[insight]
Think Big. Start Small. Scale Fast.
Most AI projects don’t fail because the models don’t work.
They fail because the organization didn’t think far enough ahead… or they tried to do all of it day one.
If you’re serious about driving real impact with AI you need to operate on three timelines at once:
Think big. Start small. Scale fast.
Let’s break that down.
Think Big (18 Months Out)
AI is not a feature. It’s an operating model shift.
When I say “think big,” I don’t mean abstract vision statements. I mean:
[ ] What business problem are we actually trying to solve?
[ ] What business process are we aligning to?
[ ] What does the end-state architecture look like?
[ ] What does governance, data cleanliness, and control need to look like?
[ ] What does this unlock in 18 months if we get it right?
In most organizations, the real gravity isn’t the AI.
It’s the data.
The data estate is often monstrous. Legacy ETL pipelines. Fragmented systems. Half-migrated platforms. Shadow processes. Manual workarounds.
If you don’t think about the end-state — the platform, the data model, the governance — you’ll build tactical AI solutions that can’t scale.
Recently, I was on the phone with an organization migrating from a monolithic .NET ETL application into a microservices architecture using a strangler pattern. That’s a big move. It’s the right move. But it’s big.
The long-term goal?
Better data. Cleaner architecture. AI-ready systems.
That’s thinking big.
But here’s the trap: you don’t need all of that finished before you start delivering value.
Start Small (90 days)
If “think big” is the 18-month view, “start small” is the next 90 days.
The question becomes:
What is the one problem we can solve right now that creates disproportionate impact and builds momentum?
In this case, we didn’t try to decompose the entire ETL application at once.
We isolated a single high-leverage function: missing data in the quote to cash process.
Why?
[ ] It had real data integrity issues.
[ ] It touched systems that ultimately need to migrate anyway.
[ ] It had high business value.
[ ] Fixing it would surface immediate ROI.
That’s how you choose where to start.
Not based on what’s easiest.
Not based on what’s most technically interesting.
But based on:
[ ] Impact
[ ] Strategic alignment
[ ] Architectural relevance to the long-term goal
In this specific example, we also identified a gap where the sales team had previously captured valuable lead data — but staffing constraints killed the process.
So instead of hiring more people, we used agentic technologies to capture and structure that data automatically.
Humans used to do it.
They couldn’t anymore.
AI now could.
That’s a 90-day win.
And 90-day wins build belief.
Scale Fast (Build for what's next)
Here’s where most organizations get this wrong.
They start small… but they build in a way that doesn’t scale.
If you do the first phase correctly, momentum will create demand.
Leads start flowing.
Sales gets better inputs.
Revenue impact becomes visible.
Then the organization asks:
What’s the next function? What’s the next application? What’s the next bottleneck we can unlock?
If you didn’t design Phase 1 with scale in mind — with clean APIs, extensible data models, governance baked in — you’ll end up rebuilding everything.
Scaling fast doesn’t mean moving recklessly.
It means:
[ ] Architecting intentionally.
[ ] Building reusable components.
[ ] Creating patterns that can be replicated.
[ ] Designing for the 18-month outcome while executing in 90-day increments.
When momentum hits, you should be able to expand — not replatform.
The Real Discipline
AI transformation isn’t about launching one big initiative.
It’s about disciplined sequencing.
[ Think big ] Define the long-term architecture and business outcome.
[ Start small ] Deliver a high-impact win in 90 days.
[ Scale fast ] Build in a way that compounds value instead of creating technical debt.
Most organizations either:
[ ] Think big and never start.
[ ] Start small but stay small.
[ ] Or scale chaos.
Very few do all three well.
The ones that do?
They don’t just “experiment with AI.”
They operationalize it.
