Media plan for budget allocation

Use this workflow when you need to turn scattered channel results, audience evidence and campaign constraints into a budget decision that someone can approve. It is for paid media managers and marketing leads planning a defined campaign, not for producing a generic channel recommendation.

You will produce a one-page media plan, three spend scenarios and a written recommendation. Use the same input pack and decision rules each time, so the recommendation can be checked and revised when results arrive.

1. Build one campaign input pack

Create a document called Campaign media plan inputs. Do not start by asking the model to choose channels from a broad objective. Give it the facts that govern the decision.

Add these fields:

• Campaign objective: the measurable action, such as qualified lead submission, product page purchase or completed registration.
• Campaign period: start date, end date and any launch, sale or event dates.
• Total working budget: media spend only. List production, agency and technology costs separately.
• Primary audience: market, job role or customer group, geography, language, device and key exclusions.
• Offer and message: the product, proposition, proof points and mandatory wording.
• Channel performance: recent spend, impressions, clicks, conversions, conversion value where relevant, cost per result, frequency and attribution method for each channel.
• Audience evidence: customer research, first-party audience size, search demand observations, CRM segments and site behaviour. Label each source and date.
• Constraints: channel commitments, minimum spends, brand safety rules, tracking gaps, legal review needs, landing page limits and capacity to follow up leads.
• Decision rule: the result you will prioritise. For example, maximise qualified leads within a fixed budget, or protect efficiency while testing one new audience.

Keep observed data separate from assumptions. Put Observed or Assumption at the start of every line. This prevents a forecast being treated as a reported result.

2. Check the data before you use it

Compare channel results only when the definitions match. A platform-reported conversion and a CRM-qualified lead are not interchangeable. Nor are results from different campaign lengths, audience sizes or attribution settings.

Use this short audit before building scenarios:

Check	What to inspect	What to do if it differs
Conversion event	The action counted in each channel	Recalculate to one event, or label results as non-comparable
Date range	Campaign dates and seasonality	Use matching periods where possible
Attribution	Click-through, view-through or CRM match	State the method beside every result
Spend basis	Media spend versus total campaign cost	Use media spend for allocation decisions
Audience overlap	Whether channels reach the same people	Flag likely duplication and avoid adding channel reach together

Do not fill missing performance data with a precise estimate. Give the model a range, explain its basis and ask it to preserve uncertainty. If your operating limits or product behaviour may affect the task, check the current xAI documentation overview, as details are version-dependent.

3. Ask for scenarios, not a single answer

Paste the input pack into Grok, then use a prompt that requires traceable arithmetic and forces the model to retain constraints. Ask for three scenarios: conservative, recommended and expansion.

Use this prompt structure:

You are preparing a campaign media plan. Use only the information below.

Campaign input pack:
[paste the document]

Create three budget scenarios: conservative, recommended and expansion.
For each scenario, provide:
1. channel allocation in currency and percentage of media budget;
2. audience and objective by channel;
3. forecast range for the defined primary result;
4. assumptions behind the forecast;
5. risks, including audience overlap, saturation and tracking limitations;
6. the condition that would cause spend to move between channels.

Keep observed data separate from assumptions. Do not invent benchmarks or audience sizes. If data is missing, state what is missing and use a labelled range only where the input supports one.

Then recommend one scenario against this decision rule: [paste decision rule]. Show the allocation total and check that it equals the working budget.

The three scenarios have distinct jobs:

• Conservative protects the channels and audiences with the strongest comparable evidence. Reserve only a small, named test budget.
• Recommended balances expected delivery with enough testing to improve the next allocation decision.
• Expansion shows what you would do if more budget became available, or if the campaign must pursue broader reach. It is not permission to spend more.

Ask the model to allocate every pound or dollar. An unallocated remainder means the plan cannot be executed. If there is a reserve, name it test reserve and state what releases it.

4. Turn the recommended scenario into an approval page

Create a document called Media plan decision. Put the recommended scenario first. Keep the other two as appendices or a second section, so approvers can see the trade-off without mistaking options for instructions.

Use these headings:

1. Decision requested: approve the working budget and channel allocation.
2. Campaign outcome: objective, period and primary conversion event.
3. Allocation table: channel, audience, spend, percentage, role and success measure.
4. Why this allocation: three to five bullets tied to observed evidence or clearly marked assumptions.
5. Scenario trade-offs: what the conservative and expansion plans sacrifice or gain.
6. Measurement plan: reporting source, attribution method and who validates lead quality or revenue.
7. Reallocation rules: the conditions and review dates for moving budget.

For reallocation rules, use a condition that a manager can actually apply. For example: move the test reserve only after a channel has spent its agreed test amount, tracking is functioning, and its primary result is within the approved efficiency range. Do not write optimise based on performance. That leaves the decision undefined.

5. Check where the output may be wrong

Review the model's draft against the input pack line by line. The most common error is not bad writing. It is an assumption that has been presented as a fact.

Look for these signals:

If you see this	Treat it as	Action
A conversion forecast with no calculation or evidence	Unsupported precision	Replace it with a range or remove it
A channel recommendation not linked to an audience or objective	Generic advice	Specify the role, or exclude the channel
Channel percentages that do not equal 100%	Arithmetic error	Recalculate before sharing
Combined reach across overlapping channels	Possible double counting	State overlap risk and use deduplicated measurement where available
A recommendation that ignores a constraint	Unusable plan	Put the constraint into the allocation or explain the exception

Also check the landing page and follow-up process. There is little value in shifting budget to create more leads if the page cannot support the message, the form cannot be measured, or the sales team cannot classify lead quality.

6. Run the same review cadence during the campaign

Schedule a brief review at the intervals set in the plan. Bring the same fields each time: spend, primary results, cost per primary result, delivery, frequency, conversion quality and tracking status. Compare actuals with the scenario assumptions, not merely with last week's result.

Record every allocation change in a Media plan change log with the date, amount moved, reason, evidence and expected effect. This creates usable evidence for the next campaign rather than a collection of unexplained platform changes.

When the plan does not work

If the model returns broad channel advice, your input pack is probably missing a decision rule, comparable performance data or constraints. Add the missing field and run the scenario prompt again.

If the numbers look plausible but cannot be defended, reduce the forecast detail. Use an evidence-backed range, ring-fence a test reserve and set a review point. If tracking or lead-quality data is unreliable, pause allocation changes based on that metric until the measurement owner confirms what the result means.