What level are most revenue teams at?

L2, single-player productivity. Licences are deployed and individuals are faster, but no workflow has an owner and no metric has a pre-deployment baseline.

No. Buying an L5 governed agent platform while operating at L2 produces an expensive L2, because there is no shared workflow for the agent to join and no baseline to judge it against.

What is the fastest level to clear?

L3. Defining one workflow, one format and one handoff usually takes weeks and costs no new software.

A named metric with a baseline captured before deployment, a fixed review cadence, and an intervention rate you track alongside the outcome.

Is L6 realistic for a mid-market team?

In one loop, yes. Across the whole revenue org, rarely. Pick one loop where the feedback path is short and the data is already clean.

How often should we rescore?

Quarterly, in the same format, taking the lowest passing level across functions rather than the average.

← Benchmark

The AI Adoption Curve for Revenue Teams: L1 to L6, and Where Most Teams Stall

Six levels from personal experiments to a system that runs without heroes. Each level has an entry test, a failure mode, and a single exit criterion. Most revenue teams are at L2 and reporting L4.

Jonathan Kvarfordt · Published September 6, 2026 · 12 min read

Why trust this analysis?

The short answer

How do we tell what AI maturity level our revenue team is actually at?

Test for the exit criterion of the level below, not for the tools installed. A team is only at a level when the previous level's work survives the departure of the person who built it. Most revenue teams sit at L2, single-player usage, while reporting L4.

Decision rule

You cannot skip a level. Buying an L5 product while operating at L2 produces an expensive L2.

Operator action

Score your team on the six exit criteria below and fix the lowest failing one first.

Supporting pages

Adoption is real, measurable, and slower than the discourse the data behind this piece
The Single-Player AI Problem definition
Optimization Theater definition

Last reviewed September 6, 2026

Maturity models get abused. They become slides where every company is one level from the top and nobody names the level they are actually at. This one is built to be failed. Each level has one exit criterion, and if you cannot demonstrate it, you are not at that level regardless of what you have bought.

The full framework, with visuals and technology examples per level, lives at the AI maturity ladder. This piece is the diagnostic version.

The argument

How this benchmark breaks down

A map of the sections ahead, in the order the case is made. Schematic, not a dataset. Source-cited charts live in the research library.

L1: Curiosity

Individuals use consumer AI tools on personal accounts. No policy, no procurement, no visibility. Output quality varies by person and nobody can tell you what is being pasted into a public model.

Failure mode: leadership mistakes activity for adoption. Sixty people using a chatbot is not a capability.

L1 to L6

Each level is proved by the exit criterion below it

Levels are earned by evidence, not by tools installed. Schematic, not a dataset. Source-cited charts live in the research library.

Exit criterion: you can name what is being used, by whom, on what data. Most organisations discover a shadow AI stack here that is larger than the sanctioned one.

L2: Single-player productivity

Licences are bought, individuals get faster, and the gains stay with the individual. The strongest rep writes better emails. Nothing about the team's output changed, because the improvement never left one person's workflow.

This is where most revenue teams sit, and it is the most expensive place to sit, because the spend is real and the leverage is personal. The mechanics are in the single-player AI problem.

Failure mode: reporting team-level impact from individual-level anecdotes.

Exit criterion: one workflow where the AI output is an input to someone else's job, and the handoff is defined.

L3: Shared workflow

A defined process runs through AI with a human in the loop, and it is the same process for everyone doing that job. Call summaries land in the CRM in one format. Account research follows one template. Output enters a system rather than an inbox.

Failure mode: the workflow depends on one enthusiast. When they change roles, the workflow reverts within a quarter. This is the most common silent regression in the ladder.

Exit criterion: the workflow survived the departure of the person who built it.

L4: Instrumented

You can measure what the system produced. Not usage, not seats, not sentiment. Output volume, quality rate, human intervention rate, and the downstream metric it was supposed to move.

Most teams claim this level. Few pass it, because passing it requires a baseline captured before the deployment, and the baseline is what everyone skips. Without a before, every after is a story. That is the argument behind the pipeline truth test.

Failure mode: optimization theater, where the dashboard improves and the business does not.

Exit criterion: a named metric with a pre-deployment baseline, reviewed on a fixed cadence, that a sceptic in the room could not dismiss.

L5: Governed

The system has owners, limits and reverse gears. Kill criteria are written. A rollback plan exists per action class. Someone can pull the agent without convening a committee, and the pull is rehearsed rather than theoretical.

This is not bureaucracy. Governance is what lets you run agents against live pipeline at all, and the evidence says the teams with mature guardrails pull agents more often, not less, because they can see failures early. The compiled data sits in the rollback research and the operating document is in the rollback plan requirements.

Failure mode: governance written as policy and never rehearsed. A rollback plan nobody has executed is a document, not a control.

Exit criterion: you have executed a rollback in production, on purpose or otherwise, and can describe how long it took.

L6: Compounding

Output from one part of the system improves another part without a human moving it. Call outcomes retrain the qualification model. Won-deal patterns update targeting. Support resolutions feed enablement. The loop closes and the improvement rate itself becomes the metric.

Very few revenue organisations are here, and none of them arrived by buying it. They arrived by clearing L4 and L5 with unglamorous discipline. The loop mechanics are in GTM loop engineering.

Exit criterion: a measured improvement that no person initiated, traced to a specific feedback path.

How to score honestly

Ask each function to state their level and give the evidence for the exit criterion below it.
Reject every claim supported only by tool ownership or seat counts.
Take the lowest passing level across functions as the organisation's level. Maturity does not average, it bottlenecks.
Fix the lowest failing exit criterion. Do not buy for the level above it.
Rescore quarterly, in the same format, so the trend is visible rather than argued.

The distance between L2 and L4 is where nearly all wasted AI spend lives. It is also the cheapest distance to close, because it costs process discipline rather than licences.

Take it to the room

The short list this issue leaves you with

Pulled from the argument above, written so you can read it out in a pipeline or board review. Schematic, not a dataset.

Frequently asked questions

What level are most revenue teams at?: L2, single-player productivity. Licences are deployed and individuals are faster, but no workflow has an owner and no metric has a pre-deployment baseline.
Can we skip a level?: No. Buying an L5 governed agent platform while operating at L2 produces an expensive L2, because there is no shared workflow for the agent to join and no baseline to judge it against.
What is the fastest level to clear?: L3. Defining one workflow, one format and one handoff usually takes weeks and costs no new software.
How do we prove L4?: A named metric with a baseline captured before deployment, a fixed review cadence, and an intervention rate you track alongside the outcome.
Is L6 realistic for a mid-market team?: In one loop, yes. Across the whole revenue org, rarely. Pick one loop where the feedback path is short and the data is already clean.
How often should we rescore?: Quarterly, in the same format, taking the lowest passing level across functions rather than the average.

Share this issue

Posting to Instagram or TikTok? Copy the link, it carries the title, summary and share image.

Get the next benchmark, with the sample size attached.

Keep reading

Benchmark

The AI Revenue Team Org Chart: Which Roles Change, Which Appear, Which Go

Agents do not flatten the org chart. They move decision rights. Here is the before and after for seven revenue roles, the three jobs that appear, and the reporting line that decides whether any of it works.

Benchmark

The Real Cost of an AI SDR Is Not on the Pricing Page

Vendors sell you a seat price. Your P&L pays for list decay, deliverability repair, management overhead, and the meetings that never should have been booked. Here is the math that actually matters.