## Budget Optimization

This reference covers strategic budget allocation principles for marketing campaigns analyzed through Billy Grace. Consult this when a user asks about reallocating spend, scaling campaigns, or optimizing their marketing budget.

### Core principles

Think in terms of **marginal performance**: where is the next euro likely to perform best? The answer is not always the channel with the highest current ROAS.

- **Maintain a full-funnel mix.** A starting framework is roughly 40% TOFU, 30% MOFU, 30% BOFU, adjusted based on business maturity and goals. Starving TOFU to boost short-term BOFU results will eventually shrink the audience pool BOFU depends on.
- **Evaluate campaigns holistically** using ROAS/POAS (profitability), CPA (efficiency), and CTR (creative relevance). No single metric tells the full story.
- **Treat the account like a portfolio.** TOFU creates demand; BOFU captures it. Both are necessary.
- **Ensure every campaign has a clear funnel role** (TOFU, MOFU, BOFU). Campaigns without a defined role are hard to evaluate and optimize.

### Critical anti-patterns

#### Do not shift budget from TOF to BOF based on ROAS alone

BOFU campaigns typically show higher ROAS because they target users who are already close to converting. Shifting budget from TOFU to BOFU may boost short-term ROAS but will shrink the top of the funnel, eventually reducing the number of high-intent users available for BOFU campaigns to convert.

Instead of moving budget across funnel stages, suggest:
- Reallocating **within** TOFU channels or campaigns (e.g., from underperforming TOFU campaigns to better ones)
- Improving TOFU efficiency through better creatives, audiences, or funnel structure

#### Branded search is volume-limited

Branded search often shows very high ROAS, but it captures existing demand rather than creating new demand. It is usually already close to saturated.

- Do not recommend large budget increases for branded search based solely on high ROAS.
- Explain that high branded ROAS is expected and not a signal that scaling will maintain the same efficiency.

#### Non-branded search has nuances

For strong non-branded search performance:
- Ask whether there is room to increase impression share or search volume.
- Consider incrementality: would some of these conversions have happened without the ad?
- Suggest gradual increases while monitoring both efficiency and TOFU health.

### Creative fatigue

Watch CTR over time. If CTR drops while the audience, offer, and competitive landscape remain stable, the most likely cause is creative fatigue -- users have seen the same ad too many times.

The right response is testing new creatives or audiences, not cutting budget. Budget cuts address a symptom; fresh creative addresses the cause.

### Avoiding common pitfalls

- **Overreacting to short-term data**: killing a TOFU campaign because it has not converted in 3 days ignores that TOFU conversions take longer to materialize. Use appropriate attribution windows and allow time for conversion lag.
- **Large abrupt budget shifts (>20%)**: major budget changes can reset platform learning phases (especially on Meta and Google), causing temporary performance dips. Suggest incremental changes.
- **Ignoring creative fatigue**: falling CTR is often a creative issue, not a targeting or channel problem.
- **Decisions during open attribution windows**: recent Session Date data will look understated due to conversion lag. Base budget decisions on fully closed periods when possible.

### Communication approach

Budget optimization recommendations touch real money, so frame suggestions carefully:

- Talk in hypotheses and tests: "It could be worth testing a 10% reallocation from Campaign X to Campaign Y."
- Acknowledge uncertainty: "Based on the current data, Campaign A appears to have room for scale, but I'd recommend monitoring for 1-2 weeks after any change."
- When users ask for exact budget moves, explain that precise reallocation is best handled by Billy Grace's Automation feature, which uses ML models trained on the client's data. You can discuss the Automation output data but cannot run Automations yourself.