The work that fills a pipeline
is mostly not selling.
Research, list building, writing the same email again, note taking, CRM updates, chasing the deal that went quiet. It is the majority of a rep's week and almost all of a founder's selling time, and it is the part a model can genuinely take off your hands. This site covers that stage by stage: what to hand over, what to keep, and the prompts that hold up under a real quarter.
The funnel, stage by stage
Each stage fails differently, so each one needs a different kind of help. A brilliant email to the wrong account still fails, and a perfect list nobody follows up on is just a spreadsheet.
Prospecting
Build a list worth working, and know why each name is on it.
Outreach
Earn a reply from someone who owes you nothing.
Discovery
Stop pitching for twenty minutes and find what the problem actually costs.
Pipeline
Stop losing deals to admin and to fortnights nobody noticed.
Closing
Find the reason this does not close while you can still fix it.
Where the hours actually go back
Worth being specific rather than promising a transformation. In practice the time comes back in five places, and they are not evenly sized:
| Task | Typical week | What changes |
|---|---|---|
| Account research | 15 to 30 minutes per account | Three minutes, and it reads more than you would have |
| Writing first touches | 10 minutes each, more if you care | Two minutes, with the specific fact still in it |
| Call prep | Skipped, usually | One prompt, so it stops being skipped |
| Notes and CRM updates | Done badly at 7pm or not at all | Dictate the mess, get the structure back |
| Weekly pipeline sweep | An export and a pivot table | A question, answered |
None of that closes a deal by itself. It buys back the hours you were going to spend on admin and gives them back as selling time, which is the only honest claim in this category.
The pattern behind the prompts that work
Nearly every prompt on this site contains 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.” Without permission to return nothing, a model will manufacture a reason to follow up.
- “Argue this deal will not close.” Inverting the question beats asking for a balanced view.
- “Mark anything you inferred.” Separating fact from inference is what makes output safe to forward.
- “Quote directly.” Summarising is where the specifics that would have won the call get smoothed away.
- “Name the weakest line you just wrote.” The self-critique is usually more useful than the draft.
Add those five to whatever you already run and the output changes more than switching model would. A role by role view of where each fits, from SDR to VP, is in this guide to using Claude for sales.
If you run the team rather than a patch
Two problems that belong to whoever designs the system, not to the rep working the deal. Both are where most of the damage gets done.
Playbooks and prompts
14 prompts
By role: SDR, account executive, sales manager, customer success. Full text, copy and run.
6 playbooks
Sequences rather than single prompts: your first week, rolling it out, guardrails before anyone sends.
Past copy and paste
The point where prompting stops scaling and the model needs access instead.
What this will not do
It will not close a deal. It will not tell you whether an account is worth chasing if you have never written down what a good account looks like. It does not know your ICP, your pricing exceptions, or which competitor genuinely beats you on integrations, unless you tell it or connect something that can.
And it will not fix a broken process. If activity logging is inconsistent, a call brief built from the CRM inherits the inconsistency and states it confidently. Judgment, trust, and timing stay yours. Everything else is negotiable.