Your CRM was built backwards

When I led global customer success at HubSpot, the customer base grew from roughly 70,000 to more than 220,000 in just four years. Growth at that scale stress-tests every assumption you make about your customer.

At that point, the data required to understand a single customer sat in product logs, billing systems, support queues, and the notes a customer service representative typed based on memory. All that data created a record, but not an understanding.

AI agents are closing the gap. The CRM software category will not survive in its current form, as it is built on disjointed data that is hard to integrate usefully.

THE FUNNEL WAS AN ORG CHART DRAWN SIDEWAYS

The customer journey has evolved from a funnel to a flywheel and then a loop. Every version has the same defect: It focuses on what the company does, in the order the company does it. But that is rarely what happens.

Here’s what happens. A customer who is three months into a contract has two failed logins, an unresolved billing dispute, and an upgrade offer from someone who doesn’t know about either. In Salesforce’s State of the Connected Customer survey, 55% of customers said it feels like they’re dealing with individual departments rather than a unified company, and 56% said they must repeat themselves to different reps. That’s been tolerable in the past because software could only provide information to a human, and a human can only see one screen at a time.

NOW THE SOFTWARE DOES THE WORK

Tech companies are focused on getting Agentic AI to shift siloed data into databases while developing tools that find the information and process it into an answer. Now clients pay for the result rather than the tool. For example, when using a support agent, the client pays per issue resolved rather than per seat to use the software.

Foundation Capital’s Jaya Gupta and Ashu Garg coined this shift Service as Software and see it as a $4.6 trillion opportunity. Service as Software started in customer support because the function is high-volume, high-value, and has a measurable outcome: resolution. Gartner expects agentic AI to autonomously resolve 80% of common service issues by 2029, cutting operating costs 30%.

But software that does the work has to know the customer, and that requires rebuilding these five stages of the customer journey.

1. Discovery: The search box becomes a conversation.

For 20 years, product discovery has happened through typing a keyword into a search box that returned a page of results filtered by price. An AI agent consolidates that process, and it becomes marketing’s job to own the conversation.

2. Sales: One relationship, every channel.

People don’t communicate through one channel. We email, call, text a follow-up, and expect the other person to remember all three. Someone asks a question in chat, then gets a form, a marketing email, and a call from someone who never read the chat.

An AI agent that holds the whole relationship carries the thread across every channel, the way the best salespeople do for the customers they know by name. It can also do what no person can do at scale: notice something changed and reach out first.

3. Onboarding: The first week decides the relationship.

The contract was just signed. Everything the customer said before the purchase stays with whoever made the deal. They receive templated welcome emails, and the first time they need help it’s like the first time they’ve ever spoken to you.

An agent can carry the sales relationship forward from day one, know why the customer bought, and set up the account based on purchase decisions.

4. Support: Five hops, three systems, one outcome.

The most complicated AI deployment I’ve seen manages payment support for a gaming operator. Before the agent replies to a player, it’s already classified the intent, scored the urgency, and checked the language against risk categories that include signs of a problem. If it finds a risk, it switches playbooks, routes the case to the human queue trained for it, and writes the summary into the record at close. That’s five hops, three systems, two rulebooks, no keyword rules, and no manual triage.

5. Renewal: The relationship remembers.

Renewals are where every silo shows up at once. Your system pings you about an upcoming renewal at a set date, regardless of what’s happening with the customer. Meanwhile, the customer is still knee-deep in a recurring issue that hasn’t been fixed. An agent that holds the whole record knows the pain points and what’s being done to fix it. The renewal becomes a continuation of the relationship instead of a bill.

DO IT WHOLESALE, OR DON’T BOTHER

Moving to a solution that focuses on outcomes requires a platform that crosses systems, processes, and standard operating procedures, which is multi-hop reasoning. Multi-hop reasoning lets you reason across multiple systems for knowledge, context, and actions and know what’s relevant for each turn of the conversation. It takes user permissions, access, guardrails, and standard operating procedures into consideration.

Many leaders think that means buying an agent for each department from whichever vendor demos best. That rebuilds the funnel with better tooling, but the integration math is against you. MuleSoft’s 2026 benchmark puts the average organization at 957 applications with only 27% of them connected, and 86% of IT leaders say that AI agents add more complexity than value when they run without proper integration.

When you’re buying an agent, ask:

• Does the agent know what happened in the other department’s system?
• Can it act in the system, or only read?
• When it hands off to a person, does the person inherit the context or start over?

THE REWARD IS OUTSIZED

The customer journey was never a sequence of stages. It was a description of our own departments, drawn that way because no system could hold the real thing. What replaces it is one relationship, held in one place, acted on from the first question to the last renewal.

SaaS priced the tool. Service as Software prices the job.

Jonathan Corbin is the founder and CEO of Maven AGI.