14 Claude workflows, mapped to the funnel
Ordered the way revenue is actually built rather than by how interesting the technology is. Each workflow lists the data Claude needs to run it. Where that data is missing, the output is a confident guess, which is the single most expensive failure mode in this whole category.
Build a list worth working, and know why each name is on it.
Most pipeline problems start here and get blamed on outreach. A great email to the wrong account still fails.
Account research before the first touch
Prospect data + web SDR or founder, at list build time
Research {company}. I sell {product} to {buyer role}.
Return: headcount and growth trend, funding, the
tools they run that overlap with ours, roles they
are hiring for right now, and any public change in
the last 90 days that gives me a reason to reach out.
End with the single strongest opening line and say
which fact it is based on.Without a data connector this returns a polished restatement of the homepage. With one it returns things the homepage does not say.
Score a raw list against a written ICP
Prospect data + CRM Founder doing their own list build
Our ICP: {criteria}
Disqualifiers: {criteria}
Here are 50 accounts:
{paste or pull from the CRM}
Score each 1 to 5 with one line of reasoning. For
anything 3 or below, name the one fact that would
have to be different to make it a 5.
Check the CRM first: flag any account we have already
worked or lost, and say when.
Do not pad the list to fill my quota.The CRM check is what stops a founder emailing an account that churned last year. The no-padding line is what stops everything scoring a 4.
Inbound triage that respects the queue
Prospect data + CRM Whoever owns inbound, first thing
Here are today's inbound signups. For each, enrich
the company, check for an open opportunity or a
closed-lost in the CRM, and score against our ICP:
{criteria}.
Return three buckets: work now, nurture, ignore.
One line of reasoning per row.Inbound decays by the hour. This is the workflow with the clearest revenue number attached to doing it before lunch rather than after.
Earn a reply from someone who owes you nothing.
Volume hides the problem. If the first line could be sent to 500 companies, it is not outreach, it is postage.
A first touch built from one specific fact
Prospect data + web SDR, AE, or founder
Here is the research on {company}: {paste or pull}
Write a first email to {name}, {title}.
Rules:
- First sentence is about them, not us
- Reference one specific fact and name it
- Under 90 words
- One ask: a 15 minute conversation
- No em dashes, no "hope this finds you well",
no "I noticed you're hiring"
Then tell me the weakest line in what you just
wrote and why.The self-critique at the end is worth more than the draft. It surfaces the generic line you would have sent.
Rewrite a sequence so it is about them
Nothing, though CRM history sharpens it Founder who wrote the sequence at 11pm
Here are the four emails in my current sequence:
{paste}
Rewrite each so the subject is the prospect's
situation, not my product. One idea per email, one
ask at the end, each under 90 words.
After the rewrites, tell me which of the four you
would delete entirely, and why.Asking which email to delete gets an honest answer. Models will happily improve four emails when two should exist.
Follow-up that lands in drafts, not the prospect's inbox
Email + CRM AE, within an hour of the call
Read the last thread with {contact} and the CRM
notes from today's call.
Draft the follow-up. Their words, not our category
language. Reference the specific concern they raised
about {topic}. One ask at the end: the next meeting.
Save it as a draft. Do not send.Draft, never send, is what makes this safe to give a whole team in week one.
Know what is real, and stop losing deals to admin.
Deals do not die in a meeting. They die in the fortnight nobody noticed had passed.
Call prep that reads the whole account
CRM + email + Slack AE, before every meeting
Prepare me for the 2pm with {account}.
Pull the CRM record, every email thread with anyone
at that domain in the last 90 days, and any Slack
messages in the deal channel.
Give me: what they asked for last time, what we
promised, anything still open, the two questions
they are most likely to open with, and one thing I
have been avoiding asking them.
Quote directly. Do not summarise away specifics.The last item finds the deal risk. Everything above it is a summary the rep could have written themselves.
Pipeline review without the export
CRM Founder or sales manager, weekly
Across open pipeline this quarter:
which deals have had no logged activity in 14+ days,
which have a close date in the past, and which moved
backwards a stage.
Group by owner. Sort by amount. Flag anything where
the amount changed by more than 30% since creation.Each of these is a report someone built once and nobody maintained. Asking costs less than maintaining.
CRM hygiene, done by the thing that caused it
CRM + calendar + email, with write access AE, end of day
For each meeting on my calendar today with an
external attendee, find the matching CRM opportunity.
Write a call summary to the record: what was
discussed, decisions made, owner and date for each
next step.
If you cannot find a matching opportunity, list it
for me instead of guessing.The last line matters. Without it the summary gets attached to the closest-looking record, which is worse than an empty field.
Find the reason this does not close, while there is still time to fix it.
Optimism is the default setting of both reps and language models. Both need to be argued with.
Pressure-test the deal before you forecast it
CRM + email AE or founder, before commit
Here is everything on {account}: {paste or pull}
Argue that this deal will not close this quarter.
Be specific, use only what is in the record.
Then name the single piece of missing information
that would settle the argument, and give me the
exact question to ask to get it.Inverting the question is the whole trick. Asking for a balanced assessment returns a balanced assessment of nothing.
Competitive prep for a named deal
Documents + CRM + web AE, before a bake-off
{account} is evaluating us against {competitor}.
Read our battlecard in the docs drive and the notes
on this opportunity.
Give me the three places we are genuinely stronger
for this account specifically, the two where they
will beat us, and how I would handle those two
honestly.
No marketing language.Naming where you lose is what makes the output usable on the call. A one-sided card gets ignored by the rep who has to hold it.
Turn discovery into a mutual action plan
Call notes or transcript + CRM AE, after discovery
Transcript from discovery with {account}: {paste}
Build a mutual action plan: every step from here to
signature, with their owner, our owner, and a date.
Only include steps implied by what was actually said.
Mark anything you inferred with [assumed] so I can
check it with them.The [assumed] tag is the difference between a plan you can send today and one you have to rewrite first.
Protect the revenue you already won, which is cheaper than replacing it.
Churn is visible in usage months before it is visible in a conversation.
Renewal risk from usage, not from vibes
Product database, read-only + CRM Founder, CS, or account manager, monthly
For accounts renewing in the next 90 days, compare
active users and core-action volume in the last 30
days against the 30 days before.
Rank by decline. For the worst five, pull the CRM
owner and the date of last contact.
Tell me which declines are seasonal and which are
not, and say how you decided.Making the model explain its reasoning is what stops the ranking becoming a number nobody trusts enough to act on.
Expansion case, or an honest no
Support history + product usage CS or founder
Six months of tickets and usage for {account}:
{paste or pull}
Where are they working around a limitation that
{feature} would solve? Cite the specific tickets.
If the honest answer is that they are fine as they
are, say that instead of building a case.Expansion prompts produce fiction unless the model has permission to say no.
The pattern across all 14
Every prompt above contains at least one line that has nothing to do with the task. It is there to stop the model being agreeable: say so instead of inventing one, argue this will not close, mark anything you inferred, quote directly, do not pad the list.
Take those five lines into whatever you are already running and the output changes more than switching models would. What they cannot fix is missing data. A model with no access to the CRM will describe your pipeline beautifully and incorrectly.
Working in the Claude app rather than a terminal? Connecting the data takes about five minutes and no config files: the connector walkthrough is here.
To make a workflow stick past week two, save it where the team already works rather than in a doc nobody opens. Salesgear covers the two durable routes, Claude skills for sales teams and shared workspaces in Cowork, and the CRM-writing workflows in the pipeline section are walked through end to end in Claude plus your CRM.