A 6-step AI workflow for building better seasonal campaigns

Seasonal campaigns can boost sales and deliver a strong ROI. The most successful ones improve with each season.

From what I’ve seen, campaigns that bring together industry trends, consumer behavior and brand strategy tend to outperform those that focus on just one area. The challenge is that combining these inputs takes more time than most marketing teams can spare.

I’ve built a workflow to address that problem. It’s designed to help you create a set of prompts for LLMs and iterate on them to support specific marketing activities.

Agents aren’t just a single prompt. They’re a process where each response builds on the previous one. After some initial setup, you can run this sequence within a single chat:

• Intake prompt: Analyze the uploaded data and extract buyer motivators, anxieties and decision drivers by borrower type (first-time, move-up and refinance)
• Synthesis prompt: Identify the most effective promotion angles for addressing the borrower’s highest-anxiety moments, ranked by potential impact
• Build prompt: Develop the full promotion framework — hypothesis, campaign themes, offer recommendation, timeline and channel-specific messaging
• Refinement prompt: Adapt the campaign for specific segments or scenarios

We’ll use Claude to walk through this process using a mortgage example — a category where the buying journey is long, complex and emotional. If it works here, it can translate well to industries with more linear purchase paths.

Step 1: Define the purpose of your AI project

Begin by clearly defining your project’s purpose. In this case, the goal is to build a system that plans effective seasonal campaigns to increase mortgage applications and the number of funded loans. Treat this like a new agency kickoff. The more focused you are, the better your results will be. Clear goals at the start make everything else easier.

For a mortgage company, this means creating an AI tool that understands your loan products, borrower types, marketing trends, current rates, environmental factors and the trust signals that help turn a rate shopper into an applicant. Building a workflow ensures the results are relevant to your business.

Tip: Context and nuance matter in AI. Be wary of outputs from generic marketing agents, copy-and-paste or free solutions.

Step 2: Create your Claude project and add reference materials

Next, set up your Claude project workspace and add your initial reference materials by uploading documents, linking a database like Google Drive or connecting to an API. Choose what you share with Claude carefully and ask yourself whether each item helps you reach your project goal and whether your business’s guidelines allow data sharing.

Before using any files, check with your leadership, legal or compliance team. This is especially important in regulated industries. You need to know what your policies allow and where the limits are. If you are unsure whether something is safe to share, it’s best to leave it out.

You can also mask or index data instead of sharing raw details. For example, Google Trends shows search patterns without revealing exact numbers. In the same way, you can share campaign results as a percentage of your yearly total or show loan data as ranges instead of actual amounts. Large language models can work well with this kind of information. 

If you plan to go beyond uploading documents and want Claude to edit files, write to a database or run code, think carefully about what could happen before you let it run on its own. If you aren’t an expert in running these tools, always get outside advice before setting up these workflows.

Tip: You don’t need everything at once. You can add or remove materials as you go. If you link to a Google Drive file, share access to a copy rather than your original files.

There are many options when importing or adding files. This is the framework I use with clients to agree on what to include. You don’t need everything on day one. Start with reviews, past campaign results and any borrower research you have. Even just those three will give you a big head start.

Remember, you don’t need everything on day one. You can start with simple reference points, such as: 

• Examples and messages from one past campaign.
• Your brand guide.
• 5-10 customer reviews. 
• A consumer persona.
• An overview of what has worked in the past. 

Even just these few will give you a big head start.

Step 3: Write a good prompt (role, context, task, output, considerations)

Here’s an example prompt, but you can keep it simpler. The key is the structure (role, context, task, output and considerations). Use this section to set the ground rules for your output. It’s often skipped, but it’s important. If you have specific requirements, like always starting with a hypothesis or never giving generic advice, include them here.

Who you are
You are a promotional strategy consultant specializing in mortgage and home financing. Your expertise is in high-consideration purchase categories where buyer anxiety is the primary conversion barrier.

Business context
[Lender name] is a [XYZ regional] mortgage lender offering [purchase, refinance, FHA, VA, jumbo, etc. products]. Primary markets: [list states or metros]. What makes your brand different from competitors? [Insert key differentiators.]

Consumer context
Segment borrowers by type: first-time buyer, move-up buyer and refinance. First-time buyers are anxiety-driven (rate, qualification and trust are the primary barriers). Move-up buyers are equity-driven and sensitive to timing. Refinance borrowers need a clear financial trigger to act. When analyzing behavior, reference the uploaded persona files and segment all recommendations by borrower type.

How to use the data I uploaded
Reviews = parse for trust signals and anxiety language (use verbatim phrases when relevant). CRM data = analyze funnel behavior and drop-off patterns. Campaign results = analyze what offers drove action and what didn’t. Financial data = identify seasonal volume patterns and loan mix trends.

Create rules for your ​output
Always start with a strategic hypothesis. This is one sentence explaining why this promotion will work, given borrower psychology, my business and current dynamics. Always provide two campaign theme options. Always include a channel-specific messaging plan. Never recommend a rate discount as the primary offer without connecting it to a specific borrower anxiety or decision trigger. Never give generic category advice. Never use superlative claims such as ‘best rate,’ ‘lowest cost,’ or ‘fastest close’ in campaign language unless the client has independent, verifiable data to support the claim. Ground all recommendations in the sources I share or reference. If the uploaded data is insufficient to support a recommendation, say so and identify what data would close the gap.

Writing tone

Write clearly, punchy, channel-native. Provide guidance as if you were a senior BCG consultant presenting to a CFO and a CMO, but base all your recommendations on current best practices for digital marketing channels. Avoid phrases that sound like typical AI responses, such as “it’s not this, it’s that.”​

Step 4: Connect your research sources

Group your sources into three types: research you upload, data you collect and format as a CSV and live API sources that Claude can access with the Research feature.

Some sources may not be easy to get. You might need to share Google Docs with video transcripts, include PDFs or use an aggregation tool.

Here are some specific examples of important information:

• National Association of Realtors Profile of Home Buyers and Sellers: A survey of recent buyers covering motivators, down payment behavior, agent influence and time-to-decision.
• Bank of America Homebuyer Insights Report: An annual study on what’s keeping buyers on the sidelines, the financial sacrifices they’re making and the external anxiety drivers.
• Home Mortgage Disclosure Act and Federal Financial Institutions Examination Council mortgage data.
• Zillow research.
• Redfin Data Center: A source where you can download housing market data.

Step 5: Start writing your prompts

Ask Claude to do deep research using what you shared:

• Click the Research button at the bottom left of the chat interface until it turns blue.
• Then use the Research tool (note: you’ll need a paid plan for Claude) to find current mortgage, builder, real estate and rate promos in [XYZ market], such as first-time-mover offers, rate buydowns, closing-cost credits, first-time buyer specials and pre-approval incentives that have been active in the last 90 days. 
• Go further with the Research tool to find the most common fears, objections and frustrations expressed by first-time home buyers about the mortgage process on Reddit, personal finance forums, social media and review sites in the last 6 months.

Now, it’s time for you to write the intake prompt.

Role:
You are a promotional strategy consultant for [Lender Name] during the [season/time period] campaign planning period.

Context:
Use the uploaded borrower research, review data, CRM funnel behavior, past campaign results and the NAR and CFPB research files.

Task:
Analyze the buyer motivators, anxieties and decision drivers for first-time buyers, move-up buyers and refinance borrowers based on the uploaded data.

​Then build a promotion framework that includes:

1. A strategic hypothesis for the promotion
2. Two campaign theme options
3. Offer recommendation tied to borrower demand curve logic and the current rate environment
4. Create a three-month messaging plan broken out by: 1) Organic social, 2) PR, 3) Paid social and 4) Email

Output format:
Start with the hypothesis. Then describe the theme options. Then offer a rationale. Then, a channel-by-channel calendar and messaging plan.

Step 6: Iterate as results come in

Update your project with results after every promotion. Claude’s outputs become more precise as you add new information. However, recommendations will also become more diluted in the presence of more noise. Curate and prioritize what you upload.

For example:​

• Campaign results: This includes what ran, what converted, what the offer was and what didn’t work.
• A short recap note: This could be even a paragraph summarizing what you learned about borrower behavior.
• Any new review data collected post-campaign.
• Updated CRM data reflecting the cohort of borrowers from that promotion.

Turning campaign results into better inputs

Marketers who connect ideas and think creatively and critically gain a significant advantage. Connecting ideas was much harder and more time-intensive before the rise of LLM tools, so take advantage of the technology available to speed up great work.

Disclaimer: I’m not a legal expert. Please use this as a general guide and check any marketing plans with your legal team before proceeding.

The post A 6-step AI workflow for building better seasonal campaigns appeared first on MarTech.