Independent AI consulting · San Diego
A go-to-market system for small sales and marketing teams. One shared brain does the work that surrounds a sale (finding the lead, preparing the call, following up, being found when a buyer asks an AI) so the people can spend their hours on the conversation itself.
In development, October 2026. Parts of it run every night for one sales team today. The rest is either built and in testing, or planned and written down. Every feature below carries one of three marks, because a product page that does not say which is which is asking you to take it on faith.
Think about the hour after a good sales call. The rep has a transcript nobody will reread, a CRM record that needs updating, a follow-up email that ought to sound like the company rather than a template, and three more calls on the calendar. Very little of that hour is selling. It is the clerical work around selling, and it is exactly the kind of work AI does well, provided somebody builds it carefully and decides where it should stop.
The toolkit began as that build for one telecom sales dealer, first for the founder and then for the reps he hires. It is becoming a product because the next team needs the same thing, and building it twice by hand would be the expensive way to learn that.
Every agent in the toolkit reads from the same place: who the company is, whom it sells to, what it sounds like, and what it can and cannot promise. That context is written once, when a team comes on board, and each feature plugs into it instead of carrying its own copy. When the founder changes the pitch, every draft changes with it.
The model layer runs on Claude. Each function is exposed over MCP, so a rep can call it from inside Claude in the middle of a working day, and the same functions run on a schedule overnight without anyone asking.
The call transcript goes in. Structured deal data comes out (next steps, objections, budget, timing, who else is involved), followed by a draft follow-up in the company’s voice, in English or Spanish. Some kinds of email are deliberately left to a person, and for those the toolkit drafts guidance rather than the email.
Before a call, it pulls the deal’s history from the CRM and turns it into a short brief: where things stand, what was promised last time, and what to ask.
First-touch drafts built from the same voice profile, so a new rep’s first email does not read like a new rep’s first email.
Checks a lead against the customer profile and against where the service is actually available, before anyone spends a call on it. Being connected now to the live availability data.
Scores each open deal against a written rubric and says why, so a pipeline meeting starts from evidence rather than from optimism.
What the call surfaced goes back into the CRM on its own, with a person approving each change, so the record is current without anyone typing it in.
A new hire gets the playbook, the voice, and working prompts on day one, and the questions they ask in week one are answered from the company’s own material.
Takes a list of commercial addresses and finds which businesses actually operate at each one, including every tenant in a multi-tenant building, then the owner or manager and a way to reach them. It has worked through more than twenty thousand addresses for its first customer. On a 45-address test batch it went from finding 8 businesses to around 368, after we caught a filter that had been quietly discarding 98 percent of the matches. That failure taught us to test every new list against a known answer before trusting it.
Every scheduled job reports in. When one fails, the founder hears about it the same morning, not the following month.
Reads a deck or a document against the brand rules and flags what drifted before it goes out.
A fixed set of real buyer questions, asked from the team’s own markets, across ChatGPT, Claude, Gemini, Perplexity, and Google’s AI Overviews. It records whether the company is named, which competitors are named instead, and which sources the engines cite.
A monthly report on that panel: share of voice against named competitors, the sites the engines trust, the questions where you are invisible, and what changed since last month. Every rate carries its sample size.
For each question you are missing from, a brief for a page that could answer it. It refuses when there are no real facts behind the page, because a page that could have been written about anyone does not get cited and should not.
Checks that AI crawlers can actually read the website and its structured data, and raises a flag when an engine’s answer about the company changes.
Image and video ads that keep the same brand from one to the next, in English and Spanish, with a compliance check before anything runs and a ledger of what each ad claimed. Specified in full, and waiting its turn.
Most of the design work in this toolkit went into deciding where it stops. Most of these rules are already enforced in the code, and the rest are conditions on bringing on a second team.
Running for one sales team. Enrichment and failure alerts run every night, and the visibility panel runs on a fixed schedule. Follow-up, call prep, outreach, and qualification are built and in testing.
The visibility readout and deal review. Separating each team’s data properly, then bringing on a second team.
CRM write-back, rep onboarding, answer page briefs, site watch, and the content pipeline, roughly in that order, and each one only after a team has asked for it.
Designs and builds the toolkit, and runs the deployments it grew out of. A decade of teaching people to work with complicated systems, and a habit of testing models before trusting them (seven local models benchmarked, one passed).
Checks what the agents produce before it ships, in English and Spanish. Years of quality assurance on search and ad quality for Google projects, and evaluation of AI output against written rubrics.
We are bringing on a small number of sales teams while the toolkit is in development. If your team sells something and spends too much of its week on the work around selling, write to us.
Ask about early access