A Week Inside an AI Marketing Department, Day by Day

The week does not start on Monday morning for the automated part of the department. Over the weekend and overnight, the scheduled workflows keep running: account and contact records are enriched, timing signals are collected against shortlisted accounts, campaign results are read against their pause rules, inbound leads are scored and routed, and every send, reply and form fill is written back to the CRM. None of this waits for a person, because none of it needs one. It is the continuous execution the definition refers to.

It helps to keep that fact apart from a different commitment. Workflows running continuously is a property of the automation. Direct access to the CEO around the clock is a promise made by a person: a message in the portal reaches Aryeh, not a queue. The first is about what the workflows do while nobody is watching; the second is about who answers when a decision is needed. This piece is mostly about the first, and about the points in the week where the second steps in.

So when a person on the team opens the portal on Monday, the ground has already been prepared. The research shortlist has a draft version, the reply digest from the last sends is sorted by intent, the exceptions queue holds whatever the rules could not resolve, and the campaign status shows what ran and what paused. The week begins with reading, not with gathering.

Monday is for direction. The strategy already names the segments, the offers and the order of work, so the question is narrower: what does this week serve, and has anything changed since the last one? A person on the team reads the prior weekly summary and the open recommendations, checks the portal threads for anything the client raised, and sets the week’s priorities. Where a priority shifts, the CEO decides it with the client rather than letting the automation decide by default.

The first artifact of the week is the research list. Automation has already found companies that match the ideal customer profile, scored them, deduplicated them against the CRM and mapped the buyer roles inside each one. What automation has not done is decide whether the list is right. A person reads the shortlist before it goes anywhere. The accounts that fit the profile but would never buy, for a reason the criteria could not see, come off the list, and that reason is written into the criteria so next Monday’s shortlist does not repeat it. That review is the step that decides who is worth the sales team’s time.

Monday also sets the content and campaign work in motion. The content calendar says which pieces are due this week, and the campaign plans say which sends and placements are scheduled. Briefs are checked against the strategy, and any piece whose brief no longer fits is pulled before a draft is written. The day ends with a short list of what will be drafted, what will be sent and what needs a client decision.

Tuesday is drafting day, and most of the drafting is automated. Content drafts are written from their briefs in the client’s established terms, then checked against style rules and the claims policy before anyone reads them. Sequence drafts are written per segment from the account briefs. On a Tuesday that means the opening message for each segment is anchored in a fact from Monday’s shortlist, and each follow-up is drafted only once it has a new point to make. Contacts whose details failed verification overnight go back to enrichment instead of into the draft.

Then the drafts meet people. An editor on the team reads every content draft for accuracy, structure and whether it says anything worth reading, and rewrites or drops what does not. Sequence drafts are read for tone and for anything that sounds as if a machine wrote it. Any figure, quotation or claim about a competitor is traced to a source or removed. The automation produced the first version quickly; the reading is what makes it fit to carry the client’s name.

Edited drafts become approvals. By Tuesday evening the client’s portal shows one item per draft and per sequence, each addressed to the person who owns that channel and each carrying the brief and sources it came from, so the approver is not reading blind. The approver can approve, ask for a change or decline; a change request goes back to the editor, not into a debate. Silence is not approval. A draft still open on its date simply misses that slot.

Midweek is when approved work goes out and the first results come back. Tuesday’s approvals become Wednesday’s sends: each approved sequence starts its first wave, and later waves follow on the days the schedule sets, at a pace the sending domain can carry. A contact who replies or unsubscribes drops out of the remaining steps that moment. Approved content is published on its calendar date. Campaign sends and placements begin under the limits in the plan, and every step writes its own record, so the week’s log builds itself as the work runs.

Before each wave, a person reviews the deliverability signals and pauses sending when they change. Before a campaign launches, a person opens it the way a recipient will, in the inbox or feed the audience uses, and compares what they see with the plan the client approved. The check takes minutes; skipping it is how a broken link or an audience far larger than intended goes out under the client’s name.

Replies arrive throughout. Automation reads each one and sorts it by what the sender means, so an out-of-office notice or a polite decline is filed and recorded without a person touching it. What lands in front of a person midweek is the short stack the rules could not settle: someone who is interested, someone with a question, anyone whose meaning is unclear. That person answers in the client’s voice or passes the thread to the account owner, and the reply digest records which was which. The conversation itself stays with the client’s sales team.

Thursday leans toward operations. The exceptions queue is opened first: the leads, records and consent questions the rules set aside since Wednesday are resolved one by one, and where the same kind of exception has appeared on consecutive days, a rule change is drafted for the client to approve. Campaign results are read next. A campaign that paused itself overnight stays paused; a person reads why before deciding whether it resumes, changes or ends, and any change is approved as a new version of the plan, the same way the original was.

Friday closes the week with the weekly summary, and by Friday morning most of it already exists. Overnight the week’s sends, replies, publishes and campaign results have been reconciled against the definitions agreed in setup, and the draft is assembled section by section: research, content, outbound, campaigns and the pipeline movement since last Friday. Each figure in the draft points to the records it was counted from, so the person who reviews it opens the source instead of trusting the sentence.

The draft does not go out on its own. Friday’s review is a reading, not an edit pass: a person on the team opens the records behind any figure that moved, decides whether the movement is real or a counting change, and rewrites the sentence so that a reader who was not in the week understands it. A lead whose origin cannot be traced is described as untraced. The CEO reads the summary last, before it reaches the client, and holds it to the standard the client will.

The summary is also where decisions are raised. Anything the numbers argue for becomes a recommendation in the portal, written with its evidence and the decision it needs from the client. Pieces still waiting for approval, sequences held on deliverability and campaigns paused midweek are listed as what needs the client’s attention, so that the next Monday begins with the decisions already in view.

Read day by day, the split is easy to see. Automation did the work whose rule was written before the week began: the overnight enrichment and scoring, Tuesday’s first drafts and rule checks, Wednesday’s waves and publishes, the reply sorting and CRM writes throughout, and Friday morning’s draft. People carry the judgment: whether the list is right, whether the draft is worth reading, whether a reply deserves a person, whether a paused campaign should resume, and what a number means.

The line is not fixed forever. It is set during setup, where the client’s systems decide how much a workflow can do on its own, and it moves as evidence accumulates. A review that finds nothing week after week becomes a spot check; a class of exception that keeps recurring becomes a rule. The weekly summary is where those adjustments are proposed, and the client agrees to them before they take effect.

• Automated: the overnight preparation before Monday, Tuesday’s first drafts and rule checks, Wednesday’s waves and publishes, reply sorting and CRM writes throughout, and Friday morning’s draft summary.
• Reviewed by a person: Monday’s shortlist, every Tuesday draft and sequence, each wave before it goes, the interested and unclear replies, every pause, and the summary before it is sent on Friday.
• Decided by the client: Tuesday’s approvals, campaign plans, the rule changes proposed on Thursday and the recommendations raised on Friday.