How to allocate marketing budget across channels when sales close by phone or in person

Allocate a marketing budget on the next dollar, not the average one. Start from revenue matched to each channel's leads in your own sales, estimate each channel's marginal return from its last budget change, move money in steps of 10 to 20% a month, keep 10 to 20% for tests, and never hand everything to the best-looking channel.

Every budget review has the same table: channels down the side, spend and return across the top. The tempting move is to sort by return and shift money upward. It usually loses money, because the channel at the top of an average-return table is often the one that has already bought most of what it can, and its next dollar buys far less than its last.

This post is for businesses that close on the phone or in person: plumbing, roofing, dentists, law firms. It assumes you know what each channel returned; our post on how to calculate marketing ROI covers that number, and this one covers what to do with it.

Start from matched revenue per channel

The input to a budget decision is the revenue each channel actually produced: closed sales matched to its leads on phone or email, credited by a written rule such as the customer's earliest lead. Platform-reported conversions count calls and forms, or values attached to them, and for an offline business they rank channels by inquiries.

A platform cannot see which callers bought, or that the same person also called from a postcard, so two platforms will both claim a customer who touched both. Allocating on their numbers is allocating on cost per lead with a revenue label. Build the table from your own records instead: match sales to leads, as our post on marketing attribution for small business describes, group sales by the month of the lead, and use at least a quarter. If margins differ by type of work, convert to gross profit; a budget should chase profit, not revenue.

Average return vs marginal return

Average return is a channel's total revenue divided by its total spend. Marginal return is the extra revenue from one more dollar. Advertising has diminishing returns, so marginal return falls as spend rises and sits below the average. Budget decisions move marginal dollars, so marginal return is the one that should decide them.

The first dollars in any channel reach the best prospects: the people searching right now, the neighborhoods that buy most. Each extra dollar reaches someone a little less likely to buy, or the same people again. Eventually the channel saturates: there are only so many people searching for a plumber in your area this week, and bidding more mostly raises what you pay for each of them.

Estimating the next dollar without a model

A marketing mix model needs years of data most local businesses lack, but its logic needs one number per channel. Find the last time the channel's budget changed and held. Marginal return ≈ change in matched revenue ÷ change in spend, comparing lead months before and after. If search went from $5,000 to $6,000 a month and its matched revenue from $32,000 to $36,000, the extra $1,000 returned 4.0, against an average of $36,000 ÷ $6,000 = 6.0.

The estimate is rough. The season may have changed between the two periods, so compare with the same months last year where you can, and check that channels you did not touch held steady. A difference of one or two jobs a month is noise. A direction is all a budget step needs.

Minimum spend for readable numbers

A channel needs enough closed sales in the period you judge it on for the number to mean anything. A count of sales wobbles by about its square root from chance alone: 4 sales could easily have been 2 or 6. Below about 20 matched sales, judge over a longer window or not at all.

Minimum spend = sales needed × cost per customer ÷ months you will wait. If a channel's customers cost about $250 each, as our post on customer acquisition cost by channel shows how to work out, 20 sales in three months needs 20 × $250 ÷ 3 ≈ $1,670 a month. At $500 a month you will have six sales and an opinion, not a result. A channel that cannot be funded to its floor should get a defined test or nothing; a small, permanent, unreadable line is the commonest waste in a local marketing plan.

Keep a test budget for new channels

A test budget is a fixed share of spend, commonly 10 to 20%, set aside for unproven channels. It has its own rules: a set period, enough spend to reach the minimum, and a pass mark written down in advance that keeps or ends the channel. Without one, the budget can only shrink toward what already works.

Every proven channel was a test once, and every proven channel saturates eventually. Ring-fence the money, because a new channel always looks worst in its first month: its costs land at once and its sales close later.

• One or two tests at a time. $2,000 a month is one readable test or four unreadable ones.
• A fixed period, two to six months depending on volume, plus the time leads take to close.
• A pass mark, such as a return above break-even, 1 ÷ gross margin: 2.5 at a 40% margin.
• Recorded leads. A test whose calls reach an untracked number lands in Direct / Unknown and fails for the wrong reason.

Move money in steps, because of time lag

Move budget in steps of roughly 10 to 20% of a channel's spend a month, and let the changed month's leads close before the next step. Sales follow spend by days or weeks, so a large one-time move cannot be read until it is too late to undo, and several moves at once cannot be told apart.

A step also measures. It is a before-and-after comparison, so it updates the marginal estimate it was based on. Cut a channel by 15% and if its matched revenue barely moves, that money was buying little; if revenue drops sharply, reverse a small move cheaply. Pace steps to the median days from first lead to sale in your matched pairs: monthly if jobs close in a week, every other month if they take six. For the one channel you most doubt, our post on incrementality tests covers a deliberate holdout.

Seasons change the curves. In a peak more people are looking, so a channel absorbs more before it saturates. Plan the seasonal swing as its own decision from last year's matched revenue by month, make reallocation steps inside stable stretches, and never read a step taken as the season turns; our post on seasonal attribution covers comparing like with like.

Brand vs demand capture

Demand capture channels, such as search ads and Local Services Ads, reach people already looking. Demand creation channels, such as direct mail, social ads and radio, make people look. Attribution credits capture generously, because that is where the lead gets recorded, so cutting creation channels on matched returns alone can shrink capture channels months later.

A homeowner sees two postcards, searches your name a month later and calls from the ad; the sale is credited to search. Branded search is the extreme case: the best return in the account, and almost no room to grow, because bidding does not make more people search your name. Report it apart from non-brand, as our post on branded search explains.

A local plumber is not a national brand, and most of its spend will rightly be capture. What transfers is narrower: keep at least one creation channel funded steadily, cut it only in steps, and watch search and Direct / Unknown in the months after.

What to do with the Direct / Unknown share

Direct / Unknown is revenue that matched no recorded lead: referrals, repeat customers, walk-ins and leads nobody logged. Do not share it out among channels or budget against it. Use it as a check: if it falls after you cut a channel, that channel was producing sales your records missed.

Spreading it in proportion lifts every channel by the same factor, so it changes no ranking and no marginal estimate; it only flatters the totals, the same trap our post on blended ROAS describes. What it does flag is undercounting: mail carrying the main office number, or a vehicle wrap, produces calls matching cannot see, so their measured return is a floor. Fix the records before cutting them; our post on unattributed revenue covers how.

A worked example: five channels, $20,000 a month

A plumbing and water-heater company spends $20,000 a month across five channels. Revenue is matched to leads by lead month, averaged over July to September 2026; jobs average about $1,500 and gross margin is 40%, so break-even ROAS is 2.5. The figures are illustrative. The channel with the best average return should get less, not more.

The arithmetic. Average ROAS is matched revenue ÷ spend: $27,000 ÷ $3,000 = 9.0, $36,000 ÷ $6,000 = 6.0, $24,000 ÷ $5,000 = 4.8, $14,000 ÷ $4,000 = 3.5, $3,000 ÷ $2,000 = 1.5. Marginal ROAS is revenue change ÷ spend change: $1,200 ÷ $500 = 2.4, $4,000 ÷ $1,000 = 4.0, $4,500 ÷ $1,000 = 4.5, $2,000 ÷ $1,000 = 2.0. Matched revenue is $104,000 of $150,000, 69%, for a matched ROAS of 5.2.

The obvious move: give the best channel everything

Move Facebook's $4,000 and Nextdoor's $2,000 into Local Services Ads. On averages: $6,000 × 9.0 = $54,000 gained, minus the $17,000 the two produced, is $37,000 more a month for the same spend. On the marginal figure: the last $500 there brought in $1,200, under one job. Even if the next $6,000 did as well, $6,000 × 2.4 = $14,400 minus $17,000 is $2,600 less a month, and returns will keep falling. Pay-per-lead ads may not even spend it once local searchers run out. And the test is gone.

The better move: from low marginal to high, in steps

Rank by marginal ROAS: mail 4.5, search 4.0, Local Services Ads 2.4, Facebook 2.0. The last two are below the 2.5 break-even, so their last dollars lost money on margin despite healthy averages. The first step, for November leads:

The arithmetic. Each expected change is spend change × marginal ROAS: −$500 × 2.4 = −$1,200, $500 × 4.0 = +$2,000, $600 × 4.5 = +$2,700, −$600 × 2.0 = −$1,200. Net +$2,300 of revenue a month, $920 of gross profit at 40%, for the same $20,000. The Local Services cut reverses its last step, the safest cut there is: you have already seen what that money bought.

That is modest, and honest. The $37,000 was built on an average; the $2,300 on what the last change actually did, and its gains are upper bounds, because mail and search saturate too. In late December, once November's leads have closed, the company reads the step and either takes another or puts the money back. Steps stop when marginal returns are roughly level, because then no move improves the total. Nextdoor keeps its $2,000 until its six-month window closes; at two jobs a month, three months was too thin to judge.

CloseRev produces the matched-revenue column, not the budget. It reads two CSV files you export, closed sales and leads, matches each sale to leads on normalized phone and email, and credits it to the source of the earliest matched lead, with sources named exactly as written in the lead file. Only high-confidence matches count automatically; medium- and low-confidence ones stay reviewable, and unmatched sales stay in Direct / Unknown. If you enter monthly spend per channel, the report shows ROAS for those channels. It does not know your margins or other costs and does not recommend budgets; the marginal estimates and the steps are yours.

A monthly routine for allocating the budget

Once a month, rebuild the matched-revenue table by lead month, update each channel's marginal estimate from its last change, move one step from the lowest marginal return to the highest, and leave the test alone until its window ends. The first table takes an afternoon; after that, about an hour.

1. Match closed sales to leads on phone and email, credit the earliest lead, and keep unmatched sales in Direct / Unknown.
2. Total matched revenue or gross profit by channel and lead month, leaving out months still closing.
3. Estimate each channel's marginal return from its last budget change, and move 10 to 20% from the lowest toward the highest.
4. Write down every change and its date, so next month's estimate knows what moved.

The average return on your report says where money worked; the marginal return says where the next dollar will. Measure revenue in your own closed sales, estimate the next dollar from the last change, move money in steps you can read, and keep paying for tests. The best-looking channel is usually the fullest one.