Agentic Marketing Operations
Reimagining Trade-Show Marketing

Portfolio summary

A.G.E.N.T. helped me redesign how the work should happen. B.U.I.L.D. helped me turn that redesign into something buildable, testable, governable, and capable of producing measurable business value.

Human-centered agentic AI designed to transform trade-show marketing from a last-minute production task into a continuously prepared, governed, and learning sales workflow.

---------------------------------------------------------

Forty-eight hours before a trade show, the promotional email still isn’t finished.

The products are selected. The show dates are known. The customer relationships are there.

But the information needed to build the campaign is scattered across spreadsheets, inboxes, manufacturer materials, calendars, previous campaigns, and, most importantly, the experience of the person running the business.

For a small manufacturer’s representative like L.M. Marketing & Sales, the work gets done.

But getting it done requires searching, remembering, copying, checking, rebuilding, and coordinating the same pieces over and over.

Therefore, every campaign consumes time that could have been spent talking to retailers, supporting manufacturers, and creating sales opportunities.

That struggling moment became the starting point for a different kind of marketing system.

The Challenge Wasn’t Writing an Email

At first glance, the obvious AI opportunity seemed simple:

Use generative AI to write the email faster.

But that only automates one task.

A successful trade-show campaign requires someone to know:

Which show is coming?

Which brands will be represented?

Which products should be featured?

Which retailers should receive the message?

Which accounts need personal attention?

What product information and imagery are approved?

When should the campaign go out?

What needs to happen after the show?

The deeper problem wasn’t content creation.

It was coordination.

L.M. Marketing’s information existed, but much of it was dispersed across customer lists, trade-show schedules, brand information, product assets, past campaign content, and individual knowledge. The opportunity was to create a structured source of truth that an AI-enabled workflow could reliably use.

So I reframed the problem:

How might we redesign the campaign workflow so AI handles repetitive coordination while people remain focused on relationships, judgment, and selling?

That changed the project from an AI writing experiment into an agentic workflow transformation.

From Tools to a System

The business objective was deliberately narrow: improve the planning, creation, targeting, review, and analysis of trade-show email campaigns.

Not automate everything.

Not replace the sales relationship.

Build one workflow that matters, prove its value, and expand from there.

The strategy identified reduced campaign preparation time, more consistent execution, stronger targeting, and improved engagement as the first measures of success.

The resulting concept was not a single chatbot.

It was a coordinated system of specialized agents.

An Orchestrator Agent manages the overall campaign process.

Supporting agents can retrieve approved brand and product information, analyze retailer audiences, generate campaign content, recommend timing, and evaluate performance.

A Guardian function checks critical details such as dates, links, product claims, brand consistency, contact permissions, and approval requirements.

And then the automation deliberately stops.

The AI can prepare the work.
A human decides whether it is ready to send.

That boundary became one of the most important design decisions in the project.

Designing the Work Before Building the AI

I began with the A.G.E.N.T. framework, which is designed to shift attention away from simply adding AI tools and toward redesigning the workflow itself.

A: Audit

Understand how the work actually happens today.

I mapped where campaign information lived, how retailer lists were assembled, how products were selected, how copy was created, where approvals occurred, and where knowledge depended on memory.

G: Gauge

Determine whether the workflow is genuinely worth transforming.

Trade-show marketing presented a strong candidate because it combines repeatable activity with changing information, research, judgment, coordination across systems, and clear business outcomes.

E: Engineer

Redesign the workflow for agent-first execution.

Instead of asking a human to locate every piece of information and then prompt an AI, the new model makes data accessible so agents can retrieve, analyze, prepare, and coordinate much of the work themselves.

N: Navigate

Define the relationship between people and agents.

What can AI decide?

What requires human review?

Where should the system explain its recommendation?

When should it escalate?

For L.M. Marketing, final decisions involving customer communication, recipients, brand representation, timing, and messaging remain human responsibilities.

T: Track

Measure whether the new workflow actually creates value.

The objective is not “more AI.”

The objective is better marketing operations: less preparation time, fewer repetitive tasks, more consistent campaigns, better targeting, stronger follow-up, and more time for selling.

A.G.E.N.T. gave me the redesigned operating model.

But a workflow diagram is not a working system.

From A.G.E.N.T. to B.U.I.L.D.

The next phase moved from strategy into implementation.

The Harvard Data Science Initiative program separates the two. The Strategy phase uses A.G.E.N.T. to identify and redesign a high-value workflow. The Build phase moves “From A.G.E.N.T. to B.U.I.L.D.”, turning the scoped workflow into a working prototype while forcing concrete technology and architecture decisions.

For this project, that meant progressing through five implementation questions.

B: Blueprint
What must it do?

I translated the redesigned workflow into a buildable specification.

The blueprint defined:

the trade-show trigger
required campaign and customer data
specialized agent responsibilities
orchestration logic
validation requirements
human approval points
campaign execution
the feedback loop after deployment

The important shift was from describing what an AI might do to defining how information, decisions, systems, agents, and people would interact end to end.

U: Unlock
Make or buy?

Once the workflow was defined, the next question was architectural:

What technology should make this possible?

The solution needed a structured data layer for customer and trade-show information, access to approved brand content, AI reasoning and content generation, workflow orchestration, and an email-marketing execution platform.

The architecture was intentionally modular and practical rather than dependent on a custom enterprise application.

That meant thinking in terms of components that could evolve independently instead of creating a monolithic AI system.

I: Iterate
Where does it break?

Then came the part that diagrams hide.

The prototype had to work.

Rather than trying to automate the entire marketing operation at once, I treated the solution as a sequence of experiments:

Build a piece.

Run it.

Find the failure.

Understand why it failed.

Change the workflow.

Run it again.

The friction became evidence.

Data structure, field mapping, prompts, orchestration logic, agent handoffs, and human review were not just implementation details. They revealed assumptions in the original design.

That iteration philosophy also connected directly with my Design Thinking work: make an idea real enough to answer a question, learn from what happens, and refine it rather than waiting for a supposedly perfect solution.

L: Launch
Ready for the real world?

The goal was never uncontrolled autonomy.

The launch model was designed around a human-supervised pilot.

Agents could retrieve information.

Analyze an audience.

Prepare campaign content.

Validate data.

Recommend actions.

But external communication still required human approval.

That created a critical distinction:

“Ready for human review” is not the same thing as “ready to send.”

Human-centered service, accuracy, transparency, privacy, brand integrity, and proportional autonomy were treated as part of the system design, not policies to bolt on after the technology was finished.

D: Drive
Would your CEO believe it?

Finally, the test is not whether the agents performed interesting tasks.

It is whether the system creates a credible business outcome.

For L.M. Marketing, that means measuring:

Campaign preparation time
Does the workflow reduce repetitive coordination?

Consistency
Are campaigns prepared and executed more reliably?

Audience relevance
Are retailers receiving messages better aligned to territory, product interest, and relationship status?

Engagement
Are more retailers responding, clicking, attending, or requesting appointments?

Follow-up
Are fewer opportunities falling through the cracks after a trade show?

Human capacity
Is the sales team spending less time assembling campaigns and more time building relationships?

The Drive phase turns a technical prototype into a business conversation:

Is this useful enough to expand?

And if it is, the same foundation can extend into appointment tracking, lead prioritization, post-show follow-up, customer re-engagement, and additional sales-support workflows.

The System Prepares. The Human Makes the Call.

The difference between the original and future-state experience is significant.

Before

Search inboxes.

Open spreadsheets.

Find the show information.

Locate product assets.

Rebuild the audience.

Draft the email.

Check dates.

Check links.

Review the message.

Schedule the campaign.

Remember who needs follow-up.

Repeat.

After

The system recognizes the upcoming campaign.

Retrieves the relevant business information.

Identifies the appropriate audience.

Prepares approved product and brand content.

Generates the campaign.

Validates critical details.

Presents the work for human review.

Executes only after approval.

Captures performance.

Feeds what happened back into the next campaign.

The future-state vision keeps people in authority while allowing AI to coordinate, recommend, prepare, and monitor.

Designing the Guardrails

Giving software more agency creates a new quality question.

Traditional systems ask:

Did the software perform the required step?

Agentic systems force a different question:

Did the system make an acceptable decision within the boundaries we established?

So governance became part of the architecture.

The solution was designed around:

Human accountability
People retain responsibility for customer-facing decisions.

Accuracy
Dates, products, links, and claims must use current approved information.

Transparency
Recommendations should be inspectable and changeable.

Privacy and consent
Customer preferences, opt-outs, and permissions must be respected.

Brand integrity
Generated content must accurately represent both L.M. Marketing and its manufacturers.

Proportional autonomy
The system earns greater independence only as reliability is demonstrated.

Continuous improvement
Corrections, exceptions, and campaign outcomes become inputs into future iterations.

The point was not to eliminate judgment.

It was to determine where judgment belongs.

What I Designed

Role
Agentic AI Solution Lead

Disciplines
AI Strategy · Workflow Redesign · Design Thinking · Agent Orchestration · Solution Architecture · Human-in-the-Loop Governance · Prototyping · Quality Strategy

Program Context
Harvard Data Science Initiative, Agentic AI Strategist + Builder

Frameworks
A.G.E.N.T. · B.U.I.L.D. · Human-Centered Design

Solution Pattern
Structured Data → Orchestration → Specialized Agents → Guardian Validation → Human Approval → Campaign Execution → Performance Feedback

The Real Transformation

The breakthrough wasn’t:

AI can write a marketing email.

We already know it can.

A.G.E.N.T. forced a more interesting question:

How should this work happen if AI agents become active participants in the workflow?

B.U.I.L.D. forced the next one:

Now, how do we actually make that system work in the real world?

And that led to an even more useful question:

What happens when AI can prepare the work around the email?

When the system can gather information before someone asks for it.

When it recognizes the next trade show approaching.

When it can prepare the audience and locate the right product information.

When it can draft, validate, organize, and recommend.

When it can learn from the previous campaign.

And when the person who once spent hours assembling all of those pieces can instead spend that time asking:

Is this the right message for this customer, right now?

That is the transformation I was designing for.

Not replacing the relationship.

Making more room for it.