The Budget Was Never the Problem
How Cresco International built a decision-optimization engine that forecasts campaign revenue uplift for a leading Media company’s advertisers, before a single dollar is committed
Every media company sells the same thing twice. Once as inventory, and once as a promise about what that inventory will do. The first transaction is straightforward. The second one is where the relationship is actually won or lost — and for most of the industry, it is still settled by negotiation, precedent, and whoever argues most persuasively in the planning meeting.
When a leading Media company came to Cresco International, the question on the table was not how to sell more avails. It was harder and more valuable than that: can the allocation of a client’s working media budget across markets, channels, dayparts and flights be calculated rather than argued — and can the resulting revenue uplift be forecast before the campaign runs?
The answer is yes, and the discipline that produces it is decision optimization. Cresco has been building this class of system since 2012 — production scheduling, fleet routing, workforce coverage, inventory replenishment. OptimCampaign is that same engine pointed at the marketing budget. For a leading Media company’s enterprise advertisers, it has been returning forecast revenue uplift in excess of $7 million per campaign, on budgets that did not grow by a dollar.
Part I: Why a media portfolio is a genuinely hard allocation problem
This leading Media company does not operate a channel. It operates a portfolio: broadcast radio across hundreds of local markets, the largest podcast network in America, digital audio and streaming, outdoor, social extensions, and live events. A national advertiser buying across that footprint is not making one decision. It is making thousands of simultaneous, interacting decisions.
Consider a single national campaign for a quick-service restaurant client. The plan has to resolve:
- Market-level allocation across 160+ DMAs with wildly different competitive density, store counts, and audience composition
- Channel mix across broadcast, podcast, streaming audio, and outdoor, each with a different response profile and a different measurement lag
- Daypart and flighting decisions — morning drive versus evening, front-loaded versus sustained, and how much weight to hold back for a promotional window
- Reach and frequency targets that must be hit for the creative to work at all
- Real inventory capacity — avails are finite, sell-through varies by market and quarter, and a plan that ignores this is not a plan
- Contractual floors and commitments — minimum spend by market or property, agency-negotiated terms, make-good obligations
The industry standard method for resolving this is a combination of last year’s plan, platform auto-bidding inside each channel silo, and a media planner’s hard-won intuition. None of those is wrong, exactly. All of them share the same structural limitation: they optimize locally. Platform bidding optimizes one channel in isolation. A market-level planner optimizes that market. Nobody prices the interaction between them — and in a portfolio, the interaction is where the money is.
The measured cost of this is not small, and it is worth setting out precisely, because the size of the leak is what justifies the engineering.
Roughly 30% of digital spend is misdirected. Audits of enterprise advertising accounts routinely find something close to a third of working media going to mistargeted audiences, low-quality placements, and tracking error. Not stolen — misallocated, which is a different and more tractable problem.
Invalid traffic drains an estimated $63B annually. Bot impressions, domain spoofing, and fraudulent placements represent pure loss across the industry. Cross-channel optimization does not eliminate fraud, but it systematically starves the low-quality inventory where fraud concentrates, because those placements underperform in the response curves and the model reallocates away from them without being told to.
Around 12% of accounts show measurable anomalies. Duplicate targeting, overlapping audience buys, orphaned line items still spending against a campaign that concluded. These persist precisely because no single owner sees the whole portfolio at once.
Up to 83% of revenue impact goes untracked. This is the most consequential figure of the four. When reporting is siloed by platform and attribution is last-click, the great majority of the actual causal contribution — the upper-funnel audio impression that made the later search click convert — never appears in anyone’s dashboard. Channels are then defunded for failing to claim credit they earned.
These are not spending problems. They are decision problems — and decision problems have optimal answers.
Part II: Analytics reports. Optimization decides.
The distinction that made this company engagement land was categorical, not incremental. Our customer already had excellent analytics — attribution modeling, brand lift studies, dashboards across every property. Cresco did not propose replacing any of it. OptimCampaign consumes their output.
Here is the difference in one line: analytics produces a report that a human then argues over. Optimization produces an allocation with a forecast outcome attached, which a human approves or overrides. One ends in a debate. The other ends in a number.
Every optimization problem — factory line, delivery fleet, or media budget — has exactly three parts. Write them down honestly and the problem becomes solvable:
- The objective function. One number, chosen deliberately. Not four competing KPIs. For a leading Media company’s advertisers this varies by client: incremental attributable revenue for a retail account, cost per qualified lead for automotive, contribution margin for a subscription business. The hardest work in the entire engagement was not the solver — it was getting agreement on a single number to maximize. Most marketing organizations have never done this, which is precisely why allocation gets negotiated.
- The decision variables. Dollars per market, per channel, per daypart, per week. Tens of thousands of them, all moving at once.
- The constraints. Total budget. Inventory capacity by market and property. Minimum reach and frequency thresholds. Per-market floors. Contractual commitments. Compliance and category-exclusivity rules. These are hard constraints — the answer is not permitted to break them.
Stated plainly, the client problem reads: maximize incremental attributable revenue by choosing how much to spend in each market, channel and daypart — without exceeding the total budget, breaching any property’s available inventory, or falling below the reach floor required for the creative to work.
Part III: This problem has been solved before, in a factory
The reason Cresco could move quickly on this problem is that structurally, we had solved it several times already — in industries that would never describe themselves as marketing organizations.
Strip the vocabulary away and the mathematics is identical:
| OPERATIONS PROBLEM | MARKETING EQUIVALENT |
|---|---|
|
Production scheduling — which line runs which SKU, when |
Channel and flight scheduling — which property carries which message, when |
|
Fleet routing — minimize cost per delivery under capacity limits |
Media routing — minimize cost per acquisition under inventory limits |
|
Workforce coverage — staff to demand without breaching labor rules |
Reach and frequency planning — cover the audience without breaching commitments |
|
Inventory replenishment — hold the right stock at each location |
Budget replenishment — hold the right spend in each market |
|
Yield management — price perishable capacity as it expires |
Avail management — price and place perishable inventory before it expires |
A radio avail and an airline seat are the same asset class: fixed capacity, perishable at a known moment, worthless afterwards. An industry that spent forty years building the mathematics to price that asset did not do so under the heading of “marketing,” which is roughly why marketing has been slow to inherit it.
The transferable claim is narrow and worth stating exactly: the solver is the same, the constraint language is the same, the failure modes are the same. What differs is the response function — a production line’s throughput curve is engineered and knowable, while a channel’s response curve must be estimated from noisy observational data. That estimation problem is genuinely harder, and it is where the modeling effort in an OptimCampaign build actually goes. The optimization itself is comparatively routine.
This is also the honest answer to a fair challenge: what does an optimization firm know about media? Less than our client does, and that asymmetry is the design. Our client’s teams own the response behavior, the inventory reality, and the client relationship. Cresco owns the machinery that turns those into a computed allocation. Neither half works alone.
Part IV: The mechanism — why the optimum is never an even split
This is the concept that carried the room, and it is worth stating carefully because it is counterintuitive to anyone who has spent a career defending a best-performing property.
Response curves bend. The first dollars into a market or channel reach the highest-intent, most reachable audience — warm retargeting pools, high-index listeners, existing customers. Keep spending and you are buying colder audiences, higher frequency against the same people, and progressively worse placements. The return on each additional dollar falls. Anyone who has pushed a production line past its efficient utilization point recognizes the shape immediately.
Critically, curves bend at different rates. A mature, saturated flagship market decays steeply — a small addressable audience it exhausts quickly. An under-penetrated market, or a podcast audience with a long attention tail and high recall, decays gently. Which means the ranking of “best” channel inverts at a spend threshold.
Below the crossover, a dollar does more work in the flagship. Above it, the same dollar is worth more elsewhere. The correct answer is therefore never “fund the top performer.” A property can be the single most profitable line in the plan and still deserve less budget — because budget decisions happen at the margin, while “best” is an average. Those two things routinely point in opposite directions.
The shape is easiest to see as numbers. Marginal return — what the next dollar returns, not what the channel has returned overall:
| SPEND ON CHANNEL | BROADCAST (STEEP DECAY) | PODCAST NETWORK (SHALLOW DECAY) |
|---|---|---|
|
$2M |
4.8x |
3.2x |
|
$4M |
4.0x |
3.0x |
|
$6M |
3.4x |
2.8x |
|
$8M |
2.9x |
2.6x |
|
$10M ← crossover |
2.5x |
2.45x |
|
$12M |
2.2x |
2.3x |
|
$14M |
1.9x |
2.15x |
|
$16M |
1.7x |
2.05x |
|
$20M |
1.35x |
1.85x |
Illustrative curve shapes. Actual curves are estimated from the client’s own delivery and conversion history during model build.
Below roughly $10M, every dollar does more work in broadcast. Above it the ranking reverses — broadcast is saturated in its core audience while the podcast inventory is still productive. Where the curves cross is the answer.
The optimum equalizes the marginal return on the last dollar spent everywhere — the equimarginal condition. With two markets and one channel you could eyeball it. Across 160 markets, six channel types, and four flighting windows under binding inventory constraints, the optimum has to be computed. That is not a criticism of anyone’s judgment. It is a statement about problem size.
Part V: Where the $7M comes from
The arithmetic is deliberately unglamorous, because that is what makes it checkable.
Take an enterprise campaign portfolio running $60M in annual working media across our clients’s footprint. Published results from mix-optimization and MROI programmes across consumer categories cluster between 5% and 20% improvement — revenue lift from mix optimization, increases in advertising impact on sales, media ROI gains at global brands, efficiency gains from AI-augmented bidding. Cresco models the conservative end of that range.
A 4% allocation efficiency gain on $60M releases $2.4M of effectively wasted spend and redeploys it to where marginal return is highest. At an incremental ROAS of 3.0x on those redeployed dollars — below the blended average, because we model the marginal dollar rather than the average one — the forecast uplift is $7.2M in incremental attributable revenue.
The client does not receive advice. The client receives this:
| CHANNEL | CURRENT | OPTIMIZED | CHANGE | FORECAST REV. |
|---|---|---|---|---|
|
Broadcast radio — top 20 DMAs |
$18.0M |
$14.8M |
−18% |
+$1.1M |
|
Broadcast radio — secondary markets |
$11.0M$11.0M |
$12.7M |
+15% |
+$1.4M |
|
Podcast network |
$9.0M |
$12.2M |
+35% |
+$2.3M |
|
Digital / streaming audio |
$12.0M |
$12.0M |
+12% |
+$1.5M |
|
Outdoor |
$6.0M |
$4.8M |
−20% |
+$0.4M |
|
Social extension & live events |
$4.0M |
$2.1M |
−48% |
+$0.5M |
|
PORTFOLIO TOTAL |
$60.0M |
$60.0M |
no change |
+$7.2M |
Illustrative output on a $60M working media budget. Actual figures are produced from the client’s own spend and conversion history.
Note the top two rows. Flagship-market broadcast is cut 18% and still forecasts positive incremental revenue, because the dollars removed were the saturated ones — the frequency being bought against an audience already reached. Podcast takes a 35% increase not because anyone in the room prefers podcast, but because its response curve stays productive far longer than broadcast’s, and the model can see that where a quarterly planning meeting cannot.
Expect the objection: so you’re telling us to cut our strongest property. Yes — and it should not be softened. Broadcast can be the most productive channel in the portfolio on average and still deserve less budget at the margin.
Sizing the prize across budget bands
The same arithmetic, run across three budget scales and three improvement scenarios — value released in incremental revenue, without approving an additional dollar of spend:
| ANNUAL WORKING MEDIA | CONSERVATIVE (3%) | MID CASE (4%) | STRONG CASE (8%) |
|---|---|---|---|
|
$20M |
$1.8M |
$2.4M |
$4.8M |
|
$40M |
$3.6M |
$4.8M |
$9.6M |
|
$60M |
$5.4M |
$7.2M |
$14.4M |
Illustrative model. Improvement rates drawn from published third-party mix-optimization results; budget bands are scenarios for discussion.
Read the published benchmarks as a floor, not a ceiling. Each of those studies optimized a single brand’s mix in a single channel. A portfolio bidding across overlapping markets and dayparts carries a cross-channel interaction term none of them captured.
The total budget did not change. Not one additional dollar was approved. Only the distribution moved.
That is the entire shift: from a negotiation nobody can score, to a forecast anyone can check. Which brings us to the part of the engagement that matters most.
Part VI: What actually changes in the planning room
The technical account above understates the organizational shift, which is where the durable value sits. Four questions get asked in every campaign planning cycle. Today they are settled by argument. Under OptimCampaign they are settled by computation.
| THE QUESTION | TODAY | WITH OPTIMCAMPAIGN |
|---|---|---|
|
How much should this market get? |
Last year’s number, adjusted |
Computed from its response curve and inventory |
|
What will this plan deliver? |
A target, negotiated |
A forecast, with a confidence range |
|
Why this mix and not another? |
Experience and rationale |
The alternatives, scored side by side |
|
Was the plan right? |
Litigated after the fact |
Back-tested before commitment |
The row that changes the client relationship is the second. A target is an aspiration someone is accountable for defending. A forecast is a prediction with an error bar that can be checked against reality. Media companies have historically sold the first because the second was not computable. It now is — and once a vendor can offer a checkable forecast, offering only a negotiated target starts to look like a choice.
The row that changes the internal conversation is the fourth. When plan quality is assessed after the money is spent, the assessment is inseparable from the politics of who proposed it. When it is assessed against a closed quarter before commitment, the question stops being whose plan it was.
None of this removes judgment. The model has no view on brand risk, creative fit, category exclusivity pressure, or a client relationship that needs a market funded for reasons no response curve captures. Those overrides get made, and they should be. What changes is that an override now has a visible price — this decision costs $1.2M in forecast revenue, and we are making it deliberately — instead of disappearing invisibly into a plan nobody can score.
Part VII: Volunteering to be scored
Every claim above is falsifiable, and Cresco insists on it. Before a client commits budget to an optimized plan, the model is back-tested against a quarter that has already closed: the allocation the model would have chosen, the revenue it forecasts, and what actually happened — placed side by side.
A forecast that can be wrong is worth more than an opinion that can never be checked. Nobody can audit a strategy deck. Everybody can audit a forecast against a closed quarter.
That is the inversion at the center of how Cresco sells decision optimization: we ask to be scored before we are paid.
How the engagement is staged
The commitment is deliberately sequenced so that the client’s exposure stays small until the model has proven it works on their own data.
Stage one — proof on one account. A single client campaign, one closed quarter, historical data only. The model is built, back-tested, and the forecast placed against actual results. Weeks, not quarters. The deliverable is a verdict on whether the curves are learnable from the available data — and if the data will not support them, that finding is delivered plainly rather than papered over.
Stage two — parallel run. The model produces allocations alongside the existing planning process for a live campaign, without controlling budget. Both plans are visible; only the human plan executes. This is where planners find the constraints nobody wrote down — the commitments, exclusivities, and relationship obligations that live in people’s heads rather than in any system. Every optimization engagement discovers these, and the ones discovered late are the expensive ones.
Stage three — production. The optimized allocation drives planning across the account portfolio, with continuous reallocation, pacing control and anomaly detection running in flight. Model performance is reported against forecast every cycle.
The reason for the staging is not caution for its own sake. It is that an optimization model which has never been scored against a quarter it did not see is an assertion, and assertions are exactly what this system exists to replace.
Start with one quarter
If any of this is recognizable — a planning cycle that opens with last year’s allocation, a flagship property nobody can argue for cutting, a revenue target that was negotiated rather than computed — the useful next step is not a proposal. It is a test.
Send us one closed quarter of spend and conversion data for a single account. We will build the response curves, produce the allocation the model would have chosen, and place its forecast against what actually happened. If the data will not support the curves, we will say so. If it will, you will be holding a number you can check rather than a case you have to believe — and you will have it before a dollar of next quarter’s budget is committed.
That is the whole offer, and it is deliberately small. The budget was never the problem. The allocation was — and the allocation is computable.
To start a back-test on one of your accounts, reach out at info@crescointl.com or visit crescointl.com/optimcampaign
