantly”^[11].

Read Keyword Planner’s range as a ceiling, not a price. Google Keyword Planner gives two bid figures for each keyword. Google defines “Top of page bid (low range)” as approximately the 20th percentile of what advertisers have historically paid to show at the top of the page, and “Top of page bid (high range)” as approximately the 80th percentile, for your location and network settings, from the last 30 days^[12]. Three cautions follow from Google’s own notes. These are bids, not charges, and actual CPC is often lower than the bid^[10]. Some keywords with limited bid history do not show a range^[12]. And a new account gets forecasts built on average data for all advertisers until it has data of its own^[12]. You need a Google Ads account with billing details entered before you can use the keyword ideas^[13].

Microsoft Advertising has its own Keyword Planner, under Tools, which gives search volume, trends, performance and cost estimates once you have an account^[14]. Check the same keywords there, because Microsoft prices clicks in its own auction.

Work out the most you can afford per click. You can turn your CAC target into a CPC ceiling before you spend anything. Step 3 multiplies the CPC up to a cost per customer. Run the same arithmetic backwards:

Max affordable CPC = paid-media CAC target × click-to-lead rate × lead-to-customer rate.

Any keyword whose real CPC sits above that line costs more per customer than your CAC target allows, however good its leads look. Step 4 explains why the next customer costs more than the average one as spend grows.

The ceiling is only as firm as the two rates inside it, and until you have measured them both are assumptions. The click-to-lead rate comes from a test. The lead-to-customer rate should come from your own sales history, even a handful of deals. Figure 5 shows how far the ceiling moves when either is off. At a 5% click-to-lead rate and an 8% close rate, Example Co.’s ceiling falls to $3.00 a click, below the low end of the planner’s range for “AI contract review”; at 12% and 15% it rises to $13.50. Google defines the low end of the range as approximately the 20th percentile of what advertisers have paid, so a ceiling below it means most advertisers who showed at the top of the page paid more than Example Co. can afford at today’s funnel.

The Unit Economics Calculator gives you the CAC target; the Paid Media Budget Allocator takes the CPC you end up with.

When Keyword Planner shows no range. A keyword with too little history shows no range, as Example Co.’s “AI contract review software for legal teams” does. That is a reason to price it with a test, not to skip it. Three steps keep the test contained. Look up a broader term that does show a range, and use it only to see whether your ceiling is in the same neighborhood. Give the new term its own ad group and test budget, so its clicks are not hidden inside the average of a bigger term. Cap the spend with the daily budget: a $2,700 test over 30 days is $90 a day, and because Google may spend up to twice the average daily budget on a given day but no more than 30.4 times it in a month^[15], the month’s billed spend cannot pass $2,736.

Buy a small test to get your real CPC. You get a real CPC by buying clicks. Size the test by the decision it has to support, not by a round number. The decision here is whether the keyword’s click-to-lead rate is close to your assumption, and that needs enough leads to read. This test uses 30 leads, the conversion count Step 6 cites from Google and Microsoft before their automated bidding is judged, so a keyword that works also gives the bidding something to learn from. Whether 30 leads is enough to read the click-to-lead rate is shown below: it leaves a wide interval.

Clicks needed = leads needed ÷ expected click-to-lead rate. Test budget = clicks needed × max affordable CPC.

Using the max affordable CPC keeps the test cheap. If clicks cost more than that, the budget buys fewer clicks, but by then you already have your answer: each customer costs more than your target allows at today’s funnel. Test one keyword theme at a time. Spread across several themes, each one gets a fraction of the clicks and so fewer leads to read.

Log the test before it starts, with its budget, read date and the CPC that would stop it. The Paid Media Test Log has the columns.

Read the result against your ceiling. Put the CPC you paid and the click-to-lead rate you measured into one line: cost per customer = CPC ÷ (click-to-lead rate × lead-to-customer rate). It is Step 3’s two divisions again, and it stays under your CAC target exactly when the CPC stays under your ceiling. A test reads CPC and click-to-lead within the month. It cannot read the close rate until the leads have had time to close (Step 6), so until then the 12% is still an assumption. Apply the same line to the ends of the click-to-lead interval, not only the single number. If cost per customer is under your target at both ends, the keyword passes this test. If it is over your target at both ends, drop it or test a cheaper phrase. If your target falls between the two ends, repeat the test with more clicks before you move budget.

Read the search terms and auction insights after the test. The search terms report shows the searches that triggered your ads, which can differ from your keyword list depending on your match types^[16]. Google leaves out terms used by too few people to meet its privacy standards^[16], so the report can leave some searches out. Add each search term that is not relevant to what you sell as a negative keyword^[16].

Auction insights shows who you were competing with on Search. It gives your impression share, how often another advertiser’s ad received an impression when yours did (overlap rate), how often theirs was shown in a higher position (position above rate) and how often your ad ranked higher than theirs (outranking share)^[17]. It needs a minimum of activity, and it does not show insights when your impression share is below 10%^[17].

What should I change after the first test? Four changes cut wasted clicks or tie the ad to the keyword:

• Use exact and phrase match while you learn. Google says exact match “gives you the most steering over who views your ad, but reaches fewer searches than both phrase and broad match”, and broad match can show your ad on related searches, including ones that “don’t contain the direct meaning of your keywords”^[18]. Google recommends pairing broad match with Smart Bidding, which learns from conversions; Step 7 of the Digital Advertising Playbook covers Smart Bidding, checking the conversion and AI Max’s broader matching.
• Keep a negative list. Start from the searches you can already name as irrelevant, as Example Co. does below with “free”, “template”, “jobs” and “course”, and add what your search terms report shows. Google says negative keywords account for casing and misspellings automatically, but you must add synonyms and singular or plural versions yourself^[19].
• Keep brand and non-brand apart. Put searches for your own name in a campaign of their own, so they do not blend into the non-brand numbers you are testing (Digital Advertising Playbook, Step 3).
• Make the landing page match the keyword. Quality Score rates expected click-through rate, ad relevance and landing page experience. Google calls it a diagnostic tool and says it “isn’t an input in the ad auction”^[20], but the quality of your ads and landing page is one of the six factors in Ad Rank^[9]. The Website & Conversion Optimization Playbook covers the page.

Why are LinkedIn and Meta priced differently? On LinkedIn and Meta the auction is for an audience, not a keyword, so the keyword steps above do not apply directly. LinkedIn says “the cost required to win the auction depends on the bid and the desirability of your target audience”^[21]. It offers maximum delivery, cost cap and manual bidding^[22], and it charges for the billable event your objective sets, by the click (CPC) or per thousand impressions (CPM)^[21],[23]. Meta’s billing events are “impressions, clicks, or various actions”, and Meta weighs your bid strategy, your bid amount and the probability of getting your optimization goal to set an effective bid^[24].

The method still works. Run a small test, then divide spend by clicks. When a campaign is billed by impressions, effective CPC = CPM ÷ (1,000 × click-through rate): as an illustration, a $60 CPM at a 0.5% click-through rate is $12 a click. Then compare it with your own ceiling. For Example Co.’s LinkedIn rates in Step 3, 11% of clicks to lead and 20% to customer, the ceiling is $750 × 11% × 20% = $16.50. That is why LinkedIn’s $11 clicks still give it the second-cheapest customers.

Mistakes that make a cost estimate wrong. Five follow from the steps above:

• Reading the planner’s high range as the price. It is a bid, not a charge.
• Testing several themes on one budget, so each gets too few leads to read.
• Counting searches for your own name in a non-brand test.
• Stopping a test at a handful of leads because the first days looked good or bad.
• Setting the ceiling from a close rate you have not yet seen, rather than from your own sales history.

Here is the trap the ranking sets: if one channel has the cheapest customers, why not put everything there? Because channels do not scale in a straight line. The first dollars reach your highest-intent, best-matched audience. As you add more, you exhaust that audience and the platform serves less-qualified impressions, so cost per customer climbs. Add enough and a channel’s customers can cost more than those of one that started dearer.

The pattern is well documented. A meta-analysis of 751 short-term estimates from 56 studies found that a 1% rise in a brand’s advertising raised its sales by 0.12% on average^[25]: each extra dollar buys less than the one before. The first dollar into a channel and the ten-thousandth do not buy the same customer.

So the unit for allocation is marginal CAC, the cost of the next customer at your current spend, not average CAC, total spend divided by all customers so far. The average still includes the cheap early customers, so a channel that is cheap on average can be expensive at the margin. The allocator models each channel with a saturation curve: the next customer starts at the channel’s base CAC and gets dearer as spend grows, faster on a channel with a smaller audience.

Now the method comes down to one rule: fund each channel until the cost of its next customer reaches your CAC target, then stop. Spend where the next customer is cheap; pull back where it would cost more than a customer is worth.

Apply that rule across your channels and two things happen. Channels whose very first customer already costs more than the others’ next one drop out, correctly. And the channels that stay get funded until the next customer costs about the same on each, at the target or, if the budget runs out first, below it. That point is not a compromise: it is the most customers your money can buy. If one channel’s next customer were cheaper than another’s, moving a dollar would gain a customer. Split Example Co.’s $30,000 evenly and Microsoft’s next customer costs $656 while YouTube’s costs $1,118. Move $1,000 from YouTube to Microsoft and YouTube loses about 0.9 customers while Microsoft gains about 1.5: the same money, about 0.6 more customers. Keep moving money until no move gains anything.

Doing that by hand means iterating across five or more channels. The Paid Media Budget Allocator does it: set a budget, a CAC target or a customer goal, and it returns the split, the projected customers and blended CAC, the saturation curves, and how much better the split is than spending evenly.

Here the split is worth about 8% more customers. Better conversion and close rates are usually worth more, because they lower what every customer costs (Step 8). Allocation squeezes more out of the spend you have; the funnel changes what a customer costs in the first place. Do both, in that order.

It is tempting to switch on every channel that fits. Resist it. Each platform’s automated bidding learns from conversions and needs a steady run of them; a channel funded below that floor underperforms its own math. What the platforms publish:

• Meta: an ad set generally needs about 50 optimization events within seven days of its last significant edit to leave the learning phase^[26].
• Google, including YouTube: Smart Bidding can need up to about 50 conversion events or three conversion cycles to calibrate^[27], and Google suggests judging Target CPA over 30 days that include at least 30 conversions^[28].
• Microsoft: it recommends at least 30 conversions in the previous 30 days for Target CPA and Maximize Conversions, and a learning period of one to two weeks^[29].
• LinkedIn: no conversion threshold; at least $10 a day, and $100 for a new campaign^[30]. It recommends starting with maximum delivery bidding and moving to a cost cap when you need to control cost per result^[22].
• TikTok: at least $50 a day per campaign and $20 per ad group^[31], and it calls 50 conversions “the most significant indicator of passing the learning phase”^[32].

So a channel’s floor is the number of conversions its bidding needs, times what one conversion costs you. Optimize for an event cheap enough to reach that number, usually a lead or a booked demo rather than a closed customer; the Demand Gen & Performance Marketing Playbook covers how to choose it. Spread $5,000 a month across seven channels and each gets about $700: at Example Co.’s costs, not enough for 30 leads on any of its five channels.

At seed, run one or two channels in earnest plus at most one small test, as which marketing channels to start with sets out. Once a channel works and the budget can fund more above their floors, three to five is the working range; treat any others as experiments with their own test budget, not slices of the core. The allocator warns when your budget cannot support the channels you have switched on.

An allocation is a monthly target, not a launch instruction. Give each channel a floor and a ceiling. The floor is at least its learning floor from Step 6, or higher for a strategic presence, which many B2B teams keep on LinkedIn even when search looks cheaper on the day. A useful ceiling is the spend at which the channel’s next customer reaches your CAC target (Step 5), so no channel runs past what its customers are worth before you have evidence it scales.

Then turn the monthly numbers into daily budgets, so spend lands evenly instead of running out mid-month. The platforms treat a daily budget as an average: a Google campaign can spend up to twice it on a given day but no more than 30.4 times it in a month^[15], and Microsoft usually keeps overspend below double the daily limit and caps the month at the daily budget times its days^[33]. LinkedIn recommends a lifetime budget with lifetime pacing^[30]. Divide each monthly figure by 30.4, and check spend against a straight line every week.

The fastest way to misallocate is to optimize toward cheap leads. A channel can win on cost per lead and lose on cost per customer, and a platform told to chase form fills will find you more of the wrong ones. Optimize for the event closest to revenue that still reaches the learning floor (Step 6), and send your CRM’s qualified outcomes back to the platforms so their bidding learns from buyers, not browsers; the Demand Gen & Performance Marketing Playbook covers the setup.

Time lag matters too, especially in B2B. In Dreamdata’s 2026 benchmarks, revenue arrived on average 214 days after a buyer’s first conversion on a LinkedIn ad^[6], so month-one CAC overstates the true cost. Read each month’s spend as a cohort: its cost per customer, or its return on ad spend if you sell online, at 30, 60 and 90 days and after a year. And watch CAC payback by cohort, the months a customer’s gross profit takes to repay what it cost to win; Skok puts the warning line for SaaS at 12 months^[1].

Your first split is a hypothesis. Auctions shift, audiences tire and creative wears out, so marginal CAC drifts. Review monthly: where a channel’s cost per customer has crept above the others, move the next slice of budget toward the cheaper margin; where spend brings leads but no customers, cut it and redeploy. But move in steps, not daily swings, because large changes restart the platforms’ learning. Google suggests allowing one or two conversion cycles after a major bidding or budget change^[34]; Microsoft asks you not to change budgets during learning and to judge a bid strategy only after two to four more weeks^[29]; TikTok advises raising a budget by no more than 30% at a time once an ad group is out of learning, at least two days apart^[31].

What changes by stage?

What mistakes should I avoid?

• Defaulting to the same two or three channels. Habit is not a strategy; start from where your buyer decides, which may be Microsoft, TikTok, Amazon or Capterra (Step 2).
• Running more channels than the budget can fund. Below its learning floor, no channel optimizes (Step 6).
• Judging channels on cost per lead. Cheap leads that never close are the dearest thing in paid media (Step 3).
• Concentrating in the cheapest channel. Its average hides what the next customer costs (Step 4).
• Judging month-one numbers. In B2B, most customers close weeks or months after the click (Step 8).
• Rebalancing daily. Large, frequent changes restart learning and trade signal for noise; move monthly, in steps (Step 8).