Dot, Dash, Star — and the Floor Under All Three

By Errin O'Connor — Founder & Chief AI Architect, EPC Group · Microsoft Solutions Partner across all six designations · G2 Leader in BI consulting

If you run AI investment decisions at an enterprise and you have not adopted Allie K. Miller's Dot-Dash-Star framework, adopt it this week — it is the cleanest portfolio language in the industry. A Dot is a specific line-of-business use case — an HR bot, sales-call transcription — proven patterns that should return value in roughly three months, six for a slower enterprise. A Dash is broad workforce enablement — Copilot or ChatGPT Enterprise across the population. A Star is the transformational bet that changes the shape of the business. Different risk, different timeline, different governance — and Miller's sharpest warning is against confusing them: a Dot that takes two years to prove value isn't a slow Dot. It's a failed one.

I use her framework in client conversations, and I am about to spend an entire article agreeing with it — with one amendment from twenty-nine years in the field that changes how you fund all three letters. Dot, Dash, and Star are three shapes above the same floor. And when the floor is rotten, the framework doesn't save you — it just gives your failures better names.

Why the three-month Dot takes two years

Walk through why Dots actually miss Miller's three-month bar, because I have done the postmortems and it is almost never the model. The HR bot answers policy questions from four conflicting policy libraries across duplicate SharePoint sites — which one is true? The sales-transcription agent works beautifully and then the insights go to a dashboard fed by a semantic model with implicit measures nobody certified — which number is real? The invoice-matching Dot hits the vendor master and finds the same supplier spelled six ways across two ERPs — which entity is the entity? In every case the use case was proven, the vendor was fine, the pattern was textbook — and the timeline tripled because the team spent months doing unplanned data archaeology under a project name that said "AI."

The published numbers say this is the norm, not the exception. Industry research across enterprises with a thousand-plus employees: analytics teams working across roughly four hundred data sources on average, nearly one in five juggling more than a thousand; data scientists spending about sixty percent of their time cleaning data and another nineteen percent just finding it; sixty to seventy-three percent of collected enterprise data never used for analytics at all; eighty-three percent reporting silos. That is the floor your Dots land on. AI does not fix those numbers — AI inherits them, at the speed of a confident answer.

The floor is one build. The letters are many.

Here is the economics insight that changes the funding conversation, and it is the reason I am writing this amendment rather than a footnote. The floor is a shared asset. You build it once, and every Dot, Dash, and Star lands on it. A governed lakehouse in Microsoft Fabric — OneLake as the single foundation, lineage you can trace, a semantic layer with explicit certified measures, Purview classification and sensitivity labels enforced at grounding time, capacity provisioned so the AI experiences don't silently throttle. Fund that once and your Dots actually hit the three-month bar, because the archaeology is already done. Your Dash — Copilot across the workforce — grounds against certified truth instead of narrating the confusion more articulately. And your Star becomes fundable at all, because no board should approve a transformational bet on a foundation nobody has walked. The industry has converged on the same conclusion this year: the lakehouse has become the foundation for enterprise AI — not by replacing systems, but by unifying and governing them. Miller tells you how to sort the bets. The floor is what makes any of them payable.

And this is where her three-year adoption timeline — research to startups to enterprise, right on schedule — gets its enterprise-specific corollary: the clock only runs on schedule for organizations whose data was ready when the technology arrived. Everyone else adds their archaeology to the timeline and calls the delay "AI being overhyped." The hype was fine. The floor was missing.

What I tell clients to do

One. Adopt Dot-Dash-Star as your portfolio language — and add a fourth line item above all three: the Floor, funded as shared infrastructure, not taxed onto some unlucky Dot's budget.

Two. Before funding any Dot, run the two-week floor check on the specific data it will touch: sources, duplication, ownership, measures, labels. If the check fails, fund the fix first — it is cheaper than a tripled Dot timeline and it accrues to every future letter.

Three. Hold Miller's bar ruthlessly. A Dot that misses three months gets a data-floor diagnosis, not an extension. The diagnosis will be right more often than the extension.

Four. Sequence one Star-grade floor investment — the Fabric lakehouse, the certified semantic layer, the Purview posture — and then watch your Dot velocity compound quarter over quarter. That compounding is the entire business case, and it is the engagement EPC Group runs more than any other.

The two-week floor check (run per Dot, before funding)

Day 1–2: source census for the specific workflow (systems, owners, refresh, duplication). Day 3–5: truth test — same question asked of each candidate source; count divergent answers. Day 6–8: measure audit — explicit vs implicit; certified vs wild. Day 9–10: label/permission check — sensitivity labels present and enforced on everything the use case will ground.

Verdict: PASS → fund the Dot. FAIL → fund the fix as shared floor (it accrues to every future letter), then the Dot. Publish the verdict — the discipline is the deterrent.

Where I land

Miller built the right sorting system, and I would put it in front of every AI steering committee in the country. My amendment is the one the sorting system assumes and the field keeps skipping: the letters are shapes; the floor is load-bearing. Fund the floor once, and Dot-Dash-Star becomes what she designed it to be — a portfolio that returns. Skip the floor, and it becomes the best-organized failure taxonomy your board has ever seen.

Multiple models. One truth.

The data behind this (sources and verification)

• Miller's Dot-Dash-Star definitions and the three-month Dot bar — Insight Partners / ScaleUp:AI.
• Industry research (1,000+ employee enterprises) — ~400 data sources average, ~1 in 5 over 1,000; ~60% of data-scientist time cleaning + ~19% finding; 60–73% of collected data never used for analytics; 83% report silos, 97% of those say silos hurt performance.
• Lakehouse convergence: 2026 industry analysis — The lakehouse "is becoming the foundation for enterprise AI" — unify and govern, don't replace (CIO, June 2026).
• MIT Technology Review Insights / Metrigy — 78% of businesses say service improved; 31% of consumers agree — the perception gap ungoverned data produces.

Third-party figures above are attributed to their named sources as of the Last verified date. EPC Group audit figures are directional findings from client engagements. Items marked [VERIFY] must be confirmed before external quotation.

Frequently asked questions

Multiple models. One truth.