Coaching
Every 1 on 1 starts with the facts already on the table.
The hardest part of a 1 on 1 is the half hour before it, spent trying to remember where this rep was last month and what the two of you agreed. Maddox assembles that from what actually happened: the numbers as they stood at the last one, what was agreed there, the coaching received since, the numbers now, and what would most change the next set. The manager still runs the conversation. They just do not have to reconstruct it first. Underneath every page below is the same loop — prepare from the record, hold the session, capture what was committed to, then judge those commitments against evidence rather than recollection — and a single resolver that decides who is allowed to see whom, so that no screen in the pillar has its own private opinion about your org chart. Every BDR call and every AE meeting is graded against your own rubric as it lands, and a recorded 1 on 1 is graded, summarised for the person it was with and read by the next prep. It is built for teams where the people who sell also keep the account afterwards, which is most MSPs and most teams that sell anything worth keeping.
In the product
The coaching cards behind a graded call.
The chain
Where they were, what happened, where they are
Each of these is a separate reader with its own rules about what counts as evidence and what counts as an honest blank. Fifteen of them feed a single prep document, and not one involves a model: every figure on the sheet is either copied straight from the record or worked out from numbers already in it, so the facts are settled before a word of prose is written.
The record
What the coaching is built on
Coaching is only as good as the evidence under it. These pages are about that evidence and who is allowed to see it. Visibility is decided in exactly one place, which delegates the question “who reports to me” to the service that already owns it — because two answers to that question would disagree the first time an org chart changed, and the disagreement would be invisible until one manager saw a report another could not.
Accountability
Both sides of the conversation
A coaching system that only measures reps is a performance-management system wearing a friendlier name.
By role
It is a different job for each of them
These are not filtered views of one page. The metric sets, the goals, the ranking and in one case the whole document are different. A metric that belongs to another role cannot enter this one’s score at all: it is refused when a person tries to set it and recorded as off-role when the system proposes it, because judging a BDR against an AE’s number is worse than judging them against nothing.
The mechanism
The loop every page below assumes, and the one resolver that guards it
The pages below describe parts of a single cycle. It is worth seeing the cycle once, along with the two rules that stop a coaching system quietly becoming a surveillance one.
Prepare, hold, commit, judge — and then prepare from that
A prep document is assembled from the record before the meeting. The meeting happens. What was agreed becomes tracked commitments with a due horizon. Next cycle, those commitments are judged against evidence rather than recollection, and that judgement is part of the next prep. The loop closes, which is the only reason a follow-through rate exists at all.
Every link is a separately implemented reader with its own rules about what counts as evidence and what counts as an honest blank, and that is deliberate rather than accidental architecture: a single service that did all of it would have one opinion about missing data, and missing data means something different at every step.
Who may see whom is decided in exactly one place
Every coaching read turns a viewer into a scope decision, and there is one resolver that does it. It does not work out who reports to whom — it delegates that to the service that already owns the org chart, because re-deriving any part of it would be a second answer to the same question. The two would disagree the first time somebody changed a reporting line, the disagreement would be invisible, and it would show one manager a report another manager could not see.
It resolves to one of two modes. Team scope means lists are the authorised set and comparisons may name people. Self scope means you see yourself and team averages, never a named teammate. Self scope is treated as a fact about the resolved set rather than a permission to check separately, so a manager whose team is empty and a rep with no team permission land in the same place by the same route.
One deliberate sharp edge: where a workspace has never had its sales permissions seeded, the underlying resolver throws rather than returning nothing, and this one does not catch it. Catching would be forming a second opinion about the canonical answer. A workspace in that state has a provisioning fault that should surface, not a panel that should quietly empty.
The prep is shared, so what it may quote is constrained
Because the document goes to the person it is about, first-party material is permitted and third-party verbatim is banned outright — another rep’s words, another person’s development items, anybody else’s prior narrative, including in anonymised paraphrase. The rule is structural rather than reviewed: every read in the assembly is scoped to the subject themselves, so a read that cannot be scoped to the subject cannot be added to the document. Showing somebody what they themselves said is not a leak; it is the entire point of an accountability conversation.
Questions
The things people actually ask.
Does the AI decide what to say about my reps?
No. The agenda is assembled by deterministic code with no model involved at all, from readers that quote records. One bounded narration pass turns that assembled evidence into prose, under rules that require every claim to cite a specific finding. It cannot introduce a fact the assembly did not contain, and where the prose breaks a rule the pass is re-asked once rather than accepted.
What happens when there is not enough data?
It says so. A metric pair where either side is below the sample floor is surfaced and labelled rather than quietly dropped, and anything excluded is counted and reported so the omission can be stated out loud.
Is this performance management?
It is closer to the opposite. The same engine that says a rep did not implement what they were coached on will also say that a manager never coached it, and route that to the manager. Those are different failures belonging to different people.
Can a manager edit what the AI suggested?
Yes, and the original is preserved forever alongside the edit, with an append-only history of every change. An AI suggestion that has not been confirmed by a manager is never shown to the rep at all.
Do we have to use the whole thing?
No. Scorecards, development plans and one on ones each stand on their own, and several of the more opinionated behaviors have their own switch.
Bring a 1 on 1 you already ran.
Compare what you remembered about it with what the record actually holds. That gap is the whole pitch, and it is usually wider than a manager expects.