01
Your customers' words. Never the marketer's.
Every headline, every offer, every claim traces back to something a real customer said. The marketer's opinion is not an input.
Marchitect OS · The Map
Every marketing guess you ever paid for broke one of these rules.
Marketing has first principles. They hold no matter what you sell, who you sell it to, or what you charge. Strip away every tactic, every channel and every opinion, and this is what is left.
The whole of marketing, in one loop
Four layers. One cycle. It never stops turning.
Every piece of marketing a company does falls into one of these four layers. Most companies do them out of order, skip two, and never connect the last one back to the first. The OS does all four, in order, and closes the loop.
Layer 1
Capture the data
Sales calls, surveys, reviews, feedback. Everything your customers say about their problems and what they want. Raw, unfiltered, in their words.
Layer 2
Turn it into intelligence
AI agents read every word, rank what matters, build your buyer profiles, learn your company and your market, and construct the offer and the reason to choose you.
Layer 3
Execute with playbooks
Every channel pulls from the same intelligence. Each has a playbook that says exactly what to do, and a copywriter that writes in your customers' language.
Layer 4
Measure and optimize
Baselines for every channel. Results compared to the baseline. Improvements fed back into the playbooks. Yesterday's best becomes tomorrow's floor.
The return
What layer 4 learns goes back in as layer 1 data.
Every result comes back as a fact about your customers. The message that won, the channel that paid, the offer that closed. All of it lands on the pile as new data.
The next cycle starts from what the last one proved, not from what someone remembered. That return is the difference between an engine and a checklist.
The longer you run it and the more you feed it, the better it gets.
The first principle
Right person. Right message. Right time and place.
Show the right message to the right person at the right time and place, and marketing works. That is the whole equation.
Tactics change. Channels change. Tools change. This does not.
Everything the engine takes in and everything it makes serves these three.
01 · The equation
MME = RM + RP + RTP
But there is a catch.
RM
Right message
The problems they carry and the outcomes they want, in their own words, not the marketer's.
RP
Right person
The buyer the offer is built to serve. Not a demographic guess. The people who paid, described in their own words.
RTP
Right time and place
The channels they actually use, at the moments they are ready to listen.
02 · The catch
RM and RTP are 100% dependent on RP.
RM = f(RP)
RTP = f(RP)
The message is a function of the person. So is the time and place.
Hold the product still and change the person. Both other variables change with it.
Worked example. One software product. Two people at the same company who both have to say yes.
RP The person
RM The message that lands
RTP Where they listen
The CEO
Worried about the whole company.
Revenue, risk, headcount, and what this does to the value of the business. Nothing about the stack.
Peer groups, founder newsletters, the board deck, a trusted CFO's word.
The CTO
Worried about the tech.
Will it integrate, will it stay up, what it does to the stack and the engineers who run it. Nothing about valuation.
Engineering blogs, GitHub, developer conferences, a trusted engineer's word.
MMECEO = RMCEO + RPCEO + RTPCEO
MMECTO = RMCTO + RPCTO + RTPCTO
Same product. Two people. Two messages, two sets of places, two campaigns. Not one variable is shared.
That is what 100% dependent means. It holds for every group you sell to: husband and wife, younger and older buyers, the owner and the operator.
Get the person right first. Until you do, every dollar you spend on the message and the channels is spent on a guess.
03 · What you build once it is solved
MME = RM + RP + RTP
Solve for the person first. The message and the timing follow.
Offer
Built from what those people said they want. It is what the right message delivers.
USP
Wraps the offer. It is what sets this one apart from every other offer in front of that person.
Five variables, one order. Marketing stops being opinion and becomes an equation you can solve.
The six steps below solve it in that order.
01
Layer 1 · Capture
Listen to your customers.
The only fuel the engine accepts is what real customers said. Not what a consultant thinks they said. Three sources today.
What you do
- Paste in sales call transcripts.
- Run a deep-dive survey. The OS writes the survey for you, tuned to what it already knows about your buyers.
- Drop in reviews, support tickets, cancellation notes, any feedback that arrives on its own.
Behind the curtain
- Who is talking. On a sales call the engine works out which voice is the customer and which is your rep, then keeps only the customer's words. Your rep's pitch never gets mistaken for the customer's opinion.
- Review before it counts. What was extracted from a call is shown to you before it is saved. You keep or drop.
- No double counting. The same transcript pasted twice is caught and refused.
- Every response is graded for depth before anything is pulled from it, so a one-word answer never carries the weight of a paragraph.
What you get
A growing, permanent record of what your customers said, in their words, with every line traceable back to the call or survey it came from. This is the raw material for everything below.
Next: calls flow in straight from the phone system, no pasting.
02
Layer 2 · Intelligence
Understand what they meant.
A pipeline of AI agents reads every word and turns it into ranked, validated intelligence. This is the brain. Everything downstream depends on it.
What you do
- Press one button: analyze.
- Read the ranked list of problems, desires, and channels. Click any line to see the exact customer quotes behind it.
- Ask the research chat any question about your customers and get an answer grounded in the data, not the internet.
Behind the curtain
- A router sends each input to the right path: survey, sales call, or other.
- Four extractors run in parallel on every response: one finds the problems, one the desires, one the channels, one mines quotes worth using in copy.
- Two extractors listen for the sales side: objections, and the competitors customers name. Marketing intelligence and sales intelligence are kept apart on purpose. A price complaint is not a marketing problem.
- A pattern analyzer takes hundreds of raw labels and consolidates them into a ranked list of themes, with frequency and confidence on each.
- An avatar builder groups your buyers into profiles from the language they share. The AI groups; the code counts, so segment sizes are arithmetic, not opinion.
- A grader sorts every quote by how it can be used: a pain hook, a testimonial, a result claim, a voice sample.
What you get
- The Master List. Your customers' problems, desires, and channels, ranked, scored, and traceable to the source.
- Objections and competitors, ranked the same way.
- The Quote Bank. Every usable line your customers ever said, searchable, ready to paste.
- Avatars. Buyer profiles in their own words, with real quotes attached.
03
Layer 2 · Intelligence
Know your company and your market.
Customer language is half the picture. The engine also has to know what is true about your business and who else is competing for the same buyers. One website address starts it.
What you do
- Type in your website. The OS reads it and proposes facts. You confirm the ones you stand behind.
- Press "Research the market." A research agent searches the live web and answers eight questions a marketer needs answered before writing a word.
- Keep or drop the competitors it found. The ones you keep get profiled on their own.
Behind the curtain
- Nine kinds of company fact are pulled from your site: what you sell, prices, proof, testimonials, guarantees, how it works, credentials, capacity limits, your story. Each carries the page it came from.
- The eight market questions: who buys this, where you sell, what buyers call it, who competes, what buyers do instead, what it costs in this market, how you reach customers, and what the world says about you.
- Every finding wears a label: found on your site, found on the web, or inferred. Only pages the search actually returned count as sources. Anything else is marked as a guess. Enforced by code, not by trust.
- Competitor profiling reads each rival's site and records what they claim about themselves, so your claims can be judged against theirs.
- Hard limits. Eight searches. Five minutes. One button. Nothing runs on its own.
What you get
- A confirmed fact base that no agent in the system is allowed to contradict or exceed.
- A market brief with links behind every line.
- A competitor picture: who they are, what they say, what they charge, what customers say about them.
Next: a standing attendant that watches your market and flags what moved.
04
Layer 2 · Intelligence
Build the offer and the reason to choose you.
An offer is what you sell and how you frame it. A USP is why someone should pick you over doing nothing, doing it themselves, or hiring the other guy. The engine builds both from the intelligence above, and a copywriter writes the words.
What you do
- Pick an avatar and a product. The Offer Builder writes a complete offer: the core promise, the value stack, pricing, bonuses, guarantee, urgency, and a name. Edit any line; the copy rewrites around your edit.
- Have an idea for something you do not sell yet? Talk it through in an idea session. The OS interrogates it against your customer data and builds a concept offer you can promote to real once you commit.
- Build a USP. Four guided steps: claims, scores, proof, one sentence.
Behind the curtain
- The offer is built in three waves, each grounded in your top objections, and a coverage map shows which objections the offer answers and which it does not.
- Every number cites a confirmed fact. A price, a count, a date with no fact behind it is rejected by the validator before you ever see it. An agent may propose a new commitment, but it is labeled as an idea, not a fact, until you confirm it.
- The USP writes claims from what customers said, for each top desire against each alternative, doing nothing first. Appeal is computed from how often customers raised the problem. Exclusivity is judged against what competitors say on their own sites. Both can be overridden. Neither has to be.
- Claims without proof are kept, not hidden, and listed so you go find the facts. That is the habit: claim, then proof, then the copywriter cites the proof.
- A copywriter agent writes every line, trained on the work of eleven master copywriters and held to a strict standard: plain words, short sentences, the customer's language, no jargon.
What you get
- Offers with receipts: every section shows the quotes, themes, and facts it stands on.
- A ranked USP in three tiers: lead with these, support with these, mention if relevant. Plus a one-sentence statement your business can say out loud.
- A walk, not a form. Every multi-step page tells you where you are, what is next, and what you lose if you skip. The same walk will drive a chat interface later.
05
Layer 3 · Execution
Run the channels from one source of truth.
Every channel pulls from the same intelligence. Each has a playbook that says exactly what to do. The person executing does not need a marketing degree. They need the playbook.
What you do
- Open a playbook. Follow it.
- Today: the deep-dive survey playbook, which writes and specs a customer survey from your intelligence.
- Today: the copywriter, which writes inside every surface that produces words.
Behind the curtain
- Playbooks are agents, not PDFs. They read the live intelligence when they run, so a playbook written today is still current after next month's calls come in.
- Unlimited channels, each with its own playbook: ads, content, email, social, sales enablement. Every one draws from the same ranked problems, desires, quotes, offers, and USP.
- Sales training is a consumer too: the objections and competitors from your calls feed the people who take the next call.
What you get
Campaigns, content, ads, and sequences that all say the same thing because they all came from the same place. And a marketing operation that runs without you in the room.
Next: channel playbooks for advertising, content, and email, in that order.
06
Layer 4 · Optimize
Measure it. Improve it. Go around again.
Without a baseline, optimization has nothing to beat. The last layer sets the baseline for every channel, compares results to it, and feeds the improvements back into the playbooks.
What you do
- Connect a data source. Your CRM, your analytics, your ad accounts.
- Read one number per channel: how it is doing against its own baseline.
Behind the curtain
- Integration-agnostic. One adapter per source. The first is HubSpot; analytics, tag manager, and payments follow.
- One optimization loop per channel. Each compares results to baseline, suggests copy, creative, or targeting changes, and pushes them back into that channel's playbook.
- Every token the engine spends is metered and shown, so the cost of the intelligence is a known number, not a surprise.
What you get
A system that gets smarter every cycle. What layer 4 learns becomes new data in layer 1. The loop closes, and it keeps closing.
What makes it different
Six things no agency, and no other software, will give you.
02
Nothing is invented.
No agent in the system may make up a number, a price, a count, a date, or a quote. A claim with no source behind it is rejected by the code before a human sees it. This is a rule of the machine, not a promise from a person.
03
It walks you by the hand.
Every multi-step job tells you where you are, what to do next, what it needs first, and what you give up by skipping. You do not need to know how marketing works. The system does.
04
It closes the loop.
Capture, intelligence, execution, optimization, and back to capture. Most marketing is a straight line that ends when the campaign does. This is a wheel.
05
Your engine. Your data. Your install.
One install per company, on your own account, with your own keys. The AI engine is your choice. The knowledge the agents rely on ships inside the software. Nothing lives on someone else's server.
06
Built by a practitioner.
Not a software company guessing what marketers need. A marketer with thirty-five years of client work, encoding what actually works into agents that do it the same way every time.
What went into it
The work you never see.
The screen shows a button. Behind the button is this.
20+
AI agents
Each with one job, a written standard, and a gate that rejects its output when it breaks the rules.
780
Automated tests
Run on every change. Several exist only to stop an agent from ever fabricating.
51
Build days
Every day logged: what was built, what broke, what was decided and why.
11
Master copywriters
Their methods distilled into the copywriter agent. It writes the way they wrote.
67
Offer principles
Named, cited, and checked. The Offer Builder must say which principles it applied.
9 + 8
Facts and questions
Nine kinds of company fact. Eight market questions. The minimum a marketer needs before writing.
35
Years of marketing
Since 1987. Thirty-plus businesses founded. Close to a thousand clients. $2M of marketing education, encoded.
1
Equation
Right person, right message, right time and place. Everything else is a tactic.
The rules the machine obeys
Written down. Enforced in code. Not negotiable.
No fabrication.
No number, price, count, date, or claim without a verified source. Customer data for language. Confirmed company facts for facts. Rejected otherwise.
No over-reach.
A benefit is not inflated into a pain. No invented emotions or motives. A thin result is the correct result when the data is thin.
The customer's words, only.
Quotes are verbatim. A quote the system cannot find in the source is dropped, not paraphrased.
A human confirms.
Facts, competitors, extracted calls, promoted ideas: proposed by the machine, confirmed by a person, only then used.
Label what you infer.
On the way in, an agent may reason beyond its sources, as long as every finding says whether it was found on the site, found on the web, or inferred.
Your engine, your keys.
The AI engine is the client's choice. Every key is the client's own. The knowledge ships inside the software.
Marketing and sales stay separate.
A call carries both. The problem the product solves is marketing. The friction in buying it is sales. They feed different things and are never conflated.
The walk lives on the server.
Where you are and what comes next is worked out from the data, not remembered by a screen. A chat interface reads the same answer.
The outcome
Not a campaign. Not a launch. An engine built on what endures.
Tactics fade. Channels shift. Tools go obsolete. What your customers want, in their own words, and the discipline of building everything from that, does not. Every cycle sharpens the person, the message, and the timing, for as long as there are customers to serve.
And when the day comes that someone puts a number on your company, marketing is no longer a black hole on the expense line. It is a documented, transferable machine that runs without you. That is worth turns on the multiple.
Appendix: the agent roster
For the technical reader. Click to open.
Every agent in the software, where it lives, and what it does. Paths are relative to the repository root.
A1
Classifier
src/agents/agent-1-classifier.ts
Routes each input to the survey, sales call, or misc path.
A2
Engagement Scorer
src/agents/agent-2-engagement-scorer.ts
Grades survey responses by depth of signal before extraction.
A3
Problem Extractor
src/agents/agent-3-problem-extractor.ts
Pulls pains and frustrations as labeled elements with confidence and intensity.
A4
Desire Extractor
src/agents/agent-4-desire-extractor.ts
Pulls wants and outcomes, what the buyer is moving toward.
A5
Channel Extractor
src/agents/agent-5-channel-extractor.ts
Finds where the buyer hangs out and how they found you.
A6
Quote Miner
src/agents/agent-6-quote-miner.ts
Pulls verbatim, copy-ready language into the Quote Bank.
A7
Headline Grader
src/agents/agent-7-headline-grader.ts
Grades every quote by use: pain hook, testimonial, result claim, voice.
A8
Pattern Analyzer
src/agents/agent-8-pattern-analyzer.ts
Consolidates raw labels into ranked themes with a verbatim gate on every source label.
A8s
Sales Pattern Analyzer
src/agents/sales-pattern-analyzer.ts
The same consolidation for objections and competitors.
A9
Avatar Builder
src/agents/agent-9-avatar-builder.ts
Groups buyers by shared language. AI groups, code counts, real quotes attached, headlines grounded.
A10
Research Chat
api/chat.ts · src/services/research/
Grounded answers over the live intelligence, with pluggable sources.
CA
Call Speaker Attribution
src/agents/call-speaker-attribution.ts
Works out who is the prospect and who is the rep, windowed for long calls, human-confirmed.
CS
Call Segmentation
src/agents/call-segmentation.ts
Breaks the prospect's speech into units ready for extraction.
OB
Objection Extractor
src/agents/objection-extractor.ts
Sales-side friction: price, terms, logistics, approvals.
CX
Competitor Extractor
src/agents/competitor-extractor.ts
The rivals customers name on calls.
CF
Company-Fact Extractor
src/agents/company-fact-extractor.ts · api/scrape-company.ts
Reads the client's site into nine kinds of fact, each with its page, for human confirmation.
CC
Competitor-Claim Extractor
src/agents/competitor-claim-extractor.ts · api/profile-competitor.ts
Reads a competitor's site and records what they claim about themselves.
MR
Market Research Agent
src/services/companyResearch.ts · api/research-company.ts
Searches the live web to answer the eight market questions. Every finding labeled site, web, or inferred. Eight searches, hard cap.
OF
Offer Builder
src/services/offerBuilder.ts · api/generate-offer.ts
Three waves, seven sections, coverage map against top objections, fact citations enforced, highlight-to-rewrite.
ID
Idea Session and Brief
src/playbooks/offerIdea/ · api/offer-idea-brief.ts · api/promote-offer.ts
Interrogates a product idea against the data, extracts a typed brief, builds a concept offer, promotes it commitment by commitment.
US
USP Builder
src/services/uspBuilder.ts · api/build-usp.ts
Claims from customer data, appeal computed in code, exclusivity judged against competitor claims, facts linked, statement written.
DS
Deep-Dive Survey Builder
src/playbooks/deepDiveSurvey/
Guided session that writes and specs a customer survey grounded in the intelligence.
PAT
Pat the Copywriter
src/copywriting/
Composed into every surface that writes customer-facing words. Voice, principles, and one playbook per job. Ships inside the software.
GD
Guides
src/guides/ · api/guide.ts
Not an AI agent. The walk itself: readiness, steps, what is next, evaluated from the data.