I've spent ten years building go-to-market teams, and in that time the post-sales org chart barely changed. AI is changing it now, faster than most teams have noticed.

So I asked two senior post-sales leaders the same question within a week of each other: what does a customer success team look like in two years? One is merging Implementation, Support and Customer Success into a single team he calls Customer Outcome Architects. The other thinks the standalone CSM role will be gone by then.

They sound like opposite answers. By the end of this piece, I think you'll see they're mostly the same answer, described from two different directions.

Most of what follows comes out of my own years building go-to-market teams for B2B SaaS, my conversation with Adi Patil on the GTM & AI Operators Show, and a discussion with Mohammad Moiz. Adi leads Solutions Arhitcts at Rocketlane, ex post-sales leader at Clazar. Earlier he scaled implementation at Chargebee through its $7M to $100M ARR run. Moiz is SVP of Customer Experience at SpotDraft. He built the post sales function fro scratch there and now his team manages more than 500+ customers across the globe.

You can watch my conversation with Adi Patil on post-sales in the AI era here.

Your customer stopped caring about your org chart

Customers don't want an onboarding manager, then a support engineer, then a CSM, then an account manager. They want one go-to person, and they want the outcome they bought. Every handoff is a place where context gets lost and nobody owns the result.

"The end customer doesn't care about your org chart. They want one face who can be their go-to person."

Adi Patil, on the GTM & AI Operators Show

Every handoff is a place where context gets lost. Customers want one person who owns the result.

The bar has also moved from adoption to value. A customer paying $100K a year isn't satisfied by a product that simply works anymore. Renewal conversations used to be about usage dashboards. Now they're about what changed in the customer's business, and if you can't show that, the churn conversation starts without you.

And here's something most teams miss. Your customers have bought AI too, and most of them can't put it to work. Everyone wants it, but very few people inside your customer's company know how to deploy it on their own systems. Whoever shows up and makes it work owns the relationship, and that person is increasingly not called a CSM.

What good looks like now: 7 people, 250 customers

Adi Patil's post-sales team is a good picture of what this looks like in practice. Seven people serve over 250 customers across all three functions. Three handle implementations for a base of more than 200 customers. Two support engineers cover the whole book, with first responses under 20 minutes on most verticals.

For context, Gainsight's benchmark is around 49 customers per CSM in a mid-touch model, and that's for customer success alone. This team averages about 36 per person across three functions, and answers in under 20 minutes against an industry median of 47.

Its AI support agent now resolves about 30% of tickets, with more than half as the goal. That can sound low next to vendor claims of 70% and above, but those numbers mostly come from simple, high-volume consumer tickets. This 30% is complex B2B support inside Slack, grown slowly from 5-10%, with the agent's tone tuned so customers don't feel brushed off. That care is the point, and Klarna shows what happens without it.

A small team can cover far more customers once the right foundations are in place.

What Klarna actually teaches

You've probably heard the Klarna story as "they fired 700 people for AI and had to walk it back." That isn't what happened, and the real version is more useful.

In early 2024 Klarna said its AI assistant was doing the work of 700 full-time agents, most of them outsourced rather than on staff. Nobody was fired. A hiring freeze and normal attrition shrank the company by about 40% over two years. Then in May 2025 the CEO said the trade-off out loud: cost had become too big a factor, and the result was lower quality. Klarna started hiring human agents again, while keeping AI on the routine work.

So the lesson isn't that AI failed at support. Klarna measured cost per conversation instead of quality per conversation, pushed too far, and paid for it in customer trust. The slow climb from 5% to 30% is the opposite approach: automate routine work as fast as quality allows, and no faster.

Transform the team, or dissolve the role

There are two ways this can go. The first is to transform the team. Bring Implementation, Support and CS together and measure every function on customer outcomes. Implementation stops being judged on go-live dates and support stops being judged on tickets closed. Everyone is judged on whether the customer got what they bought. Adi calls this team Customer Outcome Architects, and he's building it now.

The second is to dissolve the role. In this view the standalone CSM disappears within a year or two. Commercial work like renewals and upsells moves to account managers. Technical work, deploying the product and AI inside the customer's systems, moves into product and engineering, to engineers who learn the customer's business and act more like product managers than ticket-takers. That's Moiz's forecast. 

The industry already has a name for that second role: the forward-deployed engineer. Palantir invented it, and AI startups have made it one of the fastest-growing roles in tech. So this isn't a fringe view. The market is already hiring for the role that would take over the CSM's technical work.

Either way, what happens next is a race. CSMs already understand the customer's business and need to learn tech and AI deployment. Engineers already know the tech and need to learn the business. Whichever group closes its gap first takes the work.

Here's where I land, and why I said they're mostly the same answer. The job title is the least important part of either view. A Customer Outcome Architect and an FDE are the same person seen from two sides: someone who understands the customer's business and can put AI to work on their systems. A CS leader who builds his own support agents and churn alerts is already winning that race from the CSM side. Whether the org chart says CSM, COA or FDE in two years, the people with both skills will hold the job.

Whoever closes their skills gap first, CSM or engineer, takes the work.

Your team has all the context now. Where do you go from here?

A year ago the common complaint in CS was that account information lived in ten different places. For a lot of teams, that's now solved. Connected tools pull the data together. At SpotDraft, a quarterly business review deck that used to take three to four days now takes about five minutes. That's one team's experience rather than a published benchmark, but nothing published comes close.

Solving that created a problem I haven't seen anyone write about yet.

"Earlier the problem was I didn't have context. Today, context is so readily available it has become a problem in itself."

Mohammad Moiz, SVP of Customer Experience, SpotDraft

When every tool surfaces everything, the skill stops being finding information and becomes filtering it: knowing what to act on and what to ignore. Being comfortable with data, analytical and a sharp critical thinker used to set a CSM apart. Now it's the minimum, because the job is sorting through what the systems surface.

The fix is the most copyable idea in this piece: a curated context layer. Here's how it works on Adi's team. Claude rebuilt the knowledge base from two years of questions and answers, Slack threads and Notion docs. Every customer has one page, covering how they bought, who the key people are, what outcomes they want, and summaries of every call. A weekly Claude agent keeps each page up to date, and anyone on the team can ask questions of it from Slack through a Notion agent. That weekly update is the filter. The team doesn't read twenty thousand things. It reads one well-kept page per customer.

More context isn't the answer. One well-kept page per customer is.

Pulling all the data together creates the overload. Curating it is what fixes it.

What to build, in order

Build these in order, because each step feeds the next.

1. A clean knowledge base. AI rebuids it from two years of real questions, Slack threads and docs. It's unglamorous and it sets the ceiling for everything after it. An AI support agent is only as good as the knowledge underneath it.

2. One context page per customer. Kept current by a weekly update and searchable from Slack. This turns the knowledge base from a help center into the team's shared memory of each customer.

3. An AI support agent. It answers routine questions in a tone that fits the customer. For B2B teams working in Slack, platforms like Pylon are built for this. Start small and let quality set the pace: the agent in this piece started at 5-10% of tickets, is now around 30%, and is aiming for 50% or more. Quality was never allowed to slip to push that number up.

4. Early warnings for at-risk customers. Built as a Claude Skill, it reviews logins, tickets and touchpoints every week, ranks the customers most likely to leave, and drafts outreach for each one using their context page. A health score tells you who has gone quiet. This tells you who, why it matters, and what to say.

None of this takes an engineering quarter. A team utilization tracker can be built in two or three hours and run every week. Use cheaper, faster models for routine extraction work and the most capable models when you need a thinking partner, matching the model to how hard the task is.

Each layer feeds the one above it, so start with the knowledge base.

The bar for you, personally, has moved

Stop using AI as an assistant. Handing your small tasks to AI is now the starting point, and everyone does it. The real change is redesigning the work from scratch around AI, so it handles all the call prep, summaries and follow-up, and you spend your time on the conversations that matter.

Here's how I think about what this does to performance. Performance used to run on a scale of 4 to 10. AI moved both ends. Everyone now starts at a 4, because anyone with the tools can produce competent work on day one. But the people who redesign their work around AI aren't capped at 10 anymore. They're working at 15. That means the solid 8 who changed nothing has quietly slipped below the new middle.

AI raised the ceiling, so solid work that hasn't changed now sits below the middle.

Research supports the first half of this. A large study of over 5,000 support agents found AI tools raised the productivity of newer agents by 34% while barely changing the output of the most experienced ones. AI lifts the floor. The 15 doesn't come from using the assistant harder. It comes from the redesign.

Here's a simple test if you're reporting on your own AI use. Saying AI helps you with "research" won't hold up much longer. The questions that count are how much time you saved, which workflows you automated and how many agents you have running. Plenty of people still default to search engines, and the ones who push AI tools as far as they go are usually the ones who taught themselves. Nobody is going to train you up to the 15.

SaaS isn't dying. Post-sales is becoming where the revenue lives

You've seen the "SaaS is dead" posts. What Satya Nadella actually said was narrower: business software is largely databases with business rules on top, and those rules are moving to AI agents. That describes a change, not a death. Nobody is building their own CRM from scratch over a weekend, and the software companies that rebuild around AI will keep their customers.

What changes most is how customers pay, and that changes who carries the revenue. Some AI vendors already charge only for results. Intercom's Fin charges $0.99 for each conversation it resolves, and Sierra charges per resolution and makes handing off to a human free. When customers stop paying $100K for a product and start paying $100K for an outcome, the team that delivers and proves the outcome stops being a cost center that protects renewals. It becomes the part of the company that earns the money.

For most of SaaS history, post-sales was what happened after the deal. In an outcome-priced world, it's the reason there's a next one.

FAQs

How is AI changing customer success?

It's changing what the team is organised around. Customers don't want to be passed from onboarding to support to customer success to an account manager. They want one person who delivers the outcome they paid for. AI now handles much of the routine work, from support tickets to call prep, so more teams are bringing Implementation, Support and CS together and measuring all of them on customer outcomes rather than tickets closed or go-live dates.

Will AI replace customer success managers?

AI won't replace the work, but it is changing who does it. One view is that CSMs grow into a broader role that owns the customer's outcome end to end. Another is that the standalone CSM role splits, with renewals moving to account managers and technical work moving to engineers who work inside customer accounts. Either way, the job goes to people who understand the customer's business and can put AI to work on its systems.

What is a good customer-to-CSM ratio with AI?

A common industry benchmark is around 49 customers per CSM in a mid-touch model, for customer success alone. With the right foundations, a team can go well beyond that. Adi Patil's team of seven covers more than 250 customers across implementation, support and customer success, about 36 per person across all three. The ratio improves only once the basics are in place: a clean knowledge base, a context page per customer and an AI agent on routine tickets.

What is a forward-deployed engineer?

A forward-deployed engineer (FDE) works inside customer accounts, learns the customer's business, and builds and deploys solutions on the customer's own systems. Palantir created the role, and AI companies have made it one of the fastest-growing jobs in tech, because many of their customers buy AI they can't deploy on their own. In post-sales, FDEs increasingly take on the technical work that used to sit with CSMs.

What skills does a customer success manager need now?

Two stand out. The first is filtering. Connected tools now surface everything about a customer, so the skill has moved from finding information to deciding what to act on and what to ignore. The second is redesigning your work around AI rather than using it as an assistant, so AI handles call prep, summaries and follow-up while you spend your time on the conversations that matter. Understanding the customer's business, and being able to deploy AI inside it, is what sets people apart now.

Where should a small post-sales team start with AI?

In this order: rebuild your knowledge base from real support history, give every customer one page of context that updates weekly, put an AI agent on routine support, then add early warnings for customers at risk of leaving. Each step feeds the next, so skipping ahead usually means rebuilding later.

How does outcome-based pricing change customer success?

It moves post-sales from protecting revenue to earning it. Some AI vendors already charge per result, such as per resolved conversation, rather than per seat. When customers pay for outcomes, the team that delivers and proves those outcomes is the one the revenue depends on, and post-sales stops being a cost center.