The Phone Call Is the Last Thing Most Businesses Never Wrote Down

On Monday, Salesforce announced a definitive agreement to acquire Cresta, a company whose voice agents handle customer calls end to end for brands like Marriott, Comcast and Alaska Airlines. Salesforce didn't disclose the price. The Information reported it at $2.7 billion. Axios counts it as Salesforce's 12th deal this year, and TNW notes it comes two weeks after a $5.7 billion agreement for the customer messaging company Braze.

Most of the coverage reads this as a voice AI story. I read it as a data story.

What Salesforce actually said

The interesting line in the announcement isn't about voice agents sounding natural. It's this: customer conversations are "too often trapped in silos, disconnected from customer context and the tools teams use to act." The stated goal is to connect every conversation to the customer and to the outcome: resolution, retention and revenue.

Salesforce's own EVP for Service Cloud, Kishan Chetan, put it plainly: "Voice is where service is hardest to get right, and where it matters most to customers."

Read that again as an owner of a 20-person company. The biggest CRM vendor in the world is paying billions because phone calls are the part of the customer relationship that never makes it into the system.

The same gap exists at every size

In almost every business we walk into, the phone is the blind spot. The CRM knows about the web form. The help desk knows about the email. The phone call that actually decided whether the customer stayed lives in one person's head, or on a sticky note, or nowhere.

Then that person goes on vacation, and the customer calls back and has to start over.

None of this needs a voice agent to fix. It needs a decision: every call ends with a record. Who called, what they wanted, what we promised, who owns the next step, and when. If that record doesn't exist, the call didn't happen as far as the business is concerned.

My take for operators

Don't start with the voice bot. Start with the record.

If you put an AI agent on your phones before your calls land somewhere structured, you've automated the silo. The agent will have great conversations that nobody else can see, act on, or learn from. That's the exact problem Salesforce says it's buying Cresta to solve, and it's cheaper to avoid than to fix.

Here's the order I'd follow:

1. Pick the one system where customer history lives. One. Not three.

2. Make every inbound call create or update a record there, even if a person types it at first.

3. Attach an owner and a next date to every open item from a call.

4. Only then add AI: transcription and summaries first, then routing, then agents that resolve simple calls on their own.

This is the same approach we take when we build a business hub for a client. The phone, the inbox, the forms and the tasks all feed one place, and the AI works off that shared record instead of next to it. It's a big part of why we can deploy a full hub in under three weeks now, where it used to take four to six months. The AI is fast. Deciding where the truth lives is the slow part, so we do it first.

Why this matters now

Enterprise vendors are buying their way into voice because it's the last channel they don't own. Over the next year, a lot of smaller businesses will get pitched AI phone agents that promise to answer every call.

Some of those tools will be good. Ask one question before you sign: where does the call go when it's over? If the answer isn't "into the same record your team already works from," keep looking.

The call is where your customer tells you the truth. Write it down.