Documentation

How it works once it is running: what is in your workbook, what happens every morning, how people answer, and the things you will want to do yourself.

What you need

A Google account, a program, and about thirty minutes for the first conversation. That is the whole list. There is nothing to install, no seats to buy, and nobody on your team creates an account at any point.

The workbook is created in your Google Drive, under your account, from the first minute. It is never in mine and never moves.

Your first conversation

You describe the program the way you would describe it to a new joiner. What is being replaced, which parts of the business it touches, who runs each one, and the date you are working to. The agent asks about the gaps as it goes rather than at the end.

Where the program is a known platform, it starts from the model for that platform and asks you to correct the part that is genuinely yours. Where it is something you built, it starts from a baseline instead: who, what, when, and the artifact each stage owes.

You do not need a charter or a plan. If you have one, send it and it gets read first, but the conversation is enough on its own.

What arrives

One Google Sheet, populated in a single pass. Nothing is written until the whole plan validates, so you never get half a program with the other half missing.

The tabs, and what is in them

TabWhat it holds
DepartmentsThe parts of the business the program touches, and who runs each one.
Process AreasWhat actually changes inside those departments. This is the level fit gaps hang off.
OutcomesThe things that have to be true for go live. Each one has an owner, a date and a stage.
MilestonesThe stage gates the outcomes roll up to.
DependenciesWhat blocks what. A real graph, not a column of notes.
Fit GapCurrent state against future state, with each gap linked to the outcome it blocks.
PeopleEverybody the program can ask a question of, with their cover and escalation path.
RolesWhat each person is accountable for, separate from who they happen to be.
RAIDRisks, actions, issues and decisions, with the decision ones dated and owned.
Agent LogWhat the agent did each morning, in the order it did it.
Agent StateWhat it is still carrying: things asked and never answered.
My ActionsWho was written to, about what, and what came back.

Outcomes, owners and dates

An outcome is something that has to be true before go live. Every one carries an owner, a date, a stage and a status. Those four are what almost every finding is about: an outcome with no owner, no date, a date that contradicts another one, or a status nothing evidences.

Status is set in the workbook using its own words: On track, Slipping, Blocked. The people answering by email see plainer ones. Both refer to the same thing.

In the workbookWhat the person reads
On trackAll good
SlippingRunning late
BlockedStuck

Dependencies and fit gaps

Dependencies are held as a graph rather than a list, so a change to one date travels to everything downstream of it. A fit gap with no date of its own still becomes an edge in that graph, which is why a gap nobody has dispositioned can block an outcome that looks fine.

This is the one that catches people out. A gap has no date, so it never appears late, and three steps later it is the go live date.

People, cover and escalation

Each person has what they own, who covers them, and who it escalates to when they do not answer. Load is read off the plan rather than from a timesheet, so the agent can tell you who is carrying too much and which items would stop if one person were away for a week.

An item with nobody to escalate to is itself a finding. So is a department with no owner.

The three agent tabs

These are the agent’s own working tabs. You can read them; you rarely need to.

  • Agent Log. What it did this morning and in what order, with counts.
  • Agent State. What it is still carrying. Things raised weeks ago that nobody has answered live here, which is why nothing drops off because a week went by.
  • My Actions. Who was written to, about what, and what came back.

What happens before eight

Every working morning, in this order:

  1. Reads every row in the workbook.
  2. Runs fifty-two finding rules across all of it.
  3. Works out which findings need a person, and which person.
  4. Sends one brief to each of those people, and nothing to anyone else.
  5. Advances the escalation clock on anything still unanswered.
  6. Writes what it did to Agent Log.

It runs whether anyone opens the workbook or not. There is no dashboard to check, and checking one is not a step in this list.

The brief

Each person gets only their own items, never the whole program. Every item says why it matters, usually who is waiting on it. If somebody has nothing that needs an answer, they get no email at all, which is the main reason the ones they do get are read.

Answering

Three buttons in the email: All good, Running late, Stuck. One tap, from the phone, with nothing to log into. The link in the email is the credential.

The same link opens that person’s own live list, if they want to see everything they own rather than just today’s question.

Escalation

Anything unanswered keeps its clock running and gets asked again the next morning. Only a real answer resets it, and only All good resets it fully. Running late and Stuck both keep it visible, because both mean somebody still has to do something.

Past a threshold it escalates to whoever that person escalates to. If the escalation path is empty, that becomes the finding.

Editing by hand

It is a Google Sheet. Type over any cell, any time, and the next run takes your value as the truth. You do not have to tell the agent, and you do not have to undo anything it did.

This is the intended way to work, not a fallback. The agent drafts; you have the last word on every row.

What it proposes, what it applies

Scheduling is proposed, never applied. When dates need to move, you get the proposal and the reasoning, and nothing changes until you accept it.

Downstream recalculation is different: when you move a date yourself, the things that depend on it are recalculated automatically, because leaving them stale would be worse than moving them. Anything that needs a judgement call comes to you instead.

People joining and leaving

Add a row, and they are in from the next morning. Remove a row, and everything they owned becomes unowned, which is deliberate, because unowned work is a finding and should not quietly vanish with the person.

Six things you will want to do

Add somebody to the program
  1. Add a row to People with their name and email.
  2. Put their name in the owner column of whatever they own.
  3. They get their first brief the next morning. Nothing else is needed. No invitation, no account.
Change a date and see what moves
  1. Type the new date straight into the Outcomes row.
  2. The next run recalculates everything that depends on it.
  3. Anything that now conflicts is raised as a finding, rather than quietly overwritten.
Mark something complete
  1. Set the status on the outcome.
  2. Put the evidence in the evidence column: a link, a document, a ticket reference.
  3. Without evidence the agent keeps treating it as open, on purpose.
Dispose of a fit gap
  1. Open Fit Gap and set the disposition: in scope, out of scope, or deferred.
  2. A gap with no disposition keeps blocking whatever waits on it.
  3. Deferred gaps come back when their date arrives.
Pause the morning brief
  1. Tell the agent in chat, or set the flag on the program row.
  2. Nothing is sent while it is paused, and nothing is lost.
  3. The escalation clock keeps counting, so you find out what built up.
Hand the workbook to somebody else
  1. It is a Google Sheet in your Drive. Change the owner in Google Drive.
  2. Tell me so the agent writes to the right place.
  3. Nothing has to be exported, and nothing lives anywhere else.

Who can see what

The workbook is yours, in your Drive, shared with whoever you share it with. People answering by email see only their own items through their own signed link. They never see the workbook, and they never see anybody else’s items.

The agent reads and writes your workbook and sends your email. It touches nothing else.

If something looks wrong

Two things worth checking before anything else. Agent Log tells you whether this morning’s run happened and what it found. Agent State tells you what it thinks is still outstanding, which is usually where a surprise comes from.

If it still looks wrong, email hello@heypm.ai with the program name and roughly when you noticed. I read those myself.

Something not covered here? These are the questions that actually come up. If yours is not one of them, ask and I will answer it, then add it to this page.
Ask a question