Back in December, I attended The Heart of Service Summit in London. Since then, Salesforce has continued to push the boundaries of its AI offerings. At this point, it seems fair to say that Agentforce is quickly becoming a central part of Salesforce’s value proposition.
Fast forward to the Agentforce Service Summit, and the momentum only seems to be growing. As always, the day was packed with real-world case studies, hands-on demos and exciting roadmap updates. Want the key highlights and takeaways? Read on!
Headline Stats 📈: Agentforce in Action
Hearing about real-world Agentforce use cases is always valuable – especially when they’re backed by clear, measurable results. Here are some of the standout stats shared during the day:
- Bionic: Achieving operational savings by reducing Average Handling Time (AHT) by 25%, while also enabling 24/7 customer support through Agentforce.
- Finnair: Streamlining onboarding with a 25% reduction in employee ramp-up time.
- Heathrow: Reporting an impressive ~95% response accuracy rate in Agentforce responses.
- Salesforce: Help and Training’s Agentforce Agent has handled nearly 500,000 conversations, averaging 32,000 per week. The agent is achieving a resolution rate of ~84%, with just 4% of conversations requiring escalation.
- Wiley: Delivering a 213% return on investment, 40% case deflection and approximately $230,000 in savings.
These examples span multiple industries, but the golden thread is clear: operational efficiency and personalised experiences, whilst freeing up teams to focus on higher-value, complex, and empathetic interactions.
Lessons Learned 💡: Insights
The stats are impressive, but they’re only part of the story. Here are some of the key takeaways for businesses starting their Agentforce journey:
- Cut through the hype: Generative and agentic AI are progressing rapidly through the hype cycle. Stay grounded, keep your goals clear and iterate with intent.
- Evolve your agents. Consider a phased approach. Start small (e.g. answering simple knowledge base questions), before opening up to more data and eventually taking action.
- Keep it simple. Generative and agentic AI solutions are non-deterministic – they may produce different outputs from the same inputs. The more complex your agent becomes, the more rigorous your testing needs to be.
- 💡 – Avoid spinning your wheels – keep your agent lean and focused. For context, Salesforce’s Help and Training Agent uses just 7 Topics and 8 Actions.
- Use the right tool for the right job. You wouldn’t use a sledgehammer to crack a walnut, so don’t use Agentforce for simple, deterministic processes. Instead, apply it where it adds real value (e.g. high-volume processes, with variable inputs, such as Tier 1 cases).
- Testing is key. Build fast, test more. Tools like Testing Centre in Agentforce 2dx can help streamline this process by testing both assertive logic and response meaning in bulk.
- Context matters. Your AI strategy must reflect your business, it’s risk appetite, and customer expectations, Tailor your approach to the need.
- Balance the challenges. Don’t ignore broader concerns, such as environmental impact, output accuracy, cybersecurity and intellectual property risks, etc. Consider how this impacts your implementation.
- Coach, but don’t control. Guardrails are important, but being overly prescriptive can lead to undesired results. Instead, think of it as coaching your agent; refine it over time.
In short, start with a clear use case, keep your design focused, test thoroughly and coach your agent to success.
Product Roadmap 🧭: Highlights for Service Cloud & Agentforce
The Agentforce Service Summit offered a valuable look at the innovations on the horizon. As always, these are forward-looking statements. This means they are subject to change and fall under Salesforce’s Safe Harbour Policy. Do not base purchasing decisions on these. If a feature catches your eye, be sure to check out the official release notes or speak with Salesforce directly for the latest updates.
Here are a few roadmap items that stood out for me:
- Export Knowledge to PDF (Winter ’26 or later). Allow teams to generate PDFs of your knowledge articles.
- Enterprise Knowledge (GA June ’25). Use Data Cloud to connect, ingest and unify knowledge content across platforms like SharePoint, Google Drive, Confluence, websites and more. This can enhance Agentforce’s grounding and response quality.
- Service Assistant Enhancements. Service Assistant is a feature which generates step-by-step guidance to resolve a customer issue. Further enhancements are upcoming.
- Actionable Service Assistant (GA June ’25). Let the assistant automatically act on the plan, saving team members more time to focus on complex steps.
- Adaptive Service Assistant (GA Oct ’25). Automatically adjust steps based on real-time context to ensure the plan continues to reflect the customer’s needs.
- Self-Learning Knowledge (Pilot/Beta June ’25; GA Oct ’25). Analyse support data (e.g., cases, interactions, etc); identify trending topics and gaps. Use this to generate content updates to improve knowledge for grounding.
- Knowledge Blocks (Winter ’26 or later). Centrally managed snippets of knowledge content which can be reused across articles as needed (i.e. one-to-many).
- Knowledge Feedback Management (Beta June’25; GA Oct ’25). Easily manage feedback on knowledge articles at scale.
- Omni-Channel Unified Routing for Voice (GA H2 ’25). Extend Omni-Channel routing capabilities to include voice, with skill-based routing and consistent cross-channel agent experiences.
From the upcoming updates, it’s clear that knowledge management and data enrichment will be a major focus. These improvements won’t just enhance self-service experiences; they’ll strengthen Agentforce by delivering rich data needed to generate accurate, personalised, outputs.
Reflection 🔄: What’s changed since the Heart of Service Summit in December?
So, what’s changed since December’s Heart of Service Summit? More than you might think!
- More Case Studies. While Agentforce is still new, we’re now seeing real-world use cases emerge. Some businesses are shifting from exploration to operationalisation.
- Shifting Challenges. Core issues like data quality, privacy, and security remain, but others are also surfacing, such as how model choices impact sustainability. As adoption grows, these priorities will continue to shift.
- From Generative to Agentic. In December, Agentforce had only just launched. Now, Agentic experiences are front and centre – moving beyond just generating responses to intelligent, action-driven agents. This shift is becoming more apparent, but we must remain mindful of the hype curve!
- Development to Testing. Building Agentforce agents can be fast but testing and refinement is where the real investment begins. Ensuring confidence, trust and performance will make the Agentic Application Lifecycle a priority for businesses.
- The Rise of Data Enrichment. High-quality, connected data is becoming increasingly critical for successful AI solutions. Features such as ‘Enterprise Knowledge’ highlight how Data Cloud is key for grounding Agentforce with accurate, relevant inputs.
One thing that hasn’t changed: the need to keep humans in the loop. Whether for escalation, complex problem-solving, or moments requiring empathy or understanding, Agentic AI should be seen as an enabler – freeing people from lower-value and repetitive tasks to focus on higher-value activities.
If there’s one takeaway from this event, it’s the continuing rise of Agentic AI and the growing clarity around how businesses are starting to unlock its potential; a trend we’re likely to see continue this year.
Summary
The Agentforce Service Summit made it clear we’re entering a new phase of AI in customer service – one that goes beyond automation and into intelligent, personalised and adaptive support. The conversation around Agentic AI is shifting from awareness and ‘what’s possible’ to ‘what’s happening’; and perhaps more importantly, ‘what’s been learned’.
Given the pace of change, we can expect continued evolution throughout the year. If you’re embarking on your Agentic AI journey, the prevailing advice is clear: understand your use case, start small, don’t overcomplicate, prioritise testing, refine and keep humans in the loop.
If you found this post helpful and want to stay abreast of upcoming features and releases, be sure to follow SFDC Penguin on LinkedIn and X. Until next time, thanks for reading! 🐧
Thank you
Thank you to Salesforce for hosting the Agentforce Service Summit and to all of the speakers who helped to make the event insightful!
Bonus Penguin Fact 🐧
The news this week has been dominated by talk of Trade Tariffs. However, it was unexpected that penguins would feature so prominently!
Tariffs have been applied to Heard Island and the McDonald Islands, a territory of Australia in the Southern Ocean. The islands are uninhabited by people, but are a valuable wildlife haven, home to a diverse array of species, including of course, penguins! The islands are known to be important breeding sites for Macaroni, Eastern Rockhopper, Gentoo and King penguins.
Learn more about Heard Island and the McDonald Islands here!
Want More?
Found this article useful? Why not share!
Or check out some of our other articles.