Dreamforce 2026 Session Recap: From Good to Great Service Transformation

During Dreamforce week, everyone in service wanted to talk about AI agents. Far fewer wanted to talk about what has to be true before an agent can deliver. That gap was the focus of the Good to Great Service Transformation session Thunder hosted on September 16 at The Fly Trap in San Francisco.

I was joined by Matt Kravitz, VP of Product Management at Salesforce, to walk a room of service and contact center leaders through a maturity framework for getting real business impact from AI. Our core message: great service doesn't start with the agent. It starts with a clear service model and a foundation built one pillar at a time. Here's a recap.

Start with your service model

Before picking channels or tools, you need to know what kind of service organization you are. We grouped service models into three patterns, and each one has a different definition of success:

The model tells you where to put your AI investment. Action-based organizations should focus on agentic self-service actions. Order-based and knowledge-centered organizations get more from augmenting their reps and technicians with contextual guidance, summarization and AI-drafted replies.

One of the most important points of the session: most organizations see themselves in all three buckets, and that's fine. But trying to make more than one of them your major focus leads to stagnation and mediocrity. Pick your primary model and build around it.

The five pillars of service maturity

Once you know your model, maturity comes down to five pillars. Each one has its own three-level journey from good to great.

1. Channel strategy: from presence to intelligence

It starts with one question: why do customers contact us? Identify your top 6–10 contact drivers, then assign each to a deliberate engagement approach: no-touch (self-service), low-touch (digital) or high-touch (voice). Order status belongs in self-service. Product questions may deserve a person.

This prompts a useful debate about "could vs. should." Just because you can deflect a contact doesn't mean you should, especially where humans add real value.

The maturity path runs from Channel of Choice (every channel offered for every intent, so customers guess which is fastest) to an Intelligent Front Door (ask "why" first, then offer self-service and the right channels) to an Agentic Portal that doesn't just recommend a channel but offers to complete the action. Our advice: dream big, but start small. Launch on your portal, replicate the logic in messaging, then scale to voice.

2. Information architecture: from library to engine of action

Most help centers are organized by product (the Manual) or by category (the Library). Both put the burden on the customer. Topic-centric sites are easy to browse, but they hand customers a "reading assignment" instead of a solution.

The most mature organizations are intent-centric. They navigate by what the customer wants to do — track a shipment, start a return, update a payment method — and launch an automated flow or agent right away. A practical tip from the session: use an LLM to draft a working baseline of your intent-based IA.

3. Search experience: from results to resolution

The customer journey often starts with search, whether on Google, your site, your help center or a GPT. Search has evolved from results (a list of keyword-matched links) to answers (direct, grounded responses based on meaning) to conversation (multi-turn reasoning that can take action or hand off with context). Complex organizations, Salesforce included, typically need a hybrid of all three to serve diverse audiences.

4. Help agents: from deflection to resolution

Help agent maturity follows three A's. Take a customer asking, "What's the status of my order?"

1. Answer: a generic response from knowledge ("Go to My Account and look for the Order section").
2. Access: a contextual answer grounded in authenticated data ("Your order has shipped!").
3. Action: orchestrated resolution ("Do you need to return or change your order?").

Each contact driver may call for a different level. The real question is where you sit today on the spectrum from answerability to actionability.

5. Omni-presence: from siloed bots to an agent workforce

Many organizations start with an AI agent in a single chat widget. Expanding to more surfaces often creates a walled experience: separate bots with separate logic, so nothing carries over between text and voice. The goal is a unified workforce: one reasoning engine that orchestrates actions across channels, portals and voice, so a customer can start on one channel and finish on another.

Where do you land? The 15-step report card

We closed by bringing it all together in a single report card. Across the five pillars, each of the three levels is a step forward, for 15 steps in total:

Matt also shared how Salesforce is putting this into practice with Casey, its help agent on help.salesforce.com, which reached a 70% autonomous resolution rate as of June 2026. It's a strong proof point for what's possible when the foundation is in place.

The biggest takeaway? Going from good to great isn't about adding more AI. It's about knowing your service model, maturing each pillar on purpose, and moving every contact driver from deflection to true resolution.