You Think You're Doing "Omnichannel." To Your Customer, You're Just an Amnesiac.

Let me tell you something that happened to me.

Last week, I spotted an ergonomic chair I liked on a furniture brand's app. I added it to my cart, stared at it for twenty minutes, thought better of it, and closed the app.

The next day, I opened my laptop, ready to buy. The website's homepage pushed — a sofa. The day after, I scrolled past their ad in my social feed — a bookshelf. The day after that, their customer service called me: "Sir, would you like to hear about our mattresses?"

I told the agent, "I was looking at a chair."

A beat of silence. "Oh, I can't see your browsing history on my end."

One brand, four channels, four greetings — and every single one treated me like a stranger.

Good lord, this is 2026, and this is what most companies call "omnichannel marketing."

One: What Does "Omnichannel" Actually Mean?

A lot of people think "omnichannel" means you've opened a Weibo account, a TikTok account, a mini-program (WeChat's in-app applets), and a private-domain group (owned WeChat-style communities).

Wrong.

That's "multichannel," not "omnichannel."

The difference between multichannel and omnichannel is the difference between "this company has five branches, and none of them knows you" and "this company has five branches, and the employee at any one of them can call you by name and tell you what you bought last time."

One-sentence definition: the core of omnichannel is "identity continuity."

When a customer moves between your website, your app, your emails, your support line, and your physical stores, they shouldn't have to reintroduce themselves — and your systems shouldn't treat them like a stranger.

Sounds simple? Almost no one pulls it off.

McKinsey has a number: 76% of customers expect brands to deliver personalized cross-channel experiences, but only 33% of companies actually do. That 43-percentage-point gap is where the opportunity lives.

Two: The Four Parts of AI-Powered Omnichannel

So what does AI actually do here?

I'll break it into four parts. Once you see them, you'll get it.

Part One: Identity Resolution

The foundation.

A single customer might register for your membership with one email, browse your site anonymously as a string of cookies, call customer service and give a phone number, and place orders in your app using a WeChat login.

To your systems, those are four different people.

The first thing AI has to do is recognize "the same person" — using deterministic methods (email and phone number matching) and probabilistic methods (device fingerprints, behavioral patterns) to stitch scattered records into one complete profile.

Without this step, nothing else matters.

Part Two: Real-Time Signal Capture

Old-school marketing was "user clicked X, push them Y." Write the rule once, let it run for a year.

The new omnichannel is "I know this user's mood, device, context, and intent — right now."

He spent three minutes in your app today without buying. He just received your marketing email and didn't open it. Right now he's crammed into a subway car with a flaky signal.

Your system has to catch all three of those signals in milliseconds and make a call: "Should I interrupt him right now? On which channel? With what message?"

Real-time doesn't mean "the next email." It means "the decision this second."

Part Three: Predictive Decisioning

This is where AI actually earns its keep.

It scores every customer: likelihood to buy, likelihood to churn, response probability for this offer, the best channel, the best moment.

But here's the big trap — the quality of the model depends on the quality of the data.

Plenty of companies spend a fortune on the most expensive AI platform, but the underlying data is dirty, broken, and scattered, so the model spits out nothing but wrong answers. It's like spending three million RMB on a top-spec sports car and filling the tank with gutter oil.

Part Four: Cross-Channel Orchestration

This is the part most easily overlooked.

Every action across every channel has to be coordinated by a single brain. The customer taps "I want to learn more" in the app, and a second later the support side sees a pop-up. The customer adds an item to the cart on the website, and a second later the app's homepage changes.

If you can't pull this off, the customer concludes: "This company's left hand doesn't talk to its right hand. Their AI is window dressing."

Three: Why Do 80% of Projects Fail?

Let me make a bet with you — if your company is about to launch an AI omnichannel project, eight times out of ten it will fail.

Not because the tech isn't good enough. Not because the budget is too small.

Because you think this is a marketing project, when it's actually a data project.

What do I mean?

I've seen too many companies start by comparing platforms, comparing features, comparing prices, and then pick a two-million-RMB SaaS. Six months after launch, conversion rates haven't budged — and complaint rates have actually gone up. Why? Because the underlying data is fragmented: one record in the CRM, another in the support system, another in the website backend, and none of them talk to each other.

AI running on top of shattered data produces "personalization" that's worse than no personalization at all.

Take this example. A customer returned a product in your app last week. The AI doesn't know it (because the return data never flowed back), so the next day it pushes a pile of similar products at him. The customer opens the app and thinks:

"I already returned it, and you're still pushing it at me?"

That's "amnesia-style personalization" — worse than doing nothing, because it actively damages the customer.

Data governance, identity cleanup, feedback loops — this unseen grunt work is the real foundation of AI omnichannel.

Four: Four Numbers That Show How Much This Is Worth

Some numbers to get you excited.

• A beverage company called Hydrant used AI to predict customer churn and lifted conversion rates by 2.6x.
• Companies making real-time AI marketing decisions see conversion rates 20% higher than peers on average, with customer acquisition costs 15% lower.
• Companies using AI for support see satisfaction scores rise 15–25% within six months, and ticket volume drop 20–30%.
• Companies that do omnichannel well see customer retention climb 3–7% and revenue growth of 5–15%.

Tempting numbers, right?

But behind every one of those numbers is a team that spent at least six months on data governance.

It's not that they picked the right platform. It's that they built the foundation right.

Five: An "80% Solution" Checklist For You

You don't have to start with a multi-million-RMB plan.

If your team is small and the budget is tight, here's a lightweight stack that will get you 80% of the upside:

1. CDP (Customer Data Platform) — gather the data and unify identities first. This is the foundation; there's no skipping it. Open-source options include Apache Pinot; commercial options include Segment and Shence (神策).
2. Journey orchestration tool — handles trigger logic and multi-channel distribution. Braze or Insider both work; pick the one that fits your channel mix.
3. Web personalization engine — A/B testing plus dynamic content. Optimizely, VWO.
4. Customer support system — connect support to the customer profile, so agents can pull up "who is this customer" in a second. Intercom, Zendesk.
5. Data feedback loop — interaction data from every channel has to flow back into the CDP, continuously feeding the model.

This five-piece stack can be maintained by a team of two to four people.

And when I say "maintained," I don't mean "launch it and you're done." I mean every quarter you retrain the model, clean the data, and fix identity stitching.

Omnichannel isn't a project. It's infrastructure. And infrastructure demands long-term investment.

Six: One Final Counterintuitive Judgment

Having written this far, I want to tell you something counterintuitive.

Over the next three years, the companies that actually win at omnichannel won't be the ones with the most advanced AI models — they'll be the ones with the most solid data governance.

AI tools are getting more open. Models are getting more homogeneous. The model you can buy, someone else can buy too. The prompt you can write, someone else can copy.

But is your data clean? Is your customer profile complete? Is your feedback loop smooth? This unseen grunt work is the real moat.

Back to that furniture brand that drove me crazy.

I later asked their COO about it. He said, "We'd love to do omnichannel too, but the data is scattered across five systems. Unifying them would take six months, and right now running campaigns is more urgent."

I heard that and thought — if you don't do this today, by next year the cost will only have doubled. And the year after that, the customer will probably already have been poached by whichever competitor "did the data governance first."

So: build the data foundation first. Then we can talk about AI.

That's the plainest piece of advice I can give to every head of marketing in 2026.