The next phase of AI in customer experience goes past the chatbot.
I listened to a very interesting conversation last week (the conversation is older but it just crossed my radar): The future of customer experience: Embracing agentic AI. McKinsey explains agentic AI as software that can make decisions and take action on a customer’s behalf, such as changing a plan or issuing a refund, without waiting for a person.
That changes the main question a company has to ask. With a chatbot, you ask whether the answer was right. With an agentic AI customer experience, you have to ask who is accountable when the AI acts.
Table of Contents
What makes AI agentic
A chatbot answers questions. An agentic AI does the next part: It acts. McKinsey’s description is software that is goal-driven, can make decisions, and take action on behalf of the customer — not just hand back information. The example in the discussion is a personal AI agent that picks up in one second, knows your history, and can book the upgrade or sort the problem on the spot.
What matters here is where those agents operate. They are not staying inside the support queue. The same kind of agent can guide a customer through onboarding or run a renewal outreach. As the McKinsey discussion puts it, these agents are not just handling support; they are selling and building the relationship too.
Once AI can act across the whole relationship, the old picture of a bot that deflects tickets is far too small.
What an agent actually does across the relationship
In post-sale, an agent might notice an account going quiet and draft the check-in for the owner to review before it goes out. Catching a customer who has hit the limit of their plan, it can tee up the upgrade conversation before anyone asks. And right after a customer has a win, the agent can queue a congratulation comms, a nomination for an internal award, or request for a reference.
None of these are support tickets.
They are moves across the relationship, the kind of work that used to sit in five different teams, now within reach of one owner who decides which of the agent’s suggestions actually go out. So, the same technology that answers a billing question can also act on renewal or an upsell.
Pointed at the whole relationship, an agent is not a help-desk tool, it is a junior team member doing first drafts of the relationship work, with a person deciding what is good enough to act on.
Acting on a customer raises the stakes
A wrong answer from a chatbot annoys a customer. A wrong action from an agent costs something real: a refund that should not have gone out, or a promise the company cannot keep. The risk moves from a bad reply to a bad outcome.
That is why McKinsey frames this moment as the move from AI curiosity to AI accountability. The pressure now is to show measurable returns, with the CFO asking what the spending actually produced. When an agent can act, accountability stops being optional. Someone has to own what the agents do.
What’s working today, and what’s stuck
McKinsey found that almost every company is doing something with AI, but most of it has not scaled. For chatbots and voice bots, about 95% of efforts are still stuck in pilots and only 5% are running at scale. The wins with more traction are in copilots and coaching, where the AI feeds a person better information in the moment, and those run further along — at 30% to 45% at scale depending on the use case.
One example McKinsey gives is coaching for frontline staff. The AI spots where someone is struggling with a certain type of issue and pushes them short targeted lessons over a few months. The results they report are the ones an executive cares about: handle time came down, and first-contact resolution went up, so customers didn’t have to call back. The pattern worth taking from this is that the biggest early wins often come from AI helping a person do better, not from AI replacing the person.
McKinsey’s answer: one owner of the whole journey
McKinsey’s own team lands on the same answer. In the discussion, they argue that there needs to be one overall ownership of the entire journey, from where the customer enters to where they end up with a person. The old setup, where a digital team owns the bot and a separate team owns the human conversation, falls apart once agents act across both.
We have made this case for unifying the post-sale experience before, and agentic AI makes it more important than ever. If AI agents are taking actions across success, community, advocacy, loyalty, and expansion, you cannot have five teams each watching their own agents. One team must own the relationship and everything the agents do inside it.
The role this creates
McKinsey is also clear about what happens to the people. The roles that matter move toward relationship management and complex problem-solving, and toward blending functions that used to sit in separate teams. They describe new work such as a customer journey architect and human-in-the-loop roles that oversee what the AI does, stepping in where needed.
That is the role we have been building toward. We call it the SCALE™ professional. SCALE is how we run post-sale as one connected program across Success, Community, Advocacy, Loyalty, and Expansion, owned by one person instead of five teams.
With agentic AI, that owner also directs and supervises the agents, and steps in when an action needs a human. Customer support stays its own team, because fixing a difficult technical problem is a different skill from owning the relationship, but the owner sees its tickets and supervises the agents working there too.
I’ve written more about the kind of role that runs this.
Agents are only as safe as the record they act from
An agent that acts is only as good as the data it acts on, and McKinsey is frank about the state of that data. A support person today may touch seven to ten systems, the data is fragmented, and that is why so many companies do not know who you are when you call.
Worse, McKinsey found that almost 70% of contact-center staff say at least a quarter of what they do is written down nowhere. It lives as folklore passed between people or buried in old call transcripts.
Point an agent that takes action at that mess and it will act on the wrong picture, fast. So, two things have to come first. One is a single, current record of the customer that every agent and every person works from. The other is a clear set of guardrails: what an agent may do on its own, and when it has to hand off to a human. McKinsey describes the same need as compliance layers and constant checks on accuracy.
Agentic AI is not an IT project
The fastest way to fail at this is to treat it as a technology project. McKinsey is blunt about it: An agentic AI rollout run as a pure IT implementation is, in their words, more or less bound to fail. The technology must be secure and has to connect to your other systems, and that is real work for IT. But whether it actually helps a customer, and whether there is a return on the money, is a business question, and it stays with the business.
What works is shared ownership of the outcome. McKinsey points to companies that tie the AI’s goals to real business results and hold both the technology side and the operations side accountable, rather than letting AI sit in a corner of IT. The question they suggest that leaders ask is what AI should own, and how you measure whether it is doing that job.
If you are the executive now responsible for this, the trap to avoid is buying a tool and assuming the customer experience will sort itself out once IT installs it, because it will not.
How to start with agentic AI
McKinsey’s advice for getting going is refreshingly plain: Start small and learn quickly, then build toward the bigger picture over time. The warning that comes with it is just as plain. Do not treat this as a technology project, and do not believe a vendor who says their tool solves everything. The hard part is changing how people work, not buying software.
On top of that, assign one owner to the relationship before you turn agents loose. Point the agents at low-risk actions first, where a mistake is cheap to fix. Get the shared record and the guardrails in place early. And measure the whole customer, not each team’s own score, so you can see what the agents are actually doing to the relationship. The aim is not a big-bang rollout but a few agents doing useful, low-stakes work well, with a person who can see and undo what they do.
Agentic AI is going to act on your customers whether the org chart is ready or not. The companies that come out ahead will be the ones that decide, on purpose, who owns the customer when the AI acts. That owner is a person, supported by agents and accountable for the whole relationship. The good news is you do not have to get all of it right at once. Begin with one owner and a short list of safe actions, and let the agents take on more as they earn the trust.
If you are working out how to bring your post-sale teams together, or how to put AI agents to work without losing control of the customer relationship, we can help. Advocacy Maven has built more than 300 customer programs across the full SCALE range: Success, Community, Advocacy, Loyalty, and Expansion. See how we work or read what our clients say about the programs we built with them.
Summary: Agentic AI Customer Experience Needs One Accountable Owner
The shift to an agentic AI customer experience changes AI from answering questions to taking action on behalf of customers. As AI agents begin influencing onboarding, retention, advocacy, loyalty, and expansion, the success of an agentic AI customer experience depends less on the technology itself and more on accountability. Companies that scale an agentic AI customer experience successfully will pair AI agents with one owner responsible for the entire customer relationship, supported by shared customer data, clear guardrails, and human oversight.
An agentic AI customer experience is not an IT project or a chatbot upgrade. It is an operating model that combines automation with relationship ownership. Organizations that establish a single accountable owner, start with low-risk actions, and expand AI responsibilities gradually will be better positioned to capture the benefits of an agentic AI customer experience without losing control of the customer journey.
FAQ 1: What is an agentic AI customer experience?
It is a customer experience where AI does not just answer questions but takes action on the customer’s behalf, like changing a plan or sorting a billing problem, with a person accountable for the result. McKinsey describes agentic AI as software that is goal-driven and can make decisions and act, across the whole journey rather than only in support. The practical difference from a chatbot is that an agent can change something in the real world, so the experience is faster but the stakes are higher. Because it acts, the payoff shows up in fewer dropped balls too: the renewal that gets prepped on time, or the unhappy customer who gets reached before they walk. That is also why it needs one clear owner, someone who decides what the agents may do and steps in when an action needs a human. Done well, the customer gets quick, capable help that still feels like one company that knows them.
FAQ 2: How is agentic AI different from a chatbot?
A chatbot responds; an agentic AI acts. A chatbot can tell you the refund policy, while an agent can actually issue the refund or change the booking. McKinsey frames it as the move from generative AI, which creates text and answers, to agentic AI, which is goal-driven and makes decisions on the customer’s behalf. The upside is that more gets resolved in one go, without a handoff to a person and a wait. The catch is that a wrong action has real consequences — not just a wrong answer — so an agentic setup needs guardrails and a person who owns what the agents do. In short, a chatbot talks and an agent does, and the doing is what raises both the value and the risk.
FAQ 3: Who should own the customer relationship when AI agents are involved?
One person, or one small team, should own the whole relationship, with the agents working under them. McKinsey makes this point directly, arguing for one overall ownership of the entire journey rather than a digital team that owns the bots and a separate team that owns the human moments. In our model that owner is the SCALE professional, who carries success, community, advocacy, loyalty, and expansion as one job. Customer support stays its own team, because solving a hard technical problem is a different skill, but the owner sees its tickets and supervises the agents there too. The reason ownership matters so much with agentic AI is that the agents take action, and an action with no owner is how a small automation becomes a public mess. The point is that no agent acts in a vacuum: a named person is accountable for what happens to the customer.
FAQ 4: What are the risks of agentic AI in customer experience?
The main risk is an agent taking a wrong action quickly and at scale, like issuing refunds it should not or making promises the company cannot keep. That risk grows when the data is poor, and McKinsey notes that agents often have to pull from seven to ten systems, with a lot of what staff actually do written down nowhere. So the failure mode is an agent acting confidently on a fragmented or out-of-date picture of the customer. The other risk is less dramatic: most efforts never get past the pilot stage, because a tool got bought and then treated as the finish line. The way to keep both in check is a single record everyone works from, and clear limits on what agents can do without a person. Agentic AI pays off when those controls come first, not after something goes wrong.
FAQ 5: How should a company start with agentic AI?
Start small and learn quickly, then build toward the bigger picture, which is the path McKinsey recommends. Pick a few low-risk actions to automate first, where a mistake is cheap to undo, rather than turning agents loose on everything at once. Get two things in place before you scale: one shared record of the customer, and clear guardrails for what the agents may do. Give the relationship a single owner so someone is accountable for the agents’ actions from day one. Treat it as a change in how people work, not just a piece of software, because the technology on its own will not fix a fragmented customer experience. And keep it out of the bucket of projects that get handed to IT and forgotten; the business has to own the outcome.
FAQ 6: Does agentic AI replace customer-facing teams?
Mostly it changes them rather than replaces them. McKinsey describes the work moving toward relationship management and oversight, with new roles like human-in-the-loop and customer journey architect, as routine production goes to the agents. The people who stay are the ones making the judgment calls the agents cannot and owning what the agents do. So, the headcount question is less about cutting people and more about who directs the agents and who is accountable when they act. In a SCALE setup, that is the one relationship owner, with support keeping its own technical team alongside.
