McKinsey Has a Phrase for This: Next Best Experience

You're Pushing Your Customers Away

A few days ago, an old friend who runs a consumer brand came over for tea.

He told me his marketing budget hadn't been cut by a single cent, yet results kept getting worse. Texts went out and fewer and fewer people opened them; campaigns launched and almost nobody took part. Then he asked: is my marketing broken?

I said, maybe it isn't your marketing that's broken — it's your understanding of your customers.

He looked puzzled. What do you mean?

So I told him about a phrase — a concept McKinsey has studied for years and written about at length. It's called Next Best Experience. In plain English, it means giving each customer the best possible next experience.

First, let's be honest about how customers are treated today.

The answer is two words: pushed.

Sales is chasing quotas, blasting discount codes into group chats; support sits at the call center waiting for customers to come to them; marketing blankets the calendar with campaigns every quarter. Three teams, none of them on the same page — and whoever feels the most performance pressure shouts the loudest.

So put yourself in the customer's shoes. Here's how one ordinary day goes:

At three in the afternoon, the research team sends a survey. At four, the loyalty team pushes an "enable auto-pay now" prompt. At five, marketing slides into your inbox with the latest bundle offer.

And the person whose day was already interrupted? By that afternoon, they're tapping "I agree to unsubscribe from everything."

That's not an experience. That's harassment.

So What Is It, Really?

Next Best Experience — I like putting it in the plainest possible language:

Before every single interaction, answer one question for the customer: "Right now, what do you actually need?"

In the old model, every move a company makes starts from "what do I want to sell."

This turns that on its head: first think about what the customer needs at this exact moment. Was that overcharge refunded? Was that last call properly followed up? Settle those first, let them calm down, and only then hand over the one thing that's genuinely useful right now.

That single step beats a hundred discount pushes.

And I'm not pulling these numbers out of thin air. McKinsey gave me three figures, and I've carried them with me ever since:

• Customer satisfaction can rise 15% to 20%;
• Revenue can grow 5% to 8%;
• Service costs can drop 20% to 30%.

A company that pulls off all three at once will flip its attitude toward the support department overnight. Support stops being a cost center. It becomes a cash cow.

The Same Customer, Two Different Lives

Concepts get dry pretty fast, so let me tell you about one customer — let's call him Qiang. He's a member of a health insurance plan, and he has diabetes.

Before this system existed, Qiang's life went like this:

A glitch in his reimbursement hit him with a transaction fee. Furious, he called support to complain — it took three calls before anyone sorted it out. And that very same day, the company pushed him an invitation to "join the diabetes management program." Still smarting, he swiped it away.

Then he forgot to pay his premium and got hit with a late fee. When the end-of-month experience survey arrived, he closed it without reading a word.

Do you see the pattern? At every touchpoint, the company handled one isolated problem and moved on. Nowhere was there a process that treated Qiang as one whole person.

Now, what does that same story look like under Next Best Experience?

A bill that looks wrong gets flagged instantly; the system refunds the overcharge automatically — and adds a coffee voucher as an apology. A claim with an error in it is flagged early, and someone follows up to help him double-check. That diabetes program no longer wears its "recommended for you!" sales face; instead, it shows up on a calm day, framed as "let's manage your blood sugar and save a little on your renewal," and it reappears in front of him.

Qiang filled in the survey. This time, he did it happily.

The company gets a better hand too: fewer calls coming in, fewer repeat complaints, and customers who would have churned are staying.

That's what I mean by "give a little, take a little — both sides win."

There's one heavier word hidden in Qiang's story: customer lifetime value.

What is customer lifetime value? It's not what today's single order is worth. It's the sum of every choice and every action that customer will make over the next three years. Every good thing you do for your customers eventually lands in that ledger.

Four Boxes, One Engine

Philosophy done; now the craft. To make this run, you need an engine. If you prise it open, the engine turns out to be four boxes.

Box one: the foundation of data.

Sweep every trace of the customer into one place — bills, transactions, app taps, support call logs. Whip that into a single "data lake," cleaned and tidied, shaped into tables you can query at any moment. Plainly: one spreadsheet holding several years of customer history.

Box two: three predictive models.

They answer three questions: will he churn? Should he be upgraded? Which channel should this message travel through?

These predictions don't live in silos. Take a customer flagged "high churn risk": he is immediately pulled out of every promotional list and placed on a "retention list" — retention perks, service fixes. Take another with "low churn risk, high upsell potential" — the system hands him a gentle upgrade invitation instead.

Box three: GenAI does the talking.

Who talks to the customer? The AI composes the message.

Crucially, it's not slotting in templates but carrying the conversation's context: it knows, for example, that it has just refunded money to Qiang, so its copy naturally says "We've rechecked your billing — have another look …" It knows the right emotional register for the moment. And the voice follows the channel: short and sharp in text, courteous in email, unhurried in the chat window.

One notch further up is Agentic AI, which drafts several variants, guesses how the customer will react, then self-corrects — growing from "fill the template" to "write truer every time."

Box four: getting the message out.

Put words out from the right channel at the right moment: text, email, in-app notification. And where it crosses to a human: your agents' screen, when a customer calls, already shows AI's suggestion.

Four boxes in place — the engine is online.

Three Real Scenarios, Let the Numbers Talk

Principles alone leave you with a question mark. So here are three real scenarios.

Case one — a payments company: predicting who's about to loosen their grip

A global payments provider fears exactly it: the giant merchant walking out the door.

It built a "data twin" for each merchant and ran it through a model to predict one thing: will this merchant's business dip in the next seven days? Then it sorted them out: those with early signs of churn got priority service and a policy cushion; those with headroom but tight cash flow got new features and a buffer to climb on.

The number that came out: merchant churn fell by up to 20% in a year.

Case two — airline, compensation that fits the passenger

Before AI, a delayed flight meant the standard coupon for everyone.

With it, the system first reads who's on the line — is this a frequent business flyer, delayed for the third time this month, or a once-a-year holiday traveler?

The former gets offered compensation with real heft, and a genuinely apologetic handling. The latter, a standard coupon.

The numbers: the ability to identify and win back the customers worth keeping — up 210%; customer satisfaction — up 800%; the share of high-value customers "thinking about leaving" — down 59%.

That's not treatment discrimination, that's respect. A customer of many years and a passenger on his first trip were never supposed to receive the same care.

Case three — telecom, learning to hear "leave me alone"

A European telecom made a firm rule: as long as a customer has an unresolved complaint, or sits somewhere inside the after-sales flow, no promotional contact with that customer — ever.

Get the issue settled first, then talk business. "Resolved" ranks above "sold."

The result: customer satisfaction immediately caught up with the market's number one.

Harder than Tech — Even McKinsey Has Scars

"Fine on paper, but is it really doable? Honestly?"

Yes — it's hard. But not primarily on the tech side.

McKinsey built an internal AI assistant for its people, called Lilli. On paper, this should have been a technical project — build, train, deploy.

But in practice, more than half of the effort on this "tech project" actually went into "getting people to use it": weaving it into daily workflows, training staff, changing habits. The model itself was the smallest piece.

I filed that away: no matter how good the model, it only counts when people trust it and actually use it.

So don't rush out to buy systems. Sort out the two human questions first — "Do the teams believe it?" and "Will they pick it up?" — and only then does the road open up ahead of you.

Six Warm-Up Moves to Get In

You don't have to start big — you just have to aim right. Here's McKinsey's checklist, compressed to six moves:

1. Run a pilot. Pull a few key datasets (CRM, billing, operations) into a small lake and run a pass — you'll learn fast whether your data is complete and clean.
2. Pick a quick win. Don't try to build the whole suite at once. Choose the single most valuable scenario — churn prediction, or an upsell — and carry one model all the way through it.
3. Stock the parts. MLOps, DevOps, MarTech — no need to buy them all up front, just enough to serve.
4. Wire the teams together. Build a cross-functional group with one unified "contact rule": marketing, sales, and support all answer the same discipline for the same customer.
5. Keep a control group. Set up two tracks from day one and track them side by side — every change becomes a number immediately.
6. Walk on two legs. Run a small pilot project so wins show up at the morning stand-up, while the big pieces — the data lake, cross-functional governance — get cleaned up in parallel.

Six moves, but one thread runs through them all: win trust with small victories, then fill in the big gaps.

A Final Word

That afternoon, my friend was quiet for a long minute, then said: "And this — will it win back the customers we've already lost?"

I said: they may be gone for good — but at least this lets you watch a customer starting to lean toward the door before he even knows he's inclined.

Actually, few companies are even on this path. And among them, even fewer have built the four boxes and taken it all the way to the end.

There's one last thing I'd leave with you:

Know what your customer wants, before he asks for it.

May you understand your customer's next move sooner than the customer does — and may the next experience you give them be the best one yet.