26 AI Predictions — I Reconciled the Accounts for You

Late last year, I came across a list of predictions that was making the rounds.

Twenty-six entries, all calls about 2026: how AI would reshape customer experience, employee experience, design, and innovation.

Honestly, my first reaction was: here we go again. Every year-end brings one of these lists — every line bold, every line unfalsifiable.

Now, eight months of 2026 are behind us.

I dug the 26 back out and checked them against reality, one by one.

No hype, no hit job. Some have already landed. Some are coming true in real time. And a few of the ones that looked scariest back then? Also coming true.

The scariest ones, it turns out, were never about how powerful AI would be. They were about how AI would screw things up.

Let's settle them line by line.

Entry One: AI Has Started to "Know" Its Customers

What is hyper-personalization?

It's a storefront that no longer lays out its wares and waits for you to browse. It gets built on the spot, for an audience of one. You click, it changes. It remembers what you bought last time. It knows which page you hesitated on for three seconds.

Starbucks has an AI engine called Deep Brew that serves every user a different set of offers. Alibaba's e-commerce platform assembles a storefront on the fly for every visitor. In the UAE, Emirates NBD pushes tailor-made financial products to each user inside its online banking.

These are not demos. McKinsey ran the numbers: companies that nail the "next-best experience" lift customer satisfaction by as much as 20% — while their service costs keep falling.

Why do customers respond so strongly to this?

An early-2025 McKinsey survey: 71% of consumers expect personalized treatment; 76% get annoyed when they realize a brand doesn't know who they are.

Right? Text a friend and get back "Sorry — who is this?", and you'll know exactly how that feels.

One step further is predictive service — AI moves before you've even opened your mouth. Telcos predict a network failure is coming and notify you in advance; sometimes you never even register that there was a problem. In Singapore, DBS uses models to flag which customers might churn, so relationship managers show up at their door with a plan in hand. Siemens builds AI into its equipment and sends service crews out before anything breaks.

From "you ask, I answer" to "I thought of it first" — that step is the watershed of customer experience.

And conversational AI is becoming the front desk for all of it.

This year, one of Gartner's predictions is coming true: nearly 75% of customer interactions will be powered by AI.

Bank of America's Erica has logged more than 3 billion cumulative interactions. Vodafone's TOBi has handled millions of support requests across multiple countries. At India's HDFC Bank, a bot called EVA handled more than 2.7 million queries in six months, serving 50 million customers.

Good grief. One month of a bot's output is more work than a human agent gets through in a lifetime.

There's a catch here, though — I'll hold that thought and pick it up in Entry Five: when a bot blows it, the cost is far greater than when a human does.

Customer experience also came with a further-out call: the boundary between online and offline would blur. IKEA lets you "place" furniture into your own living room with your phone; Sephora's AR mirrors let you try on shades without touching a thing. Two years ago these were novelties. This year, they're standard equipment.

Customers never wanted "intelligence." They wanted to be remembered.

Entry Two: AI Skills Are the New Century's Driver's License

The employee-side predictions felt the flimsiest to me at the time. Looking at them this year, they're the most solid.

Gartner's call: by 2027, about three-quarters of hiring processes will assess candidates' AI capabilities.

Note this: testing for AI ability will become like "knowing how to drive" — a default requirement, not a bonus.

Around the year 2000, "can use a computer" went from bonus to baseline. AI is replaying that scene right now.

The problem is on the front line. BCG's 2025 workplace research: among managers, more than 75% use AI tools; among frontline employees, only half.

It's not that they don't want to learn. It's that nobody teaches them.

One set of numbers from the same study is especially worth savoring: a few hours of training for employees, plus an explicit signal from leadership, lifts their enthusiasm for AI from 15% to 55%.

A few hours. From 15% to 55%. That is probably the best-value money in this year's entire training budget.

Some companies placed this bet long ago. AT&T has poured more than $1 billion into employee reskilling. Siemens and Rolls-Royce run in-house "AI academies" where everyone takes classes, from engineers to sales.

On the talent-management front, it gets even more brutal.

Unilever uses AI in hiring and cut hiring time by 75%, while the share of offers going to candidates from disadvantaged backgrounds rose 16%. IBM's HR function runs models that predict who is about to quit with 95% accuracy — and retention alone has saved it more than $300 million.

$300 million — saved by a prediction.

There's one more prediction that's easy to overlook: customer experience and employee experience would merge, into a single term — Total Experience. The two were always two faces of one coin: only employees who feel good about their work can serve customers well.

Walmart gave store associates a voice assistant — inventory questions, just ask out loud. easyJet equipped its call-center agents with real-time prompts. Japan's Mizuho connected its customer-service bots to an internal knowledge base, so the hard questions customers ask get distilled, almost incidentally, into learning material for staff.

Gartner said that by this year, 60% of large enterprises would have programs like these.

Want to serve your customers well? The next step is serving your employees first.

Entry Three: The Innovation Cycle, Cut in Half

The design-side predictions have come true most thoroughly.

Adobe's 2024 survey: 83% of creative professionals were already using generative AI. 83% as of 2024 — what's the number this year? No need to look it up.

The results are on display. Netflix and Disney use AI for concept art and storyboards, compressing the pre-production cycle. Coca-Cola uses generative AI for advertising visuals. BMW and Airbus use generative design for parts, cutting weight by 30% to 50% with zero sacrifice in strength. No human brain would imagine these structures — but a machine can compute them, and a human then judges "will this work?"

So don't panic. AI's job is to imagine ten thousand options; the human's job is to pick the one. The division of labor in creativity has changed. Creativity hasn't died.

And here comes the more interesting part: innovation is "leaking" into the hands of ordinary employees.

Microsoft's Power Platform — drag, drop, and an employee who can't write code can build an app. A small fintech in France had employees with zero IT knowledge feed prompts to it, and the customer-service bot was live within weeks. Founders in India use AI to draw up marketing assets and go toe-to-toe with big companies.

Gartner predicted that by this year, more than 80% of enterprises would have used generative-AI APIs or models.

Which means innovation is no longer just the R&D department's job. A store manager can tune an inventory model; a marketing associate can test creative concepts. A company with 3,000 employees has, in theory, 3,000 innovation nodes.

Now layer on the R&D-cycle line: BMW and Boeing say that with AI-driven design and simulation, parts development time was cut in half. PepsiCo uses AI to mine social media and consumer data to settle on new flavors — what used to take round after round of market testing now gets done in a few months. The figure McKinsey and Kearney both observe as typical: time-to-market for new products shortened by 20% to 50%.

And don't forget AlphaFold. Around 2022 to 2023, AI cracked the protein-folding problem, compressing research that used to be counted in years into something counted in weeks.

What does a halved R&D cycle actually mean? They ship 2 new products a year; you ship 4. After two years, you're a full stride ahead. After three, your rivals can't even see your taillights.

In the AI era, slow is expensive — full stop.

Entry Four: AI Takes a Seat in the Boardroom

This prediction read like science fiction when it was written. This year, it reads like news.

An IBM study: 74% of executives believe AI will fundamentally change the way they make decisions.

Banks and insurance companies use AI to stress-test economic environments. Siemens feeds real-time operating data to AI and lets the model recommend which plant should get the money first. Singapore's sovereign wealth fund uses "digital twins" to simulate policy changes — run it through the virtual world first, then land it in the real one.

Note: AI does not decide. AI lays the options, the risks, and the probabilities on the table; humans make the call. But whoever sets the table shapes the game.

On the organizational-process side, it's hyperautomation. Insurance claims: AI reads photos to assess damage, algorithms check the policy, payment goes out automatically, and humans handle only the exceptions. Amazon's warehouses have been running on this logic for a long time. Banks in Europe and Asia use robotic process automation to handle tens of thousands of loan documents and KYC (know-your-customer) checks every day.

Here's some blunt arithmetic: a competitor onboards a customer in one day with AI; your manual process takes a week. A six-fold speed gap — customers vote with their feet, and you won't even get the chance to ask "why."

Then there's the ecosystem angle. BMW, Audi, and Mercedes-Benz, rivals for over a hundred years, now pull in Intel and Mobileye to work on autonomous driving together. Amazon sells the logistics-prediction tools it honed in-house directly to third-party sellers. Microsoft and Google have turned internal AI capabilities into cloud services anyone can buy.

In the AI era, moats are not all dug. Half of them are "allied" into existence.

Entry Five: The Scariest Predictions Came True Most Thoroughly

The first four entries were all about "what AI can get done."

Now, to settle the one that stings most.

A whole cluster of predictions on that list was saying the same thing: use AI carelessly, and trust collapses fast. I thought at the time that this was overstated. This year, I take it back.

First, deepfakes. VMware's 2024 cybersecurity outlook: AI-fabricated social-engineering attacks grew 5-fold in two years, and two-thirds of large enterprises have already been hit. A fake executive holds a video call and tricks finance into wiring money — that is a real case now, not a joke.

It means everything you see and hear is, by default, possibly not real. And so Mastercard and JPMorgan are working on watermarking and anti-fraud, while Adobe and Autodesk embed "Content Credentials" into design files. The EU's AI Act passed in 2024; by August of this year, the hard obligations for high-risk scenarios are already in force. On the US side, accountability legislation targeting deepfakes is also advancing.

Second, AI content nobody reviewed. AI-generated FAQs, onboarding guides, bot replies — if they go out with no human review, then one wrong policy explanation isn't just one wrong sentence; it's a garbled message circulating under your company's name. Samsung's approach: anything published externally goes through multiple rounds of verification. Dumb? Dumb. Effective? Effective.

Third, support bots shoved onstage before they were ready. To cut costs, companies pushed undertrained bots onto the front line: they can't parse what customers say, they answer the wrong question, and when it's time to hand off to a human they absolutely refuse. The result? Customers driven back to phone and email channels, costs up instead of down, and trust in self-service overdrawn in a single stroke. Forrester put it bluntly: applying AI to customer service is dirty work, not glory work. Do it badly, and you're better off not doing it at all.

Fourth, metric overload. Plenty of customer-experience teams are still stacking up surveys, dashboards, and satisfaction scores. But what executives want this year is: how does experience affect revenue, renewal rates, costs? You push over another sheet of sentiment scores that decides nothing, and next time nobody looks at it. In airlines, telecom, healthcare, and SaaS, the leading companies have already shifted from "monitoring experience" to "using experience to drive growth."

Fifth, static journey maps. A beautifully drawn customer journey map that connects to no real data and governs no process is wallpaper. A living map is wired to product telemetry and real-time alerts — when customers churn at any given step, you know the same day.

These five all point to the same sentence:

Trust takes ten years to accumulate and one incident to collapse.

Entry Six: The Rules Are Catching Up

The more capable AI gets, the less governance remains an elective.

Forrester predicts AI-related class-action lawsuits will rise 20%. A miscalculated loan, discriminatory hiring, a bot that said the wrong thing — each one can end up in court.

IBM made a rather symbolic decision: on ethical grounds, it refused to let its facial recognition be used for mass surveillance. To sell or not to sell is business in the short term; in the long run, it's the sign above your door.

Governments everywhere are setting rules. The UAE created an AI ethics advisory board. Microsoft, Google, and OpenAI are spending heavily on detecting AI-generated content.

Over the next two to three years, the sentence "our AI is audited, fair, and accountable" will go from a bonus item to the price of admission.

Two more predictions sit further out — might as well check them too.

One is AI warfare. The US is using AI to compress operational planning, China is advancing what it calls "intelligentized warfare," and Israel uses systems to generate target lists at scale. The contest between "speed and accountability" on the battlefield will spill over into civilian life: one society grows used to rubber-stamping machine recommendations, another starts doubting every algorithm. The ethical answer technology delivers in this ultimate arena will, in the end, become the floor of everyone's trust in AI.

The other is biomedicine. The Chan Zuckerberg Initiative (CZI) is putting billions of dollars into building "virtual cell" models, aiming to simulate human biology inside a computer. AI medical devices approved by regulators already number in the hundreds. In new drug development, AI has compressed several stages by years.

The medical-device track also runs on a special logic: one failure isn't "a bad experience" — it's a human life. So the experience bar there is set almost brutally exacting. And that is exactly the bar every other industry should be holding itself to.

Finally, Closing Out the Ledger

Twenty-six entries reconciled, and it really comes down to two sentences.

First: the AI dividend is concentrating, faster and faster. McKinsey's global surveys keep confirming one thing — the ones actually capturing AI's economic gains are that small cohort of companies that treat AI as strategy and roll it out company-wide. The tools keep getting cheaper, yet the gap keeps getting wider. Why? Because the gap was never in the tools. It's in the organization.

Second: AI is not a technology; it is organizational surgery. Data foundations, employee skills, governance rules, experience design — if any one of the four is missing, AI is just an expensive demo.

Last year, those 26 entries looked to me like prophecy.

This year, they look more like a blueprint. Customer experience, employee experience, innovation speed, organizational decisions — every beam and column already has its dimensions drawn.

This year is the year to break ground.

Here's wishing you a foundation poured more solid than anyone else's.