Every Monday, someone opens four tabs, exports last week's numbers, pastes them into a spreadsheet, compares them with the week before and writes a paragraph about what changed. The paragraph is where it goes wrong: a busy person fills a gap with a reason that sounds right, or misses that the sheet stopped updating on Thursday.
A report is one of the best jobs to hand to a workflow, because it's the same job every week, as long as the workflow does it carefully. Here's the rule Lodol's ready-made reports are built on, six you can use today, and how to build your own.
The rule: compute the numbers, then describe them
A report you can trust keeps two jobs apart. The numbers come from steps that calculate: read the rows, total them, compare them with last week. Those steps work the same way on every run, so the total is the total. AI only describes what's already been worked out. If a report has commentary, the AI writing it is given the finished figures and told to quote only those, and where the data can't say why something moved, it doesn't guess.
Two habits finish the job. The figures go out with the commentary, so anyone can check a sentence against its number. And a report with nothing to read says so, because one that goes quiet when its source breaks looks exactly like a slow week.
1. A Monday briefing from your metrics sheet
The workflow: at 7:00 every Monday, Weekly metrics briefing by email reads the Metrics tab of your spreadsheet, one measure per row with this week's value and last week's, and emails you a short briefing on what changed, which way and by how much, with the numbers underneath.
What it won't do: explain why. The AI is told not to give causes, because a spreadsheet can't tell you that and a plausible invented cause is worse than none. Nor does it estimate or project, and a blank cell makes a measure incomplete, not zero. An empty sheet gets an email saying so.
2. A chart of the number everyone watches
The workflow: at 8:00 every Monday, Weekly chart of how a metric has moved reads the history you keep in Google Sheets, draws the measure you name as a line chart of its most recent readings, and emails you a link to the chart with two or three sentences on how the numbers moved.
What it won't do: forecast, set a target, give advice or explain. The readings are printed under the chart, so nothing in the note has to be taken on trust, and a row with no readable number is listed as left off, never counted as zero.
3. The week's new leads, from your CRM
The workflow: at 7:00 every Monday, Weekly HubSpot lead report to Slack posts one report to your channel: how many contacts were created in HubSpot last week, whether that's up, down or level on the week before, the split by lifecycle stage and by source, and the newest leads by name.
What it won't do: miscount. Duplicates are dropped, every contact's creation date is checked against the week, and the headline number comes from HubSpot's own total. If not every contact could be pulled back, the report says so.
On Salesforce, new in the library this week, Nudge reps about stalled Salesforce deals sends each rep a Slack message every Monday listing their open deals with no activity or stage change in two weeks, biggest first. If Salesforce can't be read, the run fails rather than looking like a quiet pipeline.
4. Search rankings, tracked week by week
The workflow: at 7:00 every Monday, Catch the pages slipping in Google each week runs one Google search for each page you list in a sheet, adds where your site sits to a rank history, and posts a summary to Slack. A page that has lost more ground than the threshold you choose gets a specific recommendation, posted as a reply in the thread.
What it won't do: give advice it can't back up. Each slipping page is opened first, because a broken page or a redirect needs fixing before any content advice, and the AI works only from the positions it measured, the results page and your page itself. The summary goes out every week, even when nothing moved. For the raw numbers, Weekly search performance export to Sheets appends last week's queries, clicks, impressions and average position, one row per query.
5. Cash and runway, before the week starts
The workflow: also new this week, Start every week knowing your cash and runway reads your Mercury accounts at 7:00 every Monday and sends one email that leads with cash, months of runway and monthly burn, followed by last week's money in and out against the average of the four weeks before, where most of it went, and every account's balance.
What it won't do: send a figure it isn't sure of. There's no AI in it: burn is money out minus money in, averaged over the last three full months, and runway is cash divided by burn, rounded down. Moves between your own accounts are left out, and if Mercury can't be read, no email goes out and the run is marked as failed.
6. What's still open, posted where the team looks
The workflow: once a week, Weekly open items digest to Slack reads your tracker sheet and posts every row not marked done to a Slack channel as one digest, each item with its owner and status. On top sit two or three sentences on what's blocked, what's overdue and who is carrying the most, written only from the list.
What it won't do: go silent. A week with nothing open still gets a message saying so, and if the summary can't be written, the list goes out anyway with a note. A row missing its task or status is skipped and logged, so the list never shrinks quietly.
Build the report your team actually reads
Add a template to your workspace and it's yours to change: point it at your own tab, move it to Friday, send it to another channel. To build one from scratch, the same shape works almost every time:
- Start on a schedule, such as every Monday at 7:00 or the first Monday of the month, in your own time zone.
- Read the source: a Google Sheets or Excel tab, HubSpot or Salesforce, Shopify orders, Google Analytics, or the reports in QuickBooks Online.
- Do the arithmetic with steps that calculate. Lodol's own steps total, average, count and group rows, and a few lines of Python in a Code step handle anything they don't.
- Add commentary only if it helps. Hand an AI step the finished figures and tell it to quote only those and not to guess at causes.
- Send it where people already read, by Gmail, Outlook or Slack, with the figures underneath, and with an If step that sends "no rows this week" when the source is empty.
Or describe it to Claude or ChatGPT, such as "every Monday at 7:00, total last week's orders on my Orders tab by region, compare each with the week before, and email me the table with two sentences that only quote those totals", and it builds the workflow in your workspace for you to check and turn on.
What it costs
Lodol charges per run, one pass of a workflow however many steps it takes. A weekly report uses four or five runs a month, so the six above come to no more than 30 between them: a sliver of Builder's smallest tier, 2,500 runs for $18 a month. The Free plan's 100 runs a month is plenty to try one or two first. Seats are unlimited, so everyone who reads a report can be shown the workflow behind it.
The writing steps run on your own Anthropic account, and the rankings searches on your own Serper account, each billed at its own rates; Lodol adds nothing.
Start with the report you dread most.
Add Weekly metrics briefing by email to your workspace and point it at your metrics tab, or browse the management and operations templates. No workspace yet? Start free.