Key Takeaways
- Pipeline hygiene is a different job from selling. This post is about one specific workflow: the recurring audit that keeps the CRM honest. Not the broader "AI for sales" picture, which we covered in Your Sales Reps Spend Half Their Day Not Selling. This one is narrow on purpose.
- Most CRM data is stale on Friday and worse on Monday. Deals sit at the same stage for weeks because nobody disqualifies them. Next-step dates expire and never refresh. Close dates slip silently. By the forecast call, the data is fiction.
- An AI coworker can audit every open deal every Friday and surface the rot. Viktor reads HubSpot or Salesforce, cross-references Gmail and Slack for last-touch evidence, and produces a per-rep cleanup list with specific recommendations: kill, push, draft a re-engage email, or escalate.
- The forecast accuracy gain is immediate and permanent. A pipeline that is audited weekly produces a forecast that holds up. Sales leaders stop guessing and start running the business off real numbers.
- The hidden gain is rep time. Reps who don't clean their CRM at all suddenly have a clean CRM. Reps who clean it manually get those four hours back. Both win.
The short version
Every sales team has the same problem. The CRM is the source of truth, except when it isn't, which is most of the time. Reps add deals quickly, forget to update them, and the system fills with phantom pipeline. Half the deals are dead. The other half have a "next step: schedule call" note from three weeks ago.
The fix is supposed to be discipline. Friday hygiene block. RevOps reminders. Stage-exit criteria in writing. Some of that helps, none of it sticks. Sales reps do not become CRM admins, and they shouldn't.
The actual fix is to put an AI coworker on the audit. Every Friday at 4 PM, Viktor pulls every open deal in the pipeline, reads the recent Gmail thread for each, checks Slack for any reference, looks at the last-activity date, and produces a per-rep list: deals to kill, deals to push, deals where Viktor has already drafted the re-engagement email. The rep spends 15 minutes reviewing the list instead of two hours auditing manually (or, more honestly, not auditing at all). Monday opens with a clean board.
This is the workflow. Below is what it actually looks like.
What "pipeline hygiene" actually means in 2026
The phrase has gotten loose. Here is the operational definition we use, the only one that matters:
A clean pipeline is one where every open deal has:
- A close date that is in the future and has been reviewed in the last 14 days.
- A next-step note that names a specific action and a specific date.
- Activity (email, call, meeting) in the last 21 days, or a documented reason it is paused.
- A stage that matches the actual customer state, not the rep's hope.
- An amount that reflects the current scope, not the original pitch.
If a deal fails any of those, it is dirty. Most pipelines are 40-60% dirty. The forecast built on top of that pipeline is a stack of guesses.
The Friday audit catches the rot before the Monday review. That is the entire game.
The Friday hygiene cron, in plain prompt form
You set this up once. It runs every Friday at 16:00 local time. The output is a per-rep DM and a leadership summary in #revops.
@Viktor every Friday at 16:00 Europe/Amsterdam:
Pull every open deal in HubSpot pipeline "New Business" with stage NOT in
("Closed Won", "Closed Lost"). For each deal, evaluate these checks:
1. close_date: is it in the past? Flag as "PAST CLOSE."
2. next_step (custom field): is the date in the past or the field empty? Flag
as "STALE NEXT STEP."
3. last_activity_date: is it >21 days ago? Flag as "DARK."
4. stage_change_date: is the deal in the same stage >30 days? Flag as "STUCK."
5. For each "DARK" deal, search Gmail for the primary contact's email in the
last 30 days. If found, propose updating last_activity_date with the
real date.
Group results by deal owner. For each owner, post a DM in their Slack
containing:
- A bulleted list of every flagged deal with the flag(s), deal value, and
current stage
- A proposed action per deal: KILL (move to Closed Lost with a reason),
PUSH (update close_date to a more honest one), CALL (rep should reach
out, draft email attached), or REVIEW (ambiguous, manager call)
- A pre-drafted "re-engage" email per CALL deal (110 words max, plain text,
references the last meaningful exchange)
- Three buttons in Slack: "kill all", "push all", "review individually"
Also post a summary in #revops listing per-rep dirty-deal counts, total
dollar value of dirty pipeline, and the three biggest deals flagged.The first time this runs, the per-rep DM is uncomfortable. Most reps are surprised at how dirty their own pipeline is. By Friday three, the muscle is built and the dirty count drops 60-80 percent.
What "drafting the re-engagement" actually means
The hardest deals to clean are the dark ones. The rep doesn't want to admit the deal is dead, but doesn't want to send a bad re-engagement email either, so the deal stays in the pipeline forever.
Viktor solves this by writing the email. Not a template. A real draft that reads the actual thread.
@Viktor for deal {{ deal_id }} (HubSpot), draft a re-engagement email to
{{ primary_contact }}:
1. Read the last 5 emails in the Gmail thread with this contact.
2. Identify the last specific commitment either side made (a date, a deliverable,
a follow-up). If none exists, identify the last topic of substance.
3. Write a 110-word email from the assigned rep that:
- References the specific commitment or topic by name (no "circling back",
no "just checking in")
- Includes one concrete update or new piece of value (a release, a case
study, a relevant data point)
- Asks one specific question that requires a yes/no
- Is signed in the rep's voice (match the prior threads)
4. Output the draft as a Gmail draft (not sent). Post a Slack message in the
rep's DM with a preview and a "send" button.
5. If the contact does not reply in 7 days, flag the deal as "RE-ENGAGE FAILED"
in the next Friday cron and propose KILL.This single workflow is the difference between a pipeline cleanup and a pipeline cleanup that actually closes deals. Some of the dark deals come back to life. Most do not, and the rep gets the conviction to mark them lost.
The leadership view: a real forecast, finally
The Friday cron also posts in the leadership channel. The shape:
Pipeline Hygiene Report, Friday {{ date }}
Total open pipeline: $4,820,000
Clean pipeline: $2,140,000 (44%)
Dirty pipeline: $2,680,000 (56%)
- Past close date: $920,000 (43 deals)
- Stale next step: $610,000 (28 deals)
- Dark >21 days: $830,000 (37 deals)
- Stuck same stage >30d: $320,000 (12 deals)
Top 3 dirty deals by value:
1. Acme Corp ($240k, dark 38 days, owner: Sarah)
2. Globex Industries ($180k, past close 21 days, owner: Mike)
3. Initech ($165k, stale next step, stuck Stage: Proposal 41 days, owner: Jen)
Per rep:
Sarah: $720k pipeline, 62% dirty (7 deals flagged)
Mike: $1.1M pipeline, 48% dirty (11 deals flagged)
Jen: $890k pipeline, 41% dirty (6 deals flagged)
...A VP of Sales who walks into the Monday forecast call with this in hand has a different conversation than one who doesn't. The number on the dashboard is no longer the real number. The clean number is the real number, and that is what the forecast uses.
Why this works for HubSpot, Salesforce, and Attio equally
The pattern is the same across CRMs because the data shape is the same: deals, stages, dates, owners, activities, contacts.
Viktor connects to all three through OAuth and reads the deal records directly. The prompt above is written for HubSpot field names; for Salesforce, swap close_date for CloseDate, next_step for NextStep__c if you have a custom field, and last_activity_date for LastActivityDate. For Attio, the field names live in your specific schema and Viktor reads them automatically.
The cleanup actions (move stage, update field, draft email) all happen through the same OAuth connection. There is no middleware. Viktor is the operator.
Where the rep's role changes
The skeptical reader is asking: "If Viktor is auditing my pipeline, am I being audited?"
The answer is no, but the framing matters. Viktor is the rep's coworker, not the manager's spy. The default setup is that the per-rep DM is private to the rep. The leadership channel posts aggregate counts and the top 3 dirty deals, not per-rep dirty-by-name lists.
This is a deliberate choice. Pipeline hygiene works when reps trust the tool. The moment Viktor becomes a snitch, reps game it (artificial activity logs, fake next-steps to dodge the audit). Keep the tool on the rep's side. The aggregate dashboard is enough for leadership to see the overall health and have a conversation with reps who are consistently dirty.
If your culture allows full transparency (some do, some don't), you can change the setup to post per-rep details to leadership. We don't recommend it for the first 90 days. Build trust first.
What the manager does with this
The Monday morning routine for a sales manager who has the Friday cron:
- 8:00 AM: open Slack, read the leadership summary from Friday
- 8:10: identify the 2-3 reps with the dirtiest pipelines this week
- 8:20: 1:1 with each, not to scold, to help (most dirt is "I don't know what to do with this deal" not "I'm hiding them")
- 9:00: forecast call, with a clean number
Compare to the prior routine:
- 8:00 AM: open the CRM dashboard, see the dirty number, sigh
- 8:30: ask the RevOps person to pull a hygiene report (they will deliver Wednesday)
- 9:00: forecast call, with a number everyone in the room privately knows is wrong
- 9:30: spend the rest of the day chasing reps for updates that should have been there
The first version is a manager doing manager work. The second version is a manager doing data-cleanup. The Friday cron is what makes the difference.
A separate cron: stale-deal triage at quarter end
The Friday weekly cron is the steady state. There is also a useful one-time cron that fires 14 days before quarter end:
@Viktor 14 days before each quarter end:
For every open deal with close_date in the current quarter, do a deeper read:
1. Last 10 Gmail emails with the primary contact, last 60 days.
2. Last 5 Slack messages in any shared channel referencing this account.
3. Last 30 days of CRM activity (calls, meetings, notes).
For each deal, output one of:
- "WILL CLOSE" with a confidence (0-100%) and the strongest evidence
- "WILL SLIP" with the reason and a proposed new close_date
- "WILL DIE" with the reason and a recommendation to mark Closed Lost
- "UNCLEAR" with a list of 3 questions the rep should answer
Post the per-deal verdicts to each rep's DM. Post the aggregate (will-close
total, will-slip total, will-die total) to #revops.
The intent of this cron is NOT to predict the forecast. The intent is to
force a real conversation 14 days before the quarter ends so reps can act,
not justify.The teams that run this find that 20-40 percent of "will close" deals at the start of the cron get reclassified after the rep does the conversation. That is the value: the conversation happens earlier and the reclassification happens before it ruins the quarter.
Safety and approval
Pipeline hygiene is internal data work, but the actions can affect customer-facing communication, so the same review-first principle applies.
Hard rules for the cron:
- No deal-stage changes auto-applied. Viktor proposes; the rep clicks. The "kill all" button in the DM moves the rep through the list one at a time with a confirm.
- No emails auto-sent. Drafts only. Even the re-engagement emails wait for the rep to click send. This is non-negotiable, see Don't Let Your AI Agent Act Without Asking for why.
- Internal Slack posts and Notion log entries are auto-create. Internal artifacts. Auto-approve.
- Quarterly review of the prompt. Sales motions evolve. The thresholds (21 days dark, 30 days stuck) should match your sales cycle, not a generic default.
The cost of getting hygiene wrong is a deal incorrectly killed and a customer incorrectly contacted. The review-first default contains both.
What this looks like at three scales
A 5-rep team. The Friday cron lives in one #revops channel and the manager doubles as the hygiene reviewer. Pipeline cleanups happen in the Friday-Monday cycle. Forecast accuracy improves within two weeks.
A 25-rep team. Per-rep DMs become essential. The leadership channel gets the aggregate view. The RevOps lead reviews the prompts monthly and tunes thresholds per pod. Forecast variance drops from 25-35% to 8-12% within a quarter.
A 100+ rep team. Per-pod aggregations make sense (regional, product line, segment). The cron runs in waves. The RevOps function shifts from manual hygiene chasing to prompt and threshold tuning. The CRO sees a real forecast for the first time.
Frequently Asked Questions
Will reps resent the cron?
Not if you set it up as their tool, not the manager's tool. The first three weeks include a per-rep onboarding where they see the DM, edit the thresholds, and own the cleanup pace. Reps usually become advocates within a month because the dirty deal count drops and their pipeline feels lighter.
Will the AI mark something Closed Lost that should have been kept?
The cron never auto-marks. It proposes. The rep confirms. The default is conservative, "REVIEW" rather than "KILL," for any ambiguous case. False positives in the cleanup list are normal; false negatives (missed dirt) are the bigger risk and that is what the cron is built to catch.
Does this work if our team uses Pipedrive, Close, or another CRM?
Yes. The pattern is the same. Pipedrive has expected_close_date, next_activity_date, update_time. Close has equivalent fields. Viktor reads the schema on connect and adapts the prompt. The leadership summary format does not change.
Can we make the leadership channel post per-rep dirty-deal totals by name?
Yes, but we recommend not for the first 90 days. The trust dynamic matters. Once the team understands the cron is on their side, you can add the per-rep view if your culture supports it.
How does Viktor decide a deal is "dark"?
The default is no email, no meeting, no call, no Slack reference for 21 days, with no documented "paused" reason in a custom field. You can change the threshold. Some teams use 14 days for transactional motions, 35 for enterprise.
What about deals where the contact is on parental leave or PTO?
The prompt above looks for a "paused" reason field. If your CRM has one (or you add one), reps can mark deals as legitimately paused with a return date. The cron skips those until the date passes.
How long until forecast accuracy improves?
Two weeks for visible improvement, one quarter for the discipline to lock in. The biggest delta is in week three, when the historic dirt has been cleaned and what's left is the live pipeline. From there, hygiene becomes a maintenance function, not a recovery function.
