how to build a self-improving sales operating system

i built a self-improving sales operating system in the last 6 months for our sales team

specialised agents run the weekday grind, draft in our voice, and fold learnings back into a private gtm repo. we approve what ships and what gets learned.

the stack

• ledger
• playbooks
• operators
• self-learning

ledger - CRM (attio)

system of record for companies, deals, people, tasks, live status

playbook - gtm repo (github)

the living brain with ~20 folders and 50+ skills

• playbook/: pricing, ICP/IBP, objections, POC plans, renewals, churn, competitive summary
• product/: positioning, feature truth, technical reference
• competitors/: landscape + battlecards
• legal/: contracts, DPAs, NDAs, POC licenses, partner agreements
• outbound/: ICP, sequences, campaign learnings, cold-call / email playbooks
• ops/: attio data model, quoting, call-learning, win/loss, self-heal loops
• enablement/: tone of voice, customer proof, founding-sales reference
• partners/: partner context and program planning
• marketing/: messaging, decks, collaterals, events
• hiring/: role specs
• ae-onboarding/: ramp plans
• deliverables/: customer/partner artifacts (proposals, POC plans, PDFs)
• design/: design system for decks and deliverables
• leads/, reports/, forecasts/, updates/, scripts/, mcp/: lists, analysis, dated updates, automation, external context bundle

operators - grok bot + cursor

they read ledger + playbook, do the day’s work, and update the CRM.

this is the primary daily interface for the sales team.

self learning - grok bot routines

agents review calls, wins, losses, product changes, and outbound results.

they open draft PRs that update the playbook e.g. objections, positioning, ICP, skills.

i review and merge.

next week’s operators run on a sharper brain.

operators pull from ledger + playbook to draft work.

learning writes durable improvements back into the playbook.

nothing silently rewrites the brain.

the weekday clock

scheduled routines fire in order so prep and follow-up happen without me remembering to ask.

morning (before calls)

• due-today CRM briefing
• slack “needs you” sweep
• new-customer first-touch drafts
• usage check-in drafts for prospects who signed up for our app but aren’t moving
• call prep for the day’s customer / prospect meetings
• monday: pipeline weekly update draft

during the day

• meeting brief bot: morning digest + sweeps when new calls land
• post-call follow-up drafts every couple of hours
• silent-setup / silent-using follow-ups for orgs that stalled after signup

end of day

• #sales eod standup draft ready to paste
• overnight: tidy billing / customer IDs so accounts stay linked
• friday: catch org / billing gaps before the weekend

how it feels in practice

i walk into drafts already shaped for:

• what to prep
• who to follow up
• what to post at eod

i edit, send, and move on.

the learning loops (playbook learner) are separate: they improve next week’s brain.

the weekday clock is what keeps this week’s deals moving.

the bots

daily sales ops bot

• runs the weekday sales grind on a schedule
• drafts morning briefings, new-customer follow-ups, call prep, and post-call notes
• keeps pipeline updates and the end-of-day sales standup ready to paste
• tidies billing / customer IDs so deals and accounts stay linked
• you review and send

meeting brief bot

• watches the calendar for customer and prospect calls
• builds a short brief before each one: context, history, what to push for
• morning digest plus sweeps when new meetings land

events bot

• plans conference and dinner gtm: who to meet, dinner vs 1:1
• drafts invites and follow-ups for events
• you point it at an event or list, then send and book

prospecting bot

• works ICP leads and finds ways into accounts
• researches buyers, builds lists, drafts outreach and reconnects
• you decide who gets time; it does the digging and first draft

odd-jobs bot

• handles one-shot sales tasks that don’t fit a routine
• champion lists, referral asks, event side-quests, weird deal questions
• keeps messy non-recurring work out of the specialised bots

outbound doctor

• audits outbound performance and spots what’s underperforming
• ranks the next plays: displacement, referrals, cold-call, intros, event pre-books
• you still own the sequences; it tells you what to fix or run next

playbook learner

• learns from calls, wins, losses, and stuck enterprise deals
• turns patterns into draft updates to pricing, ICP, and objection playbooks
• you review and approve so the team sales brain improves over time

full tooling

• CRM (attio.com)
• email (gmail)
• messaging service (slack)
• calendar (google)
• call recordings / notes (fathom + attio + miniti.app)
• a work board for prospecting tickets (linear.app)
• an outbound sequencer (reply.io)
• product usage / billing data (lightdash.com + getmeasure.com)
• version control (github)
• agent chats with scheduled routines (grok.com)
• a coding agent that can open draft PRs against the repo (cursor.com)

prep

• a gtm repo as the sales brain
• playbooks written so an agent can follow them
• skills / recipes per job (call prep, follow-up, standup, outbound, learning loops)
• tone-of-voice docs so drafts sound like you
• scheduled routines (without clocks you just have chatbots)
• clear bot roles: one job per bot
• draft-only rules: agents prepare, you approve
• human approval as the product: the system compounds when good drafts and good playbook edits get accepted

design principles

• agents are specialised operators (a small fleet, one job each)
• the repo is the brain
• humans only spend time on decisions that need a human
• every time the agent produces an imperfect draft, it needs to be improved with skills or context

what you get over time

• every approved draft and every learning PR improves next week’s output
• the team stops re-deriving the same objection answers
• new AEs inherit a living sales brain
• knowledge stops dying in slack threads and one-off chats