Sales and business systems
A GTM engine: account research, daily signal scans and a memo in my inbox before 8am
Give it a target list. It enriches every account, scores them on timing rather than size, watches their websites and job boards every day, writes the morning memo, and drafts the outreach that reps actually send.
Most AI content is about making machines do machine things. The bigger win is pointing it at commercial work, which is where I have spent my career. This is the system I built for it.
Five layers, built in order
1. Enrichment. A company name goes in. The pipeline scrapes the homepage, about page, services and careers pages, summarises what the business actually does, and scores it against the ICP with the disqualifiers written just as carefully as the criteria. Everything below the threshold is cut before a single credit is spent on contact data. Raw lists run at roughly half fit; this runs far tighter, because qualifying before enriching is the whole trick.
2. Scoring on timing, not fit. Recent funding, hiring for the function I sell into, a new executive in the seat, a technology on the stack that pairs with the offer, headcount growth. An account carrying three of those is worth several times the effort of one carrying none. Scores decay, because a funding round from last year says nothing about this quarter.
3. Daily scans. Every morning the engine re-reads the websites, careers pages and news of the accounts on the list and diffs them against yesterday. New job posting for a data role. Pricing page changed. Executive announced. The signals arrive before anyone asks for them.
4. The memo. By 08:00 the day’s memo is in my inbox and my Telegram: what changed, which accounts moved up, what to do about it, and three drafted messages ready to send. The memo cites the page it read, so I can check it in ten seconds.
5. Outreach. Per-persona sequences generated from the account’s own language, with the value hypothesis written for the role, not the company. Nothing sends automatically. A rep reads, edits, sends. The system’s job is to make the first draft good enough that the edit takes a minute.
Cheap model first
A small model sorts the obvious misses at cents per thousand records. The frontier model only reads what survives. Pointing an expensive model at fifty thousand raw records is a bill you did not need to pay, and the same rule runs through every system on this site.
The bigger version
The same stack produced a full market-entry plan for a consumer drinks concept: two-market analysis, a regulatory risk that would have killed one product line, unit economics and a phased roadmap. The output that mattered was the three gaps it exposed: no customer acquisition cost, no product liability cover, and a personal legal check before incorporating.
The principle
AI is at its best when it tells you what you have not thought about. The job is not to ask it for a plan. The job is to ask it what is wrong with the plan.