How to Build a 6-Step AI Workflow for Seasonal Marketing Campaigns

Most seasonal campaigns fail before the first creative brief is written — because the team is working from gut instinct instead of structured intelligence. A new practitioner-developed framework published on Martech.org by Ed Poppe, Founder and Fractional Marketing Leader at Poppe Marketing, lays out a six-step system for turning scattered CRM data, market research, and brand context into a repeatable AI-powered campaign planning engine. If you’re still using AI as a glorified copywriter on your seasonal pushes, this workflow is the operational upgrade you’ve been missing.

What Happened

Ed Poppe, who built digital growth programs for enterprise brands including Tempur-Pedic, Build-A-Bear, La Quinta, and American Airlines, published a detailed methodology on April 10, 2026 via Martech.org showing how to use AI — specifically Claude — as a strategic campaign planning system rather than a content generation tool.

The framework isn’t theoretical. It’s built from the same operational thinking that runs performance marketing at scale: define what you want the AI to know, give it the right inputs, structure how it reasons, and then close the feedback loop with real campaign data. Poppe uses a mortgage lending company as his worked example throughout, making the methodology concrete and sector-specific rather than aspirationally generic.

Here’s how the six steps break down:

Step 1: Define Project Purpose. Before touching any AI tool, you establish the mission scope in plain language. In Poppe’s mortgage example, the objective is building a system that understands loan products, borrower types, market trends, interest rates, and trust signals — with a stated goal of increasing applications and funded loans. This sounds obvious, but most teams skip it. Without a clear project purpose, AI outputs drift toward generic advice that nobody can act on. The project purpose is the filter that every subsequent prompt gets run through, and it’s what keeps a multi-month campaign planning system coherent as new team members and new data get added over time.

Step 2: Create a Claude Project and Upload Reference Materials. This is where the infrastructure work happens. You create a dedicated Claude Project — Anthropic’s workspace feature that maintains persistent context across conversations — and populate it with your organization’s actual knowledge base. Poppe recommends consulting legal and compliance teams before uploading anything in regulated industries, noting that teams can mask sensitive information or use indexed data rather than raw figures. The reference materials you load at this stage are the foundation everything else is built on — which is why Step 2 is really two jobs: finding what you have and deciding what’s safe to share.

Step 3: Write Effective Prompts with Five Components. Poppe’s prompt structure is the most technically actionable part of the framework. Each prompt should include a role (define Claude as “a promotional strategy consultant specializing in mortgage financing”), context (the current business situation), task (the specific thing you need done), output format (what you want back and in what structure), and considerations (standing guardrails like “always start with a strategic hypothesis” and “never recommend generic advice”). Structured prompting like this converts a general-purpose large language model into a role-aware strategic partner that applies professional judgment rather than producing averaged-out marketing advice.

Step 4: Connect Research Sources. Inputs are grouped into three categories: uploaded research (past campaigns, borrower personas, brand documents), formatted data (CSVs of campaign performance and CRM exports), and live API sources (real-time market data feeds where available). Poppe lists specific recommended inputs for the mortgage vertical: National Association of Realtors research, Bank of America Homebuyer Insights, mortgage disclosure data, Zillow research, and competitive intelligence on rate promotions and builder partnerships. The point is to give the AI system enough external market context that it can reason about conditions the organization itself doesn’t have proprietary data on.

Step 5: Start Writing Prompts — Beginning With Market Research. With the project loaded and role-defined, the first active prompts use Claude’s Research tool to investigate current market conditions, competitor promotions, and buyer concerns by segment. The intake prompt asks Claude to analyze buyer motivators, anxieties, and decision drivers across customer segments. This is where the system starts generating actual strategic intelligence rather than generic content. The output of this phase is a set of segment-specific briefs that can anchor the full campaign — messaging architecture, promotional timing, channel priorities, and objection-handling frameworks.

Step 6: Iterate as Campaign Results Come In. The feedback loop is what separates a one-time AI experiment from a compounding intelligence system. After campaigns run, you feed results back into the Claude Project: email performance dashboards, paid media results, updated CRM data, customer review data, and campaign recap notes. Each iteration improves the system’s recommendation precision for the next seasonal push. The spring campaign results become training context for the back-to-school cycle. The Q4 holiday data sharpens Q1 retention campaign strategy. The system gets better every cycle if — and only if — you close the loop.

Poppe’s framing throughout is explicit: “context and nuance matter in AI,” and the workflow is specifically designed to combat the generic outputs that result when marketers prompt AI with no brand context, no performance history, and no strategic framing. The methodology emerged from Poppe’s direct experience scaling digital growth across multiple enterprise brands, and it reads like a framework built by someone who has paid the price for unstructured AI adoption.

Why This Matters

The reason this framework addresses the problem that matters most is that it tackles the single biggest failure mode in AI-assisted marketing: shallow context. Most marketing teams using AI today are effectively running it cold — no persistent memory of past campaigns, no CRM integration, no structured role definition. They get outputs that are technically coherent but strategically hollow, and they wonder why their AI-generated campaign briefs sound indistinguishable from competitor briefs written by the same AI with slightly different prompts.

For agencies, this workflow changes the economics of seasonal campaign planning. The research phase — competitor analysis, customer segmentation, market trend mapping — typically consumes 30 to 40 percent of pre-campaign hours. A well-configured Claude Project that already holds your client’s past campaign data, brand guidelines, and research corpus compresses that work dramatically. You’re not starting from scratch every season; you’re building on a knowledge base that grows smarter with each campaign cycle. For agencies running five or more seasonal clients, that compression effect is significant enough to change what’s profitable at what price point.

For in-house teams at mid-market brands, the value is different but equally significant. Most in-house teams lack the headcount to run rigorous seasonal research across every channel and every audience segment. A structured AI project becomes the equivalent of a research analyst who never forgets a past campaign result, can synthesize competitive intelligence on demand, and doesn’t require onboarding when the marketing director changes.

The framework also directly challenges a widespread assumption in the marketing AI space: that better prompting alone drives better outputs. Poppe’s model makes clear that prompting is only one of six components. The quality of the reference materials — the CRM data, the past campaign dashboards, the customer personas, the competitive intel — is equally important. AI is only as smart as the context you give it. Teams investing exclusively in prompt engineering without investing in data curation are leaving most of the value on the table.

There’s a compliance dimension here that deserves serious attention. Poppe explicitly flags data privacy as a non-negotiable consideration before Step 2, and this is particularly relevant for regulated verticals: financial services, healthcare, insurance, and legal sectors. Many enterprise teams are still operating without a clear internal policy on what data can be shared with third-party AI platforms. This framework forces that conversation to happen before any AI-powered campaign planning begins — which is the right sequence, not an afterthought you manage after you’ve already uploaded sensitive data.

The framework also has structural implications for how teams think about seasonal campaigns. Most seasonal campaigns are treated as discrete projects: plan, execute, wrap, move on. Poppe’s iterative model treats each campaign as an input into a continuously improving system. The Q4 holiday campaign results become training context for the Valentine’s Day push. The spring home buying season data informs back-to-school positioning for adjacent product lines. This is a fundamentally different operating model — one that compounds in value over time rather than resetting to zero with each new season.

For solopreneurs and small agencies, the barrier to entry is surprisingly low. Claude Projects is available in paid tiers, and the reference materials you upload don’t require engineering integration — PDFs, CSVs, Google Docs exports, and even copied-in text work. A single-person agency running three to five clients could build a dedicated Claude Project for each client in a matter of hours, with materials that immediately improve the AI’s strategic usefulness on the very next brief.

The Data

The table below maps the nine categories of reference materials Poppe recommends loading into a Claude Project for seasonal campaign planning, based on his mortgage lending example as published in Martech.org. This serves as a practical checklist for standing up a project in your own vertical — adapt the specific sources to your industry, but the category structure transfers directly.

To put the measurement stakes in context: Martech.org research on email marketing published in April 2026 found that 78% of marketers say email is very or extremely important to organizational success, yet only 46% can measure the ROI of their promotional email programs. Among teams that do measure, 60% report returns exceeding $10 for every $1 spent — with 13% reporting $40 or more per dollar. The gap between teams that measure and teams that don’t is not primarily a technology gap. It’s a data infrastructure gap: teams that can’t measure can’t load credible performance data into an AI system, which means they also can’t benefit fully from a framework like Poppe’s. The implication is direct: fixing your measurement infrastructure is not separate from AI readiness. It is AI readiness.

Real-World Use Cases

Use Case 1: Regional Home Goods Retailer Building a Spring Sale Campaign

Scenario: A regional home goods chain with 40 locations runs quarterly promotional campaigns tied to seasonal home improvement cycles. Their marketing team of four handles all channels including email, paid search, social, and in-store signage. Campaign planning currently takes three weeks per cycle and relies heavily on the marketing director’s institutional knowledge, which creates a single point of failure whenever team members turn over.

Implementation: The team creates a Claude Project and uploads: the last three years of spring campaign dashboards (email open rates, click rates, conversion rates, average order value by segment), their customer segmentation model from the CRM (age group, purchase frequency, category affinity), Google and Yelp review exports to capture actual customer language, a competitor promotion tracking spreadsheet, and their current brand guidelines with approved campaign creative. They write a role prompt defining Claude as a “retail promotional strategy consultant with deep expertise in seasonal home goods marketing” with a standing instruction to “always connect recommendations to specific customer segments, never to the general audience.” The first prompt asks Claude to synthesize buyer motivators by segment for the spring home improvement season and identify the top three anxiety points that delay purchase. Subsequent prompts request a six-week promotional calendar with channel-specific messaging frameworks and timing rationale grounded in the performance history uploaded.

Expected Outcome: The research and strategy phase compresses from three weeks to three to five days. Campaign briefs arrive pre-populated with customer language drawn directly from actual reviews, segment-specific messaging angles grounded in CRM data, and a competitive gap analysis the team previously couldn’t staff. After the campaign runs, performance data gets fed back into the project before summer planning begins, so the system is already smarter for the next cycle without requiring anyone to manually synthesize what worked.

Use Case 2: B2B SaaS Company Aligning Campaign Planning to Fiscal Buying Cycles

Scenario: A mid-market B2B SaaS company selling procurement software sees predictable buying spikes at fiscal year-end for most clients and during budget-setting periods at the start of the calendar year. Their demand generation team runs paid and content programs but struggles to adapt messaging quickly enough to match where different buyers are in the evaluation cycle at any given point in the season.

Implementation: The team builds a Claude Project loaded with: closed-won deal data by month (anonymized at the individual level but segmented by company size, vertical, and deal duration), sales call notes organized by objection type and deal stage, product positioning documents and competitive battle cards, quarterly win/loss analysis from the sales team, and search keyword performance data segmented by intent category — awareness versus evaluation versus decision. The role prompt defines Claude as a “B2B demand generation strategist specializing in enterprise software procurement cycles” with a standing instruction to “always anchor recommendations to pipeline velocity impact, not just lead volume.” Initial prompts ask Claude to map seasonal buyer anxiety peaks to content topics and to identify which ad angles drove the highest pipeline-to-close conversion rate in the prior Q4 push.

Expected Outcome: Campaign briefs arrive with objection-aware messaging that maps to the actual evaluation stage the buyer is likely in at each point in the seasonal calendar. Content themes align to what sales is actually hearing in calls, reducing the persistent disconnect between marketing messaging and sales enablement materials. The feedback loop means each Q4 cycle builds on actual win/loss intelligence from the prior year, so the system progressively improves its understanding of what accelerates deal velocity at each buying stage.

Use Case 3: Mortgage Lender Running Spring Home Buying Season Campaigns

Scenario: This is Poppe’s direct example from the Martech.org article. A mortgage lender experiences its highest application volume during the spring home buying season — March through June — and needs campaigns that speak credibly to meaningfully different borrower profiles: first-time buyers facing down payment anxiety, move-up buyers navigating equity calculation complexity, and refinancers evaluating rate timing.

Implementation: Following Poppe’s six-step framework precisely: define the project purpose (increase applications and funded loans), upload NAR research, Bank of America Homebuyer Insights, Zillow market data, CRM demographics by borrower segment, past campaign performance by channel, Google review exports, and rate promotion calendars from prior years. Set Claude’s role as a “promotional strategy consultant specializing in mortgage financing” with explicit guardrails including “never recommend generic rate messaging” and “always address the specific anxiety of each borrower segment.” Use Claude’s Research tool to pull current market condition analysis, then run intake prompts by borrower type to surface motivators, trust blockers, and decision drivers grounded in the uploaded research sources.

Expected Outcome: Campaign briefs are segment-specific from the start — not a single “spring home buying season campaign” but distinct strategic documents for first-time buyers, move-up buyers, and refinancers. Rate promotion windows get matched to historical funnel behavior data, reducing spend during demonstrably low-intent periods. Customer review language gets incorporated into ad copy and email subject lines, improving resonance without requiring expensive testing cycles to discover what language actually connects with each audience.

Use Case 4: E-Commerce Fashion Brand Managing Quarterly Inventory Campaigns

Scenario: An online fashion retailer operates on four major promotional cycles per year: winter clearance, spring new arrivals, summer sale, and fall collection launch. Each cycle requires campaign strategy across email, social, and paid channels, with different inventory velocity priorities by product category that shift based on what sold through and what didn’t in the prior season.

Implementation: The Claude Project gets loaded with: SKU-level sell-through data by season aggregated by category, customer cohort analysis from the CRM covering new versus repeat buyers and category preferences, email engagement benchmarks by segment and campaign type, social performance data by content format and creative approach, and inventory forecasts for the upcoming season by category. The role prompt defines Claude as a “fashion e-commerce promotional strategist with expertise in inventory-driven campaign sequencing” with a standing instruction to “always prioritize inventory velocity and margin impact alongside brand positioning, and flag recommendations that may conflict with current stock levels.” Opening prompts ask Claude to identify which customer segments respond best to clearance messaging versus new-arrival storytelling, and which channels drove the highest repeat purchase rate in the prior comparable season. The system then generates a sequenced promotional calendar that layers inventory priorities with segment-specific messaging windows.

Expected Outcome: Promotional sequencing improves materially — clearance messaging leads with high-inventory SKUs in channels and segments most responsive to discount framing, while new-arrival messaging is timed to peak engagement windows for the brand’s highest-LTV customer cohort. Email and paid channel strategies stop operating in silos because both are drawing from the same AI-synthesized intelligence about what drove revenue in the prior season. Each quarterly push feeds data back into the project, so the system develops an increasingly accurate model of which segments drive margin versus volume in each seasonal context.

Use Case 5: Healthcare System Running Annual Open Enrollment Campaigns

Scenario: A regional health system runs annual open enrollment marketing for its insurance plans. The buying window is fixed — typically six to eight weeks — the audience is tightly segmented by age, health status, and prior plan history, and the regulatory constraints around what can be communicated and how are significant. Prior campaigns have relied heavily on one or two experienced team members who understand the compliance requirements, creating capacity and knowledge-transfer risk.

Implementation: Following Poppe’s compliance-first guidance, the team works with legal to determine what member data can be shared with external AI platforms, and opts for masked, aggregated CRM segments rather than individual-level records. They upload: past enrollment campaign performance by channel and segment, consumer research on healthcare plan decision drivers by age cohort drawn from publicly available sources, de-identified member personas developed in collaboration with the compliance team, competitor plan comparison data from publicly available filings, and CMS regulatory guidelines relevant to plan marketing communications. The role prompt defines Claude as a “healthcare enrollment marketing strategist with expertise in regulatory compliance” with a hard-coded instruction to “flag any recommendation that may conflict with CMS marketing guidelines before proceeding.” Prompts focus on messaging angles by age cohort and on the critical distinction between current-member retention messaging versus prospective-member acquisition messaging, which require different regulatory treatment.

Expected Outcome: Campaign briefs arrive with compliance flags built into the strategic recommendations, reducing legal review cycles and eliminating the back-and-forth that historically consumed two to three weeks of pre-campaign time. Messaging by age cohort reflects the uploaded research on what actually drives healthcare plan selection at different life stages, replacing marketing department assumptions with documented evidence. The system improves with each enrollment season as de-identified performance data gets added back to the project — and the institutional knowledge that previously lived in two team members’ heads now lives in a documented, transferable AI project.

The Bigger Picture

Poppe’s six-step framework lands at a moment when the AI marketing tooling landscape is rapidly consolidating while simultaneously expanding into new functional territory. As Martech.org reported, Canva recently acquired both Simtheory — an agentic AI collaboration platform — and Ortto, a customer data and marketing automation platform, with the explicit strategic goal of becoming a system where “AI becomes the connective tissue between creative, data and execution.” That framing maps almost exactly onto what Poppe’s framework is operationalizing at the team level: AI as the connective tissue between campaign intelligence, strategy, and execution — not as an isolated content generation tool that starts fresh with every session.

The direction of the market is clear: AI in marketing is moving from single-task tools — generate this email, write this ad — toward integrated workflow systems that maintain context across the full campaign planning lifecycle. The teams adapting to this shift now, by building persistent AI projects loaded with structured data and establishing feedback loops from campaign performance, are building a compounding operational advantage over teams still using AI as a one-off content tool.

The AI martech news roundup from Martech.org cataloging over 60 product launches and partnerships between March and early April 2026 reflects this acceleration. Autonomous agents, generative engine optimization, and agentic marketing platforms dominated the coverage — all pointing toward AI systems that operate with increasing autonomy, cross-session context persistence, and integration depth across the marketing technology stack. The category is moving fast enough that teams building campaign planning workflows today should architect them to accommodate agent-driven execution tomorrow, not just AI-assisted planning.

There’s a deeper signal in the data infrastructure piece worth naming explicitly. The email ROI measurement gap — where only 46% of marketers can measure promotional email returns despite 78% acknowledging email’s organizational importance — reveals a structural problem: many marketing teams don’t have the data infrastructure to feed AI systems properly in the first place. Poppe’s framework implicitly requires organized, accessible campaign performance data to exist before the AI system can do its job. For teams without clean campaign dashboards, structured CRM exports, or organized past campaign performance data, building this AI workflow functions as a forcing function for cleaning up their data operations. That side effect is not a small thing — it’s arguably as valuable as the AI capability itself, because it creates the measurement infrastructure that all future AI-powered marketing depends on.

The practitioner shift this points toward is important to state plainly: successful AI marketing is increasingly a data management discipline as much as a prompting discipline. Teams treating AI integration as purely a front-end workflow change — better prompts, faster copy generation — will hit a ceiling quickly. The frameworks that compound in value are the ones that treat structured, organized, continuously updated data as the core asset and AI as the reasoning and synthesis layer built on top of it.

What Smart Marketers Should Do Now

1. Audit your campaign data before you build any AI project. Before opening a Claude Project or any equivalent AI workspace, inventory what data you actually have and how accessible it is: past campaign dashboards, CRM exports, customer research documents, brand guidelines, competitive intelligence files. If these don’t exist or aren’t organized, your first project is not the AI setup — it’s the data audit and organization. The quality of your AI-powered campaign planning is a direct function of the quality of data you can put into it. Teams that skip this step get expensive generic outputs that look like AI-generated content because that’s exactly what they are.

2. Define a role and a standing set of guardrails for every AI project you build. Poppe’s five-component prompt structure — role, context, task, output, considerations — is the minimum viable prompt architecture for strategic marketing work. The “considerations” component is where most teams leave value behind. Write explicit standing instructions that live in the project setup, not just individual prompts: “always start with a strategic hypothesis before tactical recommendations,” “never recommend a tactic without connecting it to a specific audience segment,” “always flag recommendations that conflict with our established brand positioning.” These guardrails convert a general-purpose AI workspace into something that thinks like a senior strategist who knows your business.

3. Build one vertical-specific Claude Project per client or product line, not a general-purpose AI workspace. The power of this framework comes from deep context specificity. A single Claude Project that knows everything about your retail client’s past three years of seasonal campaigns — including what failed, what the post-campaign notes said, and what the CRM showed about segment behavior — is worth ten times more than a general-purpose AI workspace with no persistent memory of your specific business. Resist the temptation to consolidate everything into one project. The performance gap between generic AI context and specific, structured AI context is the gap between outputs you use and outputs you rewrite entirely.

4. Establish a campaign results feedback loop as a formal process, not an intention. Poppe’s Step 6 — iterating as results come in — is the step most teams will skip under post-campaign pressure to move to the next project. It is also where the compounding value lives. Build a standing process: within two weeks of every campaign wrap, a designated team member uploads performance data to the AI project using a standardized template. That template should include at minimum: email performance metrics by segment, paid media results by campaign and audience, CRM conversion data by funnel stage, and any new customer review or survey data collected during the campaign period. This is the step that turns a useful AI tool into an intelligence system that improves over time rather than plateauing after the first cycle.

5. Resolve your internal AI data policy before you build, not during. Poppe explicitly flags compliance as a prerequisite — not a step you revisit if something goes wrong. Before loading CRM data, customer records, proprietary financial data, or competitive intelligence into any external AI platform, get explicit written sign-off from legal and information security on what’s permissible under your organization’s policies and any applicable regulations. In regulated verticals — financial services, healthcare, insurance — plan to operate with masked or aggregated data, compliance-reviewed synthetic personas, or indexed data structures that avoid exposing raw records. Build your AI project within those constraints from the beginning. Attempting to retrofit compliance after the fact creates legal exposure and may require rebuilding the project architecture entirely.

What to Watch Next

Claude Projects and competing persistent-context AI workspaces are worth monitoring closely through Q2 and Q3 2026. Anthropic’s project feature is currently the most mature implementation of persistent-context AI for marketing campaign planning, but OpenAI, Google, and emerging vertical-specific tools are all building toward equivalent functionality. By Q3 2026, expect at least two of the major platforms to announce natively integrated CRM connectors that will make data loading significantly more accessible than the current manual upload model — lowering the barrier to entry for teams without dedicated data engineering resources.

Canva’s integration of Simtheory and Ortto will be worth watching as it progresses from acquisition announcement to actual product integration. If the combined platform delivers on the stated promise of connecting AI-assisted creative with customer data and campaign automation in a unified workspace, it represents a significant workflow consolidation opportunity for marketing teams currently operating across three or four separate tools. Watch for product announcements in late Q2 2026 on how these acquisitions surface in the Canva product interface and what the data connectivity model looks like in practice.

The autonomous marketing agent category, reflected in the 60-plus product launches tracked in the April 2026 AI martech roundup, is moving from experimental to deployment-ready faster than most enterprise marketing teams are prepared for. Track which platforms are announcing native campaign execution capabilities — the ability for AI agents to not just plan campaigns but actually deploy them across channels with defined guardrails — as this is the next major workflow frontier after AI-assisted campaign planning matures.

Regulatory developments around AI-generated marketing content in financial services and healthcare deserve close attention through the remainder of 2026. Both sectors are in active regulatory evaluation on disclosure requirements for AI-assisted campaign content. Teams in regulated verticals should track FTC guidance updates and any sector-specific regulatory signals, as compliance requirements could materially reshape how AI campaign planning workflows are structured and documented.

The measurement infrastructure gap identified in the email ROI research will likely drive a wave of investment in marketing analytics platforms capable of feeding structured performance data into AI systems. Watch for announcements from major CRM and marketing automation vendors around native AI project integration — specifically live data connections that enable real-time context updates rather than periodic manual uploads. When that capability lands at scale, the feedback loop in Poppe’s Step 6 becomes dramatically more powerful and significantly less labor-intensive to maintain.

Bottom Line

Ed Poppe’s six-step AI workflow framework, published via Martech.org on April 10, 2026, is one of the most operationally concrete AI marketing methodologies to surface in this cycle — precisely because it treats data infrastructure as foundational rather than optional. The core insight is that AI-powered seasonal campaign planning is not primarily a prompting problem; it is a data curation and system design problem. Teams that build persistent AI projects with structured data inputs, defined strategic roles, and closed performance feedback loops will compound advantage over teams running AI on generic prompts with no campaign memory and no feedback mechanism. The measurement gap documented in concurrent research — where only 46% of marketing teams can measure promotional email ROI — will close fastest for teams treating AI integration as a forcing function for better data operations, not a shortcut around them. Start with one vertical, build the data layer first, define the role and guardrails before the first prompt, and close the feedback loop after every campaign. The teams doing this rigorously in 2026 will have an increasingly difficult-to-replicate intelligence system by 2027.