The AI Readiness Issue

Our inaugural issue: what AI readiness actually means in 2026, the five most common blockers we see, and why the organizations furthest along started with governance, not models.

What does this digest cover?

The top patterns we observed across AI readiness engagements in the first quarter of 2026. Five themes: data foundation as the dominant blocker, the 30-day audit pattern, the value of executive sponsors, the cost of skipped metric certification, and where AI vendors are pricing in 2026.

Welcome to the first Thinklytics Digest. We’re a group of seasoned analytics and AI consultants working out of Austin, Texas. No fluff, no vendor hype, no conference buzzwords. Just straight talk about what’s actually moving the needle in enterprise AI, and what’s not.

• 60% AI projects to be abandoned through 2026. Gartner projects that 60% of AI projects will be abandoned through 2026 due to lack of AI-ready data. This is not a prediction about AI capability - it is a prediction about data infrastructure.

Source: Gartner, 2025

Lead Essay: AI Readiness Is About Data, Not AI

By 2025, the biggest question I’m hearing from C-suite leaders is simple: Are we really ready for AI? Most have approved the budget and watched plenty of vendor demos. But when it’s time to run pilots, they just get stuck.

Here’s the hard truth: you’re just not ready yet. And nope, it’s not about the AI tech itself.

AI readiness? It all boils down to the data. The companies that actually crush it with AI didn’t just dive into model-building. Nope, they sorted out their data first. We’re talking clear metrics, rock-solid quality checks, and knowing exactly where the data came from, that’s data lineage. They built their AI on a solid data foundation, not as an afterthought. Skip that step, and AI won’t take hold.

Here’s the thing: a lot of companies dive into AI projects before their data is anywhere near ready. They kick off pilots, run into roadblocks, and then scramble to fix messy data on the fly. That scramble means they cut corners on prepping the data, which tanks the quality and dooms the whole AI effort before it even gets going.

Here’s the thing: if your AI projects keep hitting walls, just stop for a second. Establish your data foundation first. Then, when that’s solid, dive back in. It’s tough advice, sure, but frankly, it’s the only way to make real progress.

The five data-layer failures that kill AI initiatives

• Undefined metric semantics
• Identity resolution gaps
• Lineage gaps at inference
• Feature store governance gaps
• Infrastructure-data coupling

Each failure is independently sufficient to stall an AI initiative. Most stalled initiatives have two or more.

Source: Thinklytics Research, 2026

Pattern Watch: Five Data Issues That Kill AI Projects

When we dug into our 2025 projects, the same five data-layer problems kept tripping us up and stopping AI from going live. Here’s what we ran into:

• Undefined metrics (29 of 47 stalled projects). We saw 29 out of 47 AI projects hit a wall because they didn’t establish clear KPIs. Without solid goals, the models end up training on messy, unreliable data. That’s a huge reason why projects get stuck.

• Identity resolution gaps (25 of 47). Here’s the deal: a lot of times, customer or product info doesn’t line up across systems. So, your models end up training on duplicates or phantom records, which throws the whole analysis off. It’s a messy problem we see all the time.

• Lineage gaps (22 of 47). Nearly half the time, companies hit a wall trying to trace AI outcomes back to the original data. That makes audits a headache and trust harder to earn because there’s no clear paper trail.

• Feature governance gaps (18 of 47). Out of 47 AI features we looked at, 18 weren’t properly versioned or tracked. So basically, these models can drift and degrade without anyone catching it.

• Infrastructure-data mismatch (15 of 47). Here’s a common snag we hit: running AI on platforms that weren’t built for it. It’s messy. You get lag, version chaos, and access control problems. Basically, it’s like forcing a tools working in ways they weren't designed for, nothing works the way it should.

Here’s the thing: when teams nailed all five key factors before even starting the modeling, they got AI into production 39 times out of 48. But if they skipped those steps? They barely made it, only 7 times out of 100. That gap is massive.

• 95% Enterprise GenAI projects that fail to deliver measurable ROI. 95% of enterprise generative AI projects fail to deliver measurable ROI. The five data-layer failures described in this digest account for the majority of those failures.

Source: MIT NANDA Initiative, 2025

Case Snapshot: Express Scripts

Express Scripts manages pharmacy benefits for 100 million people. In 2024, they had three machine learning pilots, risk stratification, drug interaction prediction, and prior authorization automation, that had been stuck in limbo for over a year.

Here’s the real pain point: member identity resolution. Express Scripts went on an acquisition spree, scooping up several companies. That meant the same person ended up with different IDs across various systems. Their models were trying to make sense of mashed-up, incomplete data. No wonder the results were all over the place.

We put together a member identity solution in just 14 weeks. Then, six weeks after launch, all three pilots fired up again. Our risk stratification model was live in 12 weeks, and the other two models rolled out within six months.

This wasn’t about launching new models or flashy platforms. It came down to one simple thing: truly knowing who the AI was actually talking to.

Recommended Reading

Let’s talk about Data-Centric AI, a concept Andrew Ng has been pushing hard. The idea is simple but powerful: focus on your data, not just the models. Instead of spending hours tweaking algorithms, the real win comes from cleaning up and improving your data. It’s frankly the biggest shift in AI right now. If you want to geek out on this more, head over to datacentricai.org.

The AI Readiness Checklist is basically your company's quick reality check. It helps you see how you’re doing with data quality, governance, infrastructure, talent, and culture. Think of it like a fast way to figure out what’s working and what needs fixing before you jump into AI.

Why AI Projects Fail: We dug into 300 AI projects that flopped. The biggest reason? Shoddy data quality. It sounds obvious, but it’s a huge deal breaker. If your data’s messy or missing chunks, your AI’s odds of success drop off a cliff. [Harvard Business Review]

We put out the Thinklytics Digest every month. If you want to get it, just drop a quick email to [email protected]. Easy.

Frequently asked questions

What does this digest cover?

The top patterns we observed across AI readiness engagements in the first quarter of 2026. Five themes: data foundation as the dominant blocker, the 30-day audit pattern, the value of executive sponsors, the cost of skipped metric certification, and where AI vendors are pricing in 2026.

Who is this digest for?

Heads of data, CTOs, and CIOs who want a fast read on what's actually happening across multiple companies' AI readiness journeys rather than one company's case study.

How is this different from the long-form articles?

The digest is shorter, more strategic, and aggregates patterns across engagements rather than going deep on one topic. The long-form articles linked from each section are where the detailed treatment lives.

Is this published monthly?

Currently quarterly. The cadence will increase to monthly once we have a deeper inventory of pattern-level observations.

Where can I see the case studies behind these patterns?

The digest references case studies for each pattern. The full case study library is at case studies. Each pattern is anchored to specific engagements where we can publish details.

How does Thinklytics use these patterns in client work?

The patterns inform our engagement scoping. When a new client describes a problem that matches a pattern we've seen 5+ times, we know what the failure modes are and can scope the remediation accordingly. Read more at data foundation.

What was the most common stall point in Q1 2026?

Metric layer drift. Pilots ran on a clean cohort, production hit the real warehouse, and the metric inconsistencies surfaced at scale. Fixing it took 6 to 10 weeks per environment on average.

Where can I see the engagements behind these patterns?

Each pattern in the digest references one or more case studies. The full library is at case studies. Particularly: Kaiser Permanente Metric Governance for the metric layer pattern, Express Scripts AI Readiness for the data foundation pattern.