The free universal sales skills assessment engine.

Bring your Anthropic key. Peer runs the rest.

Peer reads every call your recorder already captures, grades the selling skill each moment shows against one universal standard, and hands the graded record to the AI your team already asks. Hosted by us, run on your key, set up in four screens.

Start free The four steps

Your Anthropic key · your note taker · your Reps · running in one sitting

You pay Anthropic for the tokens, about $0.27 per graded call at today’s prices. Peer charges nothing to run.

Seller · recorded call

routespecialist graderevidenceblind re-markreturn

competency—

grade—

evidencequote · —

blind re-mark sampled ✓agreement logged

an atomic observation — the data type your stack doesn’t produce

What it is

An engine for grading demonstrated selling skill, on your key.

Peer is hosted. The engine runs on our servers and calls Anthropic with the key you paste; prompts, rubrics, the Standard and the grading logic never leave the server, and the engine returns grades, never the how. One Claude model, pinned by us (claude-opus-4-7): you never pick a model, and a key from any other provider is refused on paste with one sentence saying why.

Every grade names the skill underneath the moment, the level the Rep showed, and what the next level up looks like. It is a map of where each Rep’s skill is and where it can go, not a script for the next call.

Setup

Four steps, each one screen. The engine is running at the end of the fourth.

1 · Paste your Anthropic key

Checked on the spot with one cheap live call. You see the pinned model and the rough cost per graded call. Stored encrypted, never shown back beyond its prefix, revocable any time.

2 · Connect your note taker

tl;dv, Fireflies, Fathom, Gong, Avoma, Zoom or Google Meet with a key; Teams, Otter, Read.ai or anything else by transcript upload or API. Your first call list appears on screen before you leave the step.

3 · Your Reps and their roles

Reps are pre-filled from the recorder’s participants. Confirm them and tell us the role each Rep plays, from a short plain-English list with a suggested default.

4 · Running

Peer grades your most recent eligible call and shows it, calibration badge included. From here every new call is graded as it lands.

One progress strip, the next step always obvious, every error one sentence saying what to do. Nothing asks you to understand versions, prompts, frameworks or the MCP. Connecting Peer to Claude or ChatGPT is a bonus on the Running page, not a step.

What you get

A graded record of each Rep’s skill, with the moment that earned every grade.

You bring

An Anthropic API key from console.anthropic.com (a Claude subscription is not one)
The note taker you already run
Your Reps, and the role each one plays

that is the whole list

Peer returns

Atomic observations: competency, grade, polarity, the quoted moment and its timestamp
Each Rep’s level per competency, and its direction over time
What a Rep has never been observed doing, named rather than skipped
The whole record over MCP, to Claude, ChatGPT or Claude Code

grades, never the how

See it before you sign up

Same AI, same question. One of them can see Peer.

app.peerlab.ai/demo/side-by-side Open in a new tab ↗

The demo workspace: a synthetic team, real engine output, every grade with the moment that earned it. Nothing is recorded.

Alongside, never instead

Keep your stack. Peer adds the one input it cannot produce.

You keep

Recordings, trackers, pipeline signals
Scorecards and methodology, yours
Every dashboard your team reads

none of this is Peer’s job

Peer adds

A reference grade on every moment of selling, from outside your building
The quoted moment behind every grade
A calibration state on every answer your assistant gives

returned by MCP or API

Gong and Chorus hold the macro view: the whole call, what happened. Peer adds the micro view: each skill graded moment by moment against one standard, the same for everyone. Your assistant gets both.

Why scores bunch in the middle

Generic AI plays it safe. By design.

Language models reach for the consensus: your strongest and weakest calls land three points apart, and the coaching goes generic. Calibration puts something underneath built to tell the difference.

Calibration

Every grade says what it rests on.

Each grade carries one of three states. Uncalibrated: no human has yet marked a sample of this workspace’s calls. Calibrated, to a date: trained assessors blind re-marked a sample and agreement was within tolerance. Drifting: agreement on the latest sample fell below it. The state shows in the workspace and on every answer over MCP, stated as a fact, never as a plea.

Human review credits

One credit: one graded observation blind re-marked by a trained human assessor, agreement logged
Packs of 20, 50 and 100 at £4 a credit, £80 minimum
Spent two ways: a re-mark you ask for, or a sample rate you set (we suggest 10%)

the only thing Peer charges for

Why you will want it

A grade with a calibrated badge is one you can put in front of a Rep, or a promotion panel, and stand behind
A human weighs every instance in a call; the engine grades one moment per competency. Judging the same moment, the engine lands within one rung of a trained assessor 95.5% of the time, and a marked sample is how your workspace finds out where the two part
Credits are a balance, never a plan: buy them when you want the state to move

read the study → · the engine stays free either way

What this unlocks

Questions you can finally answer with evidence.

“Is the new cohort actually ramping?”

Skill grades week by week, not activity counts.

“Who are my genuinely strong Reps?”

Graded evidence that survives a promotion debate.

“Can this team open the enterprise motion?”

Skill readiness for a new play, before you bet a quarter.

“Did the enablement work?”

The same competency, measured before and after.

Free · bring your Anthropic key

Paste your key. The engine runs in one sitting.

No plan to choose and nobody to talk to. You need an Anthropic API key and a note taker; the rest is four screens. Human review credits are there when you want a grade to carry a calibrated badge.

Start free Connect it to your assistant