We’re coming up to three years since AI landed in the mainstream consciousness with the release of ChatGPT. But with the promise of this technology changing everything, where are we really after multiple years of hype?
In this year’s edition of IBM’s annual State of Salesforce report, we see a comprehensive reality of the challenges that businesses implementing AI within the Salesforce ecosystem are facing. This also provides a picture of what actions organizations should take to drive success leveraging this truly game-changing technology.
The State of Salesforce
IBM’s State of Salesforce report is one of the largest in the entire ecosystem, interviewing 1,200+ Salesforce customers to paint a picture of the biggest challenges and opportunities that the businesses of today face.
Previous editions of this report have focused on the disconnect between Salesforce and siloed data, implementing GenAI, and taking advantage of cross-cloud and industry cloud implementations.
It’s no surprise to learn that the contents of this year’s report are based around the practical implementation of AI. No hype, no wild numbers, just the reality.
Earlier in the year, I predicted that Agentforce would be a slowburner, and the fact that its pricing model is usage-based means that customers have to see ROI before they scale up. This is the reality that Salesforce are facing, and although they would like to see much faster growth, customers aren’t necessarily ready.
With only 33% of AI initiatives meeting ROI, 62% worrying about unpredictable costs, 21% feeling they have the right agentic AI governance, and only 26% saying they have most customer data in Salesforce, there are many organic issues that the Salesforce ecosystem must consider.
These statistics back up exactly what has happened since Agentforce was announced over a year ago: a lack of product ROI, a swathe of implementation issues (such as disparate or dirty data), and a backlog piling up with technical debt fixes.
But regardless of current issues, the future is AI, and with the impact it is already having on the job market, we can assume that there is much more disruption to come.
So what can be done? Let’s dive in.
1. Focus on Outcome-Based ROI
A big problem with agents and AI is that the technology can do almost anything. This entire technological transformation is unlike anything that came before.. Sure, there was the cloud computing craze, but this was “simply” moving from on-premise to cloud servers, still dealing with a database.
On the flip side, agents and AI are a totally new and emerging technology, and until you start experimenting to understand what is possible, it’s hard to understand the potential.
Although AI can be anything you want it to be, there are defined use cases that most organizations are familiar with. In the State of Salesforce report, IBM outlines the top front office and back office investments in terms of ROI (percent saying investments have somewhat or significantly increased because of AI):
Front-Office
- 70% Customer service
- 65% eCommerce
- 63% Sales enablement
- 56% Contact centre support and service
- 51% Distribution channels
Back-Office
- 64% Marketing automation
- 61% IT (platform and service management)
- 57% Security and compliance monitoring
- 55% Revenue operations
- 53% Field service
However, implementing these use cases is just half the battle – measuring ROI is the other half. IBM suggest a simple approach as part of their report to get started:
- Give every AI a “job description” tied to a risk-adjusted ROI.
- Decide what this agent is supposed to accomplish and what the performance baseline is.
- Agree on the expected benefits.
- Map out what success will look like at 30, 90, and 180 days.
In other words, you need to start treating AI like an employee. Most humans don’t hit the ground running when they join a new company, they need to be effectively onboarded, and AI is no different.
2. Use Governance to Go Faster (Yes, Really)
Governance, risk, and change management are vital in the corporate world of AI for the enterprise.
Agents and AI can make some executive teams nervous, and for good reason: an out-of-control AI can cause reputational damage to your business, such as selling cars for $1 each. As a result, nearly three in four respondents said that digital labour raises the need for risk management, but only 21% strongly agree they have the governance needed. There is clearly a disconnect.
Interestingly, IBM also reports that only 16% of executives have confidence in AI, a figure which remains unchanged from last year’s report.
This year’s report recommends a mixture of human intervention, setting up guardrails and risk thresholds, logging and monitoring, as well as documentation to quell concerns.
Salesforce solutions to some of these concerns may be…
- Make “human‑in‑the‑loop by default” your policy for customer‑facing automations.
- Gate risky actions (discounting, refunds, field updates, etc.) behind approvers.
- Log agent decisions, monitor for drift, and assign owners for intervention.
- Document prompt policies and access control scopes per agent.
3. Agents Aren’t the Starting Point – Your Data Is
“Data is the new oil” was an analogy coined by mathematician Clive Humby in 2006, and his prediction couldn’t be more true in 2025. AI is seen as the technology of our generation, contextual data is the key to unlocking the true power of large language models.
This is not a new concept or idea to most organizations, but the dirty data issue is becoming a major barrier to companies adopting this technology. Only 26% of people say most customer data sits in Salesforce, and 53% cite poor data availability/quality as the top adoption barrier for agentic AI.
When companies connect external platforms (think Adobe/SAP/Oracle) to Salesforce, they’re 2.7x more likely to achieve a strong 360° view and are 2x more likely to drive cost savings with Salesforce.
Connecting mainframe data improves prediction accuracy (up to 2.6x) and boosts cost outcomes as well. For a $19B company, that “connected” posture maps to roughly $140M in additional revenue plus ~2% implementation/development savings.
Final Thoughts
This year’s State of Salesforce report by IBM is focused on cost-effective agentic AI at scale, and this topic is exactly what the industry needs to read, contemplate and action
Although there have been countless research showing the effectiveness of AI, or in some cases ineffectiveness of AI, there is very little practical implementation advice and guidance.
I hope this post is a good first step as well as the report, to better understand how to leverage AI and can help organizations to maximize value and success in using AI.
