A campaign generates 500 customers at an average cost of €80. Will doubling its budget generate 1,000 customers? Rarely. The first auctions captured the most favourable situations; subsequent ones broaden coverage, increase frequency or enter less effective inventory.
The average describes the past. Marginal cost decides the next euro.
That is the principle.
1. Definition: scaling is a series of economic steps, not a continuous increase
The term “advertising scaling” means the controlled increase of media investment and the capabilities around it to produce more valuable outcomes without breaching profitability, cash-flow, quality or market constraints.
It can be vertical—more budget on an existing set—; horizontal—new audiences, intents, geographies, offers or channels—; or operational—more creative, data, stock and sales capacity. Sustainable scaling generally combines all three dimensions.
Scaling is therefore not a platform feature. It is an allocation decision under uncertainty, with a maximum acceptable loss and a learning schedule.
2. Key figures: growing budgets, uneven returns
- IAB Europe measures €131 billion in digital advertising in Europe in 2025, an increase of +10.5% year on year. Video grew by 19.6%, social media by 19.2% and retail media by 16.7%.
- In France, the market reached €12.4 billion in 2025, according to SRI and Udecam. Eight global participants capture 83% of growth, making a scaling strategy highly dependent on a small number of automated systems.
- The Global Annual Marketing Report 2024 from Nielsen reported that 72% of surveyed marketers expected advertising budgets to increase and that two-thirds of media budgets were going to digital, yet only 38% measured traditional and digital marketing together to assess overall ROI.
- Nielsen cites a 2022 analysis of 150,000 observations of ROI and media plans: 50% of planned investment was too low on certain channels to achieve the best return, with an estimated 50% shortfall. Under-investing can therefore be as inefficient as saturation.
- The documentation for Robyn, Meta’s open-source MMM solution, explicitly models diminishing returns: every additional unit increases response, but by progressively less. It also warns that saturation parameters are difficult to identify without calibration from experiments or other ground truth.
These figures provide no universal ratio. They show why the right level depends on a curve, context and objective, not a “+20% per week” rule repeated everywhere.
3. Gate 1 — Can the unit economics support the next step?
Before changing a campaign, establish the financial ceiling. For a purchase, start with net revenue and deduct collected taxes, product or service cost, payment, fulfilment, returns, promotions and directly variable costs. For a lead, multiply customer contribution by observed qualification and close rates.
3.1. Calculate maximum CPA
A business sells for €250 net, retains a 60% contribution margin before advertising and wants to keep €40 after media. Its maximum customer CPA is therefore:
€250 × 60% − €40 = €110.
If only 25% of leads become customers, the ceiling per lead is €27.50. This threshold must also incorporate uncertainty, refund delay and creative cost when these vary with scale.
For a recurring model, do not replace observed value with an optimistic LTV. Use sufficiently mature cohorts, segment channels when retention differs and discount future revenue. A customer acquired for €500 is not profitable today simply because a spreadsheet assumes five years without churn.
3.2. Distinguish average from marginal performance
Suppose €40,000 is invested for 500 customers: average CPA is €80. An additional €10,000 tranche brings 80 more customers, or €125 for the tranche. The new average CPA becomes €86.21 and remains below a €100 threshold, but the next euro is already in an unprofitable range.
This distinction prevents a common mistake: defending an increase using accumulated historical returns while the new capital is reducing contribution. For every tranche, the table should show spend, additional outcomes, marginal cost, net revenue and marginal contribution.
The financial gate turns green when the prudent scenario remains fundable and the cost of learning is capped.
4. Gate 2 — Can the organisation absorb the demand?
Media can accelerate faster than production, stock, support or the sales team. The business then receives more late orders, poorly followed-up leads, cancellations and negative reviews, while the platform continues celebrating the initial conversions.
4.1. Measure capacity at the constraint
Map the maximum throughput of each stage: visits, forms, qualification, meetings, proposals, orders, preparation, delivery and support. The bottleneck sets the system’s temporary ceiling, even when every other function has spare capacity.
B2B example: eight salespeople can handle 40 new qualified leads a day with a call-back within two hours. Beyond that, the delay rises to twenty-four hours and the meeting rate falls from 35% to 24%. Unchanged media cost then conceals a commercial loss. To scale, the business must automate booking, reinforce the team or slow the inflow.
For retail, track availability by SKU, margin by category, delivery promise, cancellation rate and support cost. Excluding nearly exhausted products or temporarily reducing the value sent to bidding protects the customer relationship and directs the algorithm towards what the business can actually fulfil.
Cash flow matters too. Compare platform payment dates, the stock cycle, refunds and receipts. A scenario that is profitable over twelve months can cause a liquidity crisis in month three if it requires too much working capital.
Passing the operational gate requires reserved capacity, alert thresholds and a plan for the identified bottleneck.
5. Gate 3 — Can measurement recognise genuine growth?
Increasing a budget often raises attributed conversions, but some come from existing customers, demand captured elsewhere or additional exposures to the same people. Scaling requires analysis that separates claimed volume from additional outcomes.
5.1. Establish three levels of evidence
The first level is operational: spend, delivery, conversion, quality, net sale and contribution. The second is marginal: the outcome of each step compared with a baseline adjusted for seasonality and promotions. The third is causal: a control group, geo-test, conversion lift or calibrated model when the value at stake justifies it.
Google defines iROAS as incremental value divided by incremental cost. An iROAS of 1.8 means an estimated €1.80 of additional value per extra euro before contribution is applied. If contribution is 40%, the return before other costs is only €0.72.
State the uncertainty. If a test estimates 12% lift with an interval compatible with 2% to 22%, the budget does not deserve the same confidence as a tightly measured effect. Plan a decision for low, central and high scenarios rather than retaining only the flattering value.
Platforms remain useful for daily corrections. Finance, the CRM, experiments and MMM answer other questions. Bringing them together in a coherent architecture avoids asking last-click attribution to plan all growth.
The evidence becomes sufficient when the step has a baseline, primary metric, guardrail and reading window compatible with the cycle.
6. Gate 4 — Can the creative engine renew the evidence?
A budget increase exposes available audiences to the same messages more quickly. Frequency rises, the best creative fatigues and the algorithm explores weaker combinations. Without creative throughput, vertical scaling consumes its own advantage.
6.1. Work with concepts, not files
A concept combines a tension, promise, proof, treatment and call to action. Each concept produces variants in opening, format, duration and language, allowing the team to learn why it works instead of concluding that one particular file “wins”.
Plan a portfolio: proven creative supporting the core, variations intended to extend its life, new angles for expansion and brand content that develops demand. The number depends on spend, audience fragmentation and fatigue speed. It is calculated from observed throughput, not a generic quota.
Creative analysis connects attention with outcome. Start rate, retention, click and engagement diagnose the message, but sales quality, net sales and lift decide its value. Highly clicked content can attract unprofitable curiosity.
Preserve the guardrails: rights, claims, regulated sectors, prices, areas and pages. Automated generation and assembly increase the number of combinations and therefore the control surface required.
The creative gate permits an increase when the team has ready concepts, a production cadence and a protocol for detecting fatigue before performance collapses.
7. Gate 5 — Is the pool of demand still large enough?
Even excellent execution eventually covers the people who are currently reachable. The team must know whether it can intensify, broaden or must first create more demand.
7.1. Read the saturation signals
For search: impression share, query volume, new terms, position, marginal CPC and brand share. For social: unique reach, frequency, audience size, CPM, fatigue rate and new buyers. For programmatic: deduplicated reach, inventory, cross-publisher frequency, supply-chain costs and environment quality.
Diminishing returns do not automatically mean spend should be cut. A channel close to saturation may still deliver substantial absolute contribution. The decision compares its next euro with every available option, including improving the product, conversion, awareness or distribution.
MMM often represents this relationship using a response curve and carryover effect. Robyn’s documentation nevertheless recommends calibrating saturation from ground truth where possible, because several curves can explain the same historical data. A simulation remains a hypothesis to test.
7.2. Choose the path to expansion
- Intensification: more coverage or higher bidding within the existing core.
- Adjacent extension: new queries, audiences, formats, regions or closely related offers.
- Diversification: a different channel with a distinct role and dedicated measurement.
- Development: content, brand, product or distribution to enlarge the future market.
Isolating extensions reveals their initial economics. If they are immediately mixed into the core, historical returns subsidise their learning and the team can no longer tell what the step actually produced.
Expansion can begin when it has a demand hypothesis, its own economics and a stop rule.
8. The budget staircase: a six-step Logiks method
This method is a Logiks management framework. Step sizes and durations are adapted to volume, seasonality, conversion delay and platform constraints.
8.1. Step 0 — Stabilise the baseline
Choose a representative period, reconcile sales and spend, exclude incidents, then calculate average and marginal costs by segment. No increase can fix a broken measurement system.
8.2. Step 1 — Define three scenarios
The low scenario applies a deterioration in conversion, costs and quality. The central scenario extends the adjusted trend. The high scenario assumes favourable execution without replacing the cash-flow plan with optimism.
8.3. Step 2 — Fund a reversible tranche
The tranche must produce enough signal for a decision while respecting the maximum loss. A tiny increase lost in variability teaches nothing; an immediate doubling exposes the economics unnecessarily.
8.4. Step 3 — Freeze what is not being tested
Keep price, promotion, targeting, creative and pages constant as far as possible during the reading window. If a constraint requires a change, document it so that an effect originating elsewhere is not attributed to the budget.
8.5. Step 4 — Read the results after maturity
Wait for the necessary cycle, track cohorts and compare incremental spend, net customers, contribution, quality and capacity. An operational signal can justify an emergency stop, but not a claim of causal success after forty-eight hours.
8.6. Step 5 — Maintain, extend or revert
Maintain if the threshold and guardrails hold. Extend if the next step remains plausible. Revert if the economics or quality break, then retain the learning in the register instead of concealing the reversal.
8.7. Step 6 — Recalibrate the portfolio
At regular intervals, compare marginal returns across channels, short-term contribution, brand development and business constraints. A platform’s local optimum does not guarantee the global optimum.
9. Contribution simulation: choose between two routes to market
Management has an additional €30,000. Route A reinforces an established channel whose expected marginal cost is between €110 and €150 per customer. Route B opens an adjacent audience, with a still-unknown cost estimated between €130 and €210, but capacity for 400 new customers versus 180 for A.
With €90 of contribution available per acquisition, neither route appears immediately acceptable. The business can, however, improve the page and gain €20 of contribution, negotiate a €15 lower service cost, or reserve €6,000 for a trial that measures quality before committing the rest. The relevant choice combines economics with the option to learn.
Management therefore first funds €10,000 on A, whose lower range remains close to the threshold after improvement, and €6,000 on B with a strict limit. The remaining €14,000 is not pre-allocated: it is an option to be exercised after the cohorts mature if either step demonstrates positive contribution.
This mechanism avoids confusing authorised budget with budget that must be spent. It also assigns value to information: losing €1,500 on a well-designed trial can be rational if the result then prevents €100,000 being allocated to a saturated market.
Finally, the scenario retains a non-media route. If neither route clears the threshold, the remaining budget can fund conversion, creative renewal or capacity, then return to media with transformed economics.
Waiting can be an active choice.
The threshold decides. Cash confirms. Capacity absorbs. Creative holds. Evidence authorises. Then you move up.
10. Scaling management dashboard
| Gate | Primary metric | Guardrail | Question |
|---|---|---|---|
| Economics | marginal contribution | CPA, discount, return | Does the step still create a net outcome? |
| Capacity | throughput at the bottleneck | delay, stock, quality | Can the business fulfil what is acquired? |
| Measurement | additional customers or sales | interval, deduplication | What share genuinely comes from the increase? |
| Creative | contribution by concept | frequency, fatigue, compliance | Can the message withstand the pressure? |
| Demand | new reach or intent | marginal CPC/CPM, brand | Is the pool expanding or repeating? |
The weekly committee monitors execution. The monthly committee decides the steps. The quarterly review reassesses allocation and structural capacity.
11. Logiks recommendations: nine mistakes that make growth fragile
- Multiply the budget on the basis of average ROAS.
- Confuse attributed sales with additional sales.
- Send an unobserved theoretical LTV to bidding.
- Ignore sales capacity or stock.
- Change budget, targeting and creative on the same day.
- Mix expansion and the established core in one result.
- Cut all brand investment to fund the short term.
- Treat the saturation model as truth without calibration.
- Continue a step because reverting would be politically uncomfortable.
Reversibility is a strength. It protects capital and accelerates learning.
12. FAQ
12.1. By how much should an advertising budget be increased?
There is no universal percentage. The tranche must be large enough to rise above noise, small enough to respect the maximum loss and observed over a cycle compatible with conversion.
12.2. How can you tell when a campaign is saturated?
Marginal cost rises, new reach slows, frequency increases, creative fatigues or queries move away from the core. Confirm with a response curve, segmentation and, if possible, an experiment.
12.3. Should you open a new channel?
Only if it fulfils a distinct role, reaches an additional audience or moment and receives enough budget to learn. Adding an underfunded platform mainly increases fragmentation.
12.4. ROAS or CPA for performance management?
Both are incomplete without contribution. Use CPA for acquisition efficiency, ROAS for relative revenue, then marginal contribution and cash flow for the business decision.
12.5. Does automation make scaling easier?
It accelerates allocation within the scope and according to the objective provided. It does not correct a poor value signal, limited offer, empty stock or attribution that overstates the effect.
12.6. Should spend always increase when the channel is profitable?
No. Compare the return on the next euro with the alternatives and constraints. Investing in conversion, the product, brand or capacity can sometimes produce more.
13. Conclusion
Advertising scale is a staircase. Every step requires fundable economics, available capacity, credible measurement, renewable creative and a pool of demand that remains open.
The right question is not “how much can we spend?” but “which additional tranche still creates robust contribution, and what evidence will authorise the next step?”
Growth remains a decision.
14. Main sources
- IAB Europe, AdEx Benchmark 2025, 7 July 2026: https://iabeurope.eu/knowledge_hub/iab-europe-adex-benchmark-2025-report/
- SRI and Udecam, 35th Online Advertising Observatory, 10 February 2026: https://www.sri-france.org/observatoire-epub/35eme-observatoire-de-le-pub/
- Nielsen, 2024 Annual Marketing Report: https://www.nielsen.com/insights/2024/maximizing-roi-in-a-fragmented-world-nielsen-annual-marketing-report/
- Nielsen, 2024–25 Upfronts/NewFronts Guide, analysis of 150,000 observations: https://www.nielsen.com/insights/2024/upfronts-newfronts/
- Meta Marketing Science, Robyn, Key Features — saturation and marginal response: https://facebookexperimental.github.io/Robyn/docs/features/
- Meta Marketing Science, Robyn, An Analyst’s Guide to MMM: https://facebookexperimental.github.io/Robyn/docs/analysts-guide-to-MMM/
- Google Ads Help, geo-measurement and iROAS: https://support.google.com/google-ads/answer/14102986
- Google, Modern Measurement Playbook, 2026: https://www.thinkwithgoogle.com/_qs/documents/18393/For_pub_on_TwG___External_Playbook_Modern_Measurement.pdf