If you read our scene-setter on the Board vs. The Floor tension — or watched the full video interview with Salesforce's SVP of Agentforce Contact Center — you already understand the scale of the problem. Board pressure, operational reality, and a 95% pilot failure rate sitting between them.
This piece takes the conversation further. Because knowing the gap exists is only useful if you know how to close it.
Why most AI deployments fail (and it's not what you think)
The instinct when a contact center AI deployment underperforms is to blame the technology. But three failure modes are far more common — and far more fixable.
"The challenges come from three failure modes. AI deployed in isolation. Organisational gaps in knowledge ownership. And the metrics being used to define success."
— Gautam Vasudev, SVP Agentforce Contact Center, Salesforce
The first is isolation: AI deployed without access to the customer data that makes it useful. No purchase history. No case context. No policy knowledge. The second is organisational — fragmented ownership of knowledge with no accountability for keeping it current. Gartner found 61% of service leaders have a backlog of knowledge articles to update and more than a third have no formal process for revising content. In that environment, an AI agent amplifies existing problems rather than solving them.
The third failure mode is measurement. Optimising for deflection rates and declaring that as ROI is the wrong goal. Counting containment while customers quietly escalate through another channel isn't progress — it's a metric designed to tell you what you want to hear.
The data problem nobody raises in the sales meeting
There is one insight CX leaders should carry into every vendor conversation this year: the triple penalty.
"When you ground your AI in stale or badly structured data, you pay a triple penalty. Unsatisfactory responses. Frustrated escalations to human agents. And the risk of hallucination. You have to be very careful about this."
— Gautam Vasudev, SVP Agentforce Contact Center, Salesforce
The practical diagnostic Vasudev recommends is straightforward: map your top 20 high-volume customer intents to your current knowledge articles.
"That gives you a report card — whether you've got an A grade, a B grade, or a C grade in your knowledge coverage."
Most organisations discover they are not where they assumed. That discovery is uncomfortable. It is also the most valuable thing you can do before committing to a deployment timeline.
Why some contact center AI deployments reach production while others never do
Speed of deployment is not determined by the technology decision. It is determined by what happens before it.
"The single biggest differentiator between teams that reach production and teams that don't is predefined success criteria — with alignment internally before a single line of code is written."
— Gautam Vasudev, SVP Agentforce Contact Center, Salesforce
That alignment spans IT, service operations, finance, HR, and increasingly legal. "This is not a hero's journey," Vasudev says. "This is really a collaborative effort." The use cases that move fastest are the ones where the organisation already has a natural advantage: clean data, structured knowledge, high-volume queries with predictable patterns.
Reframing the board conversation
The language used with a CFO matters more than most CX leaders appreciate. "Starting with a pilot" invites scepticism. "De-risking a strategic investment" opens a very different conversation.
"For a CFO, this becomes a conversation of not 'hey, we are starting a pilot.' It becomes 'we are de-risking a broad, strategic, large investment.' And every CFO loves the word de-risking."
— Gautam Vasudev, SVP Agentforce Contact Center, Salesforce