AGENTIC MATURITY: NAVIGATING AI, AGENTS, AND DATA WITH STRATEGY

By Joe Prentis
Salesforce Core Leader, EMEA

I recently had the chance to spend time at the Agentforce Partner Summit, digging into their latest thinking and roadmap. I went in expecting a product update and roadmap. I came away with something much more valuable - a clearer picture of how to actually move forward with AI in a way that’s grounded, realistic, and (crucially) useful.

There’s been so much noise over the past twelve months, from new GPTs and plugins to copilots and assistants. It’s hard to know what’s real, what’s scalable, and what’s just hype. What I liked most about this session is that it cut through all of that noise, giving me a way to think about progress and how best to offer guidance to prospects and clients. Not in terms of features or flashy demos, but in terms of maturity and strategy.

The 5 Levels of Agentic Maturity: From AI Agent to Multi-Agent Orchestration

Agentic maturity is a useful lens because it forces us to look at where we are, not just where we want to be. The Salesforce model breaks things down into five levels:

At Level 0, you’re working with fixed rules and repetitive tasks, automations that follow strict logic with no reasoning. Think workflow rules or simple macros. A lot of organisations have been doing this for a while to become more efficient, and they still operate here.

Level 1 introduces information retrieval agents that can search and surface information, helping humans but not doing much decision-making themselves. It’s essentially AI-assisted search, and it’s where many teams first dip their toe in the water. Think of an FAQ Agent on a support site, where the Agent is retrieving information and performing actions, but in a simple use case.

By Level 2, things get more interesting. Agents start to orchestrate tasks autonomously, but only within a single domain. There is a lot of value in having Agents at Level 2, as they can significantly improve efficiency in many of our routine tasks. For example, consider a customer service team. An Agent at this level could be trained to monitor incoming support emails, automatically identify the issue type based on keywords and sentiment analysis, and then log a corresponding case with preliminary details - all without direct human intervention beyond the initial setup and monitoring. This frees up human agents to focus on more complex issues, leading to faster response times and improved customer satisfaction.

Level 3 is where orchestration spans multiple domains. Agents start to handle workflows that move across teams, systems, or data sources. Harmonised data and guardrails are very important at this level, but the payoff is huge: fewer handoffs, faster delivery, and less admin clutter. Imagine a customer being able to process a return from your website, which automates the entire process including updating your ERP and triggering the return payment to be scheduled. There is also some impressive automation coming in the Marketing space, where agents can jump right into conversations from email replies, opening up a brand new sales channel. Forget sending people to a website or having them call in - we can turn those "Do not Reply" emails into "Please Reply" giving agents the power to upsell and cross sell.

Finally, there is Level 4: multi-agent orchestration. This is where things get genuinely exciting - and complex. You’re enabling agents to work with other agents, across different systems, and across businesses, with minimal human oversight. Currently this isn’t something that is generally available but by the end of year, innovators and leaders will be doing this. Here is an overview of where this innovation is heading.

Source: https://bit.ly/4klvBJG

How to Build an Effective Strategy: AI, Data, Planning & Implementation

Here’s the thing. No one wakes up at Level 4. The companies that get there do so deliberately - by linking strategy, data, and delivery.

And that starts with two things: a clear AI strategy and a solid data strategy.

You don’t need to write a 50-page document, but you do need a shared vision.

What role do you want AI to play in your business?
What problems is it here to solve?

This isn’t just an IT conversation. The best AI strategies are born out of real business challenges. They ask questions like:

• Where are we losing time or value today?
• What decisions are repetitive, manual, or data-heavy?
• Where could intelligence or automation create a better experience - for customers or employees?
• If we were to hire a new person today for any job, what would their role be and what would we want them to do?

Once you’ve got that foundation, it becomes much easier to prioritise use cases and avoid shiny object syndrome.

Then there’s data. Honestly, most AI projects hit roadblocks here. You can’t build great agents on top of messy, siloed, or inaccessible data. If you don’t already have one, this is the moment to build a data strategy that supports your AI ambitions. That means unifying your data sources, ensuring quality and governance, and building infrastructure that can scale. AI without data strategy is like trying to make Michelin-star food with ingredients scattered across five kitchens. You need everything in one place, clean and ready to go.

What this doesn’t mean, however, which I know everyone is thinking, is that you need all of your business data to be clean, organised, harmonised, and segmented. Data and AI strategy are intertwined, and your data maturity can grow with your AI maturity. The data that you need for your use cases does need to be clean and accessible, but that can be a small proportion of the whole business and it doesn't have to be a huge exercise. Focusing on low effort, high impact wins across level 1 and 2 can really start you off on the right path with your AI journey, but it will also give you valuable lessons as you develop more capabilities.

Agentforce Updates: RAG and Flexible Pricing

What’s nice is that Salesforce have made a few updates recently which help teams progress up the curve, without jumping straight to Level 4 from a standing start.

One of those updates is RAG (Retrieval-Augmented Generation). This means you can tailor your agents to only access specific business content from help articles to pricing policies, and have them generate responses and actions based on that source material. It’s a really clever way to add domain-specific knowledge, without retraining models. It’s a clear enabler for Levels 2 and 3, and it makes the leap to something more intelligent feel achievable.

The other big shift is flexible agent pricing. Instead of paying per conversation, you’re now charged based on action. It might sound like a small thing, but it’s important. It lowers the barrier to experimentation, allowing you to spin up small use cases, test value, and iterate quickly, without a huge financial commitment.

That’s exactly the sort of flexibility businesses need right now. Because most of us aren’t trying to build futuristic AI empires. We’re just trying to solve problems and work a bit smarter.

Plan Your AI Implementation Roadmap: Next Steps and Key Takeaways

There’s still a long way to go for most businesses and for the products themselves. But what stood out this week is that it is possible to get there, and we don’t need to run before we can walk. With the right strategy, the right data foundations, and the right starting points, it’s entirely doable.

We don’t need another hype cycle. What we need now is practical momentum, grounded in real business use cases. Ones that are solving real problems and adding value to the day to day activities of your business. And that starts by asking: where are we on the maturity curve - and what’s the next step?