From Marketing Cloud to Agentforce: The POET Project Lessons

A few years ago, I led the architecture on a major enterprise transformation project we called POET — PODS Order Entry Transformation. The goal was to modernize an outdated order entry system by migrating it to Marketing Cloud, integrating it with the broader Salesforce ecosystem, and setting the foundation for the platform capabilities we’re now seeing in Agentforce and Data Cloud.

It was the kind of project that looks clean on a slide deck and gets complicated the moment real people, real data, and real business processes are involved. We got it done. But the lessons I took away from it have shaped every enterprise implementation I’ve worked on since.

I’m not going to share proprietary details about the organization or the specific system. What I will share are the universal lessons — the things that apply whether you’re migrating to Marketing Cloud, deploying Agentforce, or running any large-scale Salesforce transformation.

1. The Technical Problem Is Rarely the Hardest Problem

POET was technically complex — multiple system integrations, data migration from legacy platforms, custom order entry workflows that needed to be rebuilt in a fundamentally different architecture. But the hardest parts of the project had nothing to do with technology.

The hardest parts were:

Getting five business units to agree on a single process definition
Convincing stakeholders that “how we’ve always done it” was not a requirement
Managing the gap between executive expectations and implementation reality
Keeping momentum when the project hit its first (and second, and third) setback

If you’re planning an Agentforce deployment, the same dynamic applies. The technology works. What will determine your success is whether your organization can agree on what the agent should do, who owns the decisions, and what changes to existing processes are acceptable.

Lesson for Agentforce: Before you design topics and actions, get alignment on the business process the agent is automating. If stakeholders can’t agree on how a process works today, an AI agent isn’t going to resolve that disagreement. It’s going to amplify it.

2. Technical Debt Is a Tax on Every Sprint

The legacy system POET was replacing had accumulated years of technical debt — custom scripts nobody understood, integrations held together with scheduled batch jobs and prayer, data models that had been extended so many times the original structure was unrecognizable.

Every sprint, we spent time working around this debt. Every estimate was wrong because we’d discover another undocumented dependency. The debt didn’t just slow us down — it made the work unpredictable, which eroded stakeholder confidence, which made everything harder.

We eventually made the decision to allocate 20% of every sprint to debt remediation. Not new features — just cleaning up what was in the way. It felt slow at the time. It was the decision that saved the project.

Lesson for Agentforce: If your Salesforce org has accumulated technical debt — complex validation rules, deeply nested Flows, undocumented Process Builders, fields that are used in ways nobody can explain — you need to address that before or during your Agentforce implementation. Agents interact with your org’s configuration. If that configuration is fragile, your agents will be fragile.

Audit your Flows, clean up your automation, and document your data model. Budget time for it. It’s not optional.

3. Phase the Rollout. Then Phase It Again.

Our original POET timeline called for a big-bang migration. We learned quickly that this was not going to work. We restructured into three phases:

Phase 1: Core order entry for one business unit. Minimal scope, maximum learning.
Phase 2: Expand to three more business units with adjustments based on Phase 1.
Phase 3: Full rollout with integrations to downstream systems.

Phase 1 took longer than planned. Phase 2 went faster because of what we learned. Phase 3 went faster still. The total project duration was roughly the same as the original big-bang plan, but with dramatically lower risk and higher quality.

Lesson for Agentforce: Deploy your first agent for one use case, one channel, one business unit. Get it right. Then expand. The companies that try to launch five agent use cases simultaneously are the ones calling me to help fix things six months later.

4. Train the Architects, Not Just the Users

Most Salesforce projects treat training as an end-user exercise while assuming the technical team will figure things out. On POET, our architects needed structured training on Marketing Cloud’s architecture before they started building. We invested two weeks in enablement. It prevented months of rework.

Lesson for Agentforce: Agentforce is a new paradigm — topic-based routing, action definitions, grounding in Data Cloud. Your architects need to understand these concepts deeply before they design solutions. “Figure it out as we go” produces agents that work in dev and fail in production.

5. Measure What the Business Cares About

Early in POET, our dashboards tracked technical metrics — deployment velocity, code coverage, defect counts. Useful for the technical team. Meaningless to executive sponsors.

We restructured reporting around business outcomes: order processing time, error rates, user adoption, and customer-impacting incidents. Suddenly, the executive team was engaged.

Lesson for Agentforce: Your executives don’t care about deflection rate in isolation. They care about cost per interaction, customer satisfaction, and resolution time. Define these metrics before you launch. If your agent handles 10,000 conversations but customer satisfaction drops, that’s not a success story.

6. The Handoff Problem Is Always Bigger Than You Think

POET required handoffs between the platform team, integration team, data team, and business analysts. Every handoff was a potential point of failure.

Requirements would be documented by the business team, interpreted differently by the platform team, and implemented in a way that didn’t match what the integration team expected. We solved this with structured handoff ceremonies — short working sessions where both teams walked through the deliverable together and documented open questions. It eliminated 80% of our rework.

Lesson for Agentforce: Agent implementations involve handoffs between business, data, platform, and governance teams. Define your handoff process explicitly. Don’t assume alignment — verify it.

7. Done Is Not Done Without Feedback Loops

The real work started when actual users were processing actual orders in the new system. The first two weeks after go-live produced more actionable insight than the entire testing phase. Real users found edge cases our test scripts never covered.

Lesson for Agentforce: Your agent will say something unexpected in week one. What matters is whether you have the feedback loops to detect issues quickly and the operational capability to fix them. Build monitoring from day one. Review agent conversation logs weekly. The agent you launch is version 1.0 — plan to ship version 1.1 within two weeks.

The Through Line to Agentforce

POET was a Marketing Cloud transformation. Agentforce is an AI agent deployment. The technology is different. The organizational challenges are nearly identical.

Both require:

Stakeholder alignment before the technical work starts
Technical debt remediation to build on a stable foundation
Phased rollout to manage risk and capture learning
Architect enablement to ensure the right patterns are used from the start
Business-focused metrics to maintain executive engagement
Clear handoff processes between teams
Feedback loops to iterate quickly after launch

The companies that succeed with Agentforce are the ones who bring the discipline of proven enterprise delivery to a new technology paradigm. That’s what POET taught me, and it’s the approach I bring to every engagement.

Planning a Salesforce transformation and want to learn from someone who’s led them? Schedule a free consultation — I’ll share what I’ve seen work and help you avoid the mistakes that cost time and money.