12 Predictions on Digital, Data, Analytics, and AI (D2A2)

About two weeks ago, I was standing on a train, waiting to exit, and making the final edits to a 37-page Google doc on my phone.

These are the unglamorous, but often necessary, moments to make it happen.

On Tuesday, we finally released the 2026 D2A2 Insights Report – 12 predictions covering Digital, Data, Analytics, and AI. Our team included PhDs from Princeton and Rutgers, consultants who’ve worked with P&G, Shell, and Apple. Working alongside them really felt like a great honor.

I’ve already shared my personal AI predictions for 2026. You can read them here.

Today, I want to share a broader perspective of what our team thought – and the predictions I’m most skeptical about.

Let’s dive in!

About the Report

The D2A2 Predictions & Prescriptions Report has been running for six years now. D2A2 stands for Digital, Data, Analytics, and AI — the four major forces currently reshaping business realities.

What makes this report different from the usual prediction pieces is that it pairs each prediction with concrete prescriptions. Not just "here's what's coming" but "here's what to do about it." Leaders at firms like Honda, JP Morgan, Siemens, SAP, and Novartis use it to cut through the noise and focus on what actually matters.

The full report is available for free at D2A2.ai:

Here are the key themes that stood out:

Theme 1: AI Gets Smaller and Closer

The report doubles down on a trend I've been writing about: AI is moving out of the cloud and onto your own infrastructure.

Two predictions reinforce this.

First, Small Language Models will gain serious enterprise traction. Not every task needs a trillion-parameter model. For domain-specific work – customer service, IT helpdesk, document processing – smaller models are faster, cheaper, and often more reliable.

Second, local and offline AI reaches critical mass. Open-weight models now rival proprietary ones. Consumer hardware can run them. The tooling matured. For high-volume or sensitive workloads, the math starts favoring on-prem deployment.

This aligns with whatI shared in December. The team consensus here is strong.

Theme 2: AI Gets Held Accountable

Three predictions cluster around the same idea: AI governance is shifting from principles to proof.

The report argues that data observability becomes the backbone of AI governance. You can't trust AI outputs if you can't trace the data feeding them. Drift detection, lineage tracking, quality monitoring, context engineering – this is the new compliance baseline.

AI governance itself evolves toward measurable accountability. Boards and regulators will demand KPIs, not vague promises. Fairness scores and audit trails. Organizations need evidence that systems behave as intended to keep them in production.

And ROI frameworks – my section – will finally bring discipline to AI investments. The core argument: stop forcing productivity tools like ChatGPT into hard ROI calculations. Measure adoption instead. But for engineered AI solutions that consume real capital, you absolutely have to demand real returns.

This theme is where the report feels most urgent. The free experimentation era is ending.

Theme 3: AI Infrastructure Grows Up

The more technical predictions focus on data architecture. Data products and data mesh become standard practice (even if this will be, again, called differently). AI-native platforms replace traditional schema-first approaches.

Headless architectures decouple front-end from back-end for more flexibility and easier access for AI agents.

The short version: the boring plumbing work is finally getting attention.

Where I'm Skeptical

To be honest, there are a few predictions in there I probably wouldn't have made. But I agree with the rationales behind them – something which I really like about this format. You don't just see the predictions, but why it's predicted.

For example, the report predicts that Classical AI will complement Generative AI — and traditional ML for prediction and optimization is making a comeback alongside LLMs. Maybe. But I think most organizations still haven't really figured out GenAI yet. Adding "also do classical ML" to the mix will probably take longer than a year.

There's also a prediction that Agentic Commerce — AI agents autonomously shopping on your behalf — will be a breakout agentic use case. I'm not fully sold on that one yet. The trust gap is still too big. Plus, booking something myself is still 100x faster than doing it via ChatGPT.

I do see agentic systems getting big for buying research though. Not making the deal, but preparing a deal-ready decision. Comparison tables, flight options, you name it. What used to be a whole industry before could be swiped away by the sheer computing power of LLMs. But I expect humans to still make the final call as people aren't ready yet to let AI spend their money unsupervised.

Don't get me wrong. I love that we have these predictions in the report. 6 authors means 6 perspectives. Each of us owned our domains, but ownership didn't mean we had to agree on everything.

This is exactly how a report like this should work.

Behind the Scenes

Most of the writing happened in the weeks before and after Christmas. "I love deadlines and the whooshing noise they make as they go by."

Each author owned a few core topics. We wrote our predictions, then discussed and commented on each other's work. Everyone followed the same structure: Prediction, Rationale, Prescription.

The actual writing happened across 5 time zones with holiday travel, family time, and Google Docs notifications competing for attention.

But we still somehow made it to show up at the same time in the same Zoom call:

Big shoutout to Raef Lawson, Executive Director of the Profitability Analytics Center of Excellence (PACE), for contributing an awesome foreword. And to Steve Rosvold at CFO.University for rehearsing and hosting the full author team live on LinkedIn next week. If you want to hear us discuss (and debate) the predictions in real time, join here:

Read the Full Report

You can download the complete 2026 D2A2 Insights Report here.

It covers all 12 predictions with background, rationale, and prescriptive recommendations. Plus, a little recap on which predictions we made last year and how these turned out.

Worth a read if you're planning your AI strategy for the year ahead.

See you next Saturday,
Tobias