The five levels of company intelligence.
One page per question that matters: what each level looks like, the signs you're at it, what advancing requires, and the trap that keeps companies stuck there.
L1 · Scattered.
At L1, company knowledge lives in heads, inboxes, and files. Nothing is queryable; everything is askable, if you know whom to ask and they're not on vacation.
Signs you're here: the same question gets different answers by department; onboarding takes months because the process is oral tradition; losing one employee loses a capability. Advancing requires an inventory, not a tool: Discover, done as a map. The trap is buying software to skip the mapping, which turns L1 into L2 with receipts.
L2 · Collected.
At L2, the systems exist: a CRM, a wiki, a drive, an automation tool. They disagree, and reconciling them is a standing meeting. Most B2B companies live here.
Signs you're here: three systems hold three versions of the same customer; reports get 'cleaned up' before leadership sees them; every new tool added a silo. Advancing requires electing one kernel and demoting everything else to clients of it. The trap is the migration fantasy: waiting for the perfect future stack instead of cleaning the one you have.
L3 · Exposed.
At L3, one kernel holds truth, surfaces are legible to humans and agents, and signal flows without a human ferrying it. This is the 90-day target, and where compounding starts.
Signs you're here: a fair question about the company gets one answer regardless of who's asked; answer engines cite you; retiring a spreadsheet is a normal Tuesday. Advancing requires governance, because agents are next and they need owners. The trap is declaring victory: L3 unmaintained decays back to L2 within a few quarters.
L4 · Operated.
At L4, governed agents work from the kernel daily: answering, routing, drafting, reporting. Each has an owner, a scope, a log, and a review loop. Boring and correct is the success state.
Signs you're here: agents resolve real work without a champion babysitting them; legal has approved a process, not a pilot; the question is which agent next, not whether. Advancing requires feeding outcomes back into the kernel so the system learns. The trap is agent sprawl: shipping agent five before agent one has a review cadence.
L5 · Compounding.
At L5, the platform improves itself: every customer interaction, closed deal, and answered question adds signal the whole system can read. Few companies are here. That's not a sales line; it's the honest ceiling.
Signs you're here: new capabilities cost days, not quarters, because the substrate is clean; your company's public answers improve without a content sprint. There's no level six. Staying at L5 is the work, and it looks like disciplined operation, forever.
How do we move from L1 to L3 in 90 days?
Scope ruthlessly to the revenue path: Discover in weeks one and two, Expose and Wire only what carries pipeline, and prove L3 on that slice by day 90. Breadth comes after the checkpoint, not before.
The full method is on the Four-Move Method page. The short version: the 90-day claim isn't about working fast. It's about refusing to boil the ocean, and it holds because the first slice funds belief in the second.
Can we skip a level?
You can skip the suffering of a level, not the work of it. L1 to L3 in 90 days doesn't skip L2; it compresses it by scoping to the revenue path first. What you can't do is deploy L4 agents onto L2 disagreement. Every company that's tried bought a very expensive demonstration of the ladder.
Where are most B2B companies?
L2, by a wide margin, usually with pockets of L1 in operations and a paid-but-idle pilot pointing at L4. Self-locate using the signs above, or let the diagnostic do it with evidence. Most teams place themselves one level higher than their systems do.