Build Smart with AI — The Founder's Playbook. Session 2: Organisational memory.
Your company knows more than any one person in it. Your AI should too.
Most of what makes your business work — how you price, why you said no to that deal, what the last three lost customers had in common — lives in a few heads and a thousand chat threads. This session shows how to turn that into a living memory your team and your AI can both use, what it changes about the AI you're already paying for, and where a human must still make the call.
Retainer pricing
Current rule — replaces per-seat
One current record, versus the pile it replaces.
- When
- Wednesday, 7 October 20262:00 PM BST · 3:00 PM CEST · 9:00 AM Eastern
- Format
- Zoom webinar · Live, 45 min + Q&APrivate Q&A — nobody sees who else is attending
- Who's talking
- Sanjay Menon & Anoop E.CEO asks the founder's questions; CTO answers and demos live
- Built for
- Founders of 20–200 person companiesBusiness fluency assumed. No technical background needed
Who this is for
You've grown past the stage where you can hold the whole company in your head — and it's starting to show
This edition is written for one person: the founder or MD of a company that has outgrown "ask me". You have a team, a live product or service, real customers — and a growing sense that the business still routes through you more than it should. If a colleague could take a week off and nothing would need to be re-explained, you can skip this one.
We'll use sales and marketing as the running example throughout, because that's where the knowledge gap costs the most and shows fastest — pricing decisions, positioning, why deals are won and lost, what was promised to whom. Everything shown applies equally to operations, delivery, or finance.
Your company's most valuable asset — how it actually thinks and decides — lives in people's heads and chat threads. Your AI can't use it, your team can't find it, and your valuation is discounted for it.
You'll recognise yourself if…
- New hires take months to become useful because the "how we do things here" isn't written anywhere.
- You've answered the same question — about pricing, a client, a policy — three times this month.
- Your team tried AI, got generic answers, and quietly went back to asking you.
- An investor, buyer or bank has asked what happens to the business if you step back — and you didn't love your answer.
What you'll leave with
Three answers you can act on the same week
What can I do myself?
Switch on "search my company's tools" today — and know what it will and won't tell you.
The AI tools you already pay for can now search your Slack, email and documents and answer questions. We'll show you exactly how far that takes you, honestly, and how to start writing down what your company knows in a way that compounds.
Where will it mislead me?
Why "correct" answers from the wrong version of the truth are the real danger.
Three pricing sheets, one reversed decision in a chat thread, and a confident AI summary. We'll show the failure live, and the simple discipline that prevents it.
Where do I still need experts?
Deciding what's true, and what AI is allowed to act on, is a human job — here's where it sits.
You'll see precisely which steps need a person's sign-off, why that is what makes the whole thing trustworthy, and how to keep that from becoming a bottleneck.
The obvious question
"My AI can already search everything we have. Isn't that the same thing?"
No — and we'll say so before you have to ask. Claude's Ask your org, and the equivalents from every other vendor, are good. If you're on a business plan, switch it on this week; it takes fifteen minutes. But searching what your company has written is not the same as knowing what your company has decided. The two sit at different layers, and the session is about the second one.
Searching your tools — what the AI vendors ship
Tells you what was said
Every time you ask, the AI runs off to your chat, email and documents, reads whatever it finds, and works out an answer on the spot. Nothing is remembered between questions. The next person to ask gets the same detective work, done again — and possibly a different conclusion.
- Brilliant at "where is the contract?" and "what did the client say on Tuesday?"
- Unreliable at "what is our position and why?" — it has to guess which of several versions is current.
- Can only know what someone wrote down; the reasoning in your head never surfaces.
- A tool for people to ask questions — not something your AI assistants can build on.
Organisational memory — what this session is about
Tells you what is true, and why
Information from your tools is gathered continuously, checked, and — where it matters — reviewed by a person who decides which version wins and records the reason. The result is a small, living set of plain documents: your company's memory. People read it. Your AI reads it. Both get the same answer.
- Conflicts settled once, by a human, with the reasoning kept — not re-guessed every time.
- Captures the "why" behind decisions, which no search can find because it was never written.
- Same question, same answer, until someone deliberately changes it.
- A foundation your AI assistants can stand on, act from, and add to.
"What's our current retainer pricing for a 40-person fintech — and why did we move off per-seat?"
Search-your-tools answer
Finds a 2025 pricing deck, a newer one in someone's Drive, and a leadership thread that changed the rule. Produces a fluent summary that blends all three. Confident, cited, and wrong on the number that matters.
Memory answer
Reads one current page: the rule, the date it was approved, who approved it, and the two-line reason the team moved off per-seat. Then points to the live deal system for the numbers that change daily.
Where search wins, and we'll say it out loud: breadth and speed to start. It covers everything you've ever written, it respects who is allowed to see what automatically, and it's live in an afternoon. Memory covers only what's been curated, and takes weeks to build properly. Use both. Just don't mistake one for the other.
Smarter and cheaper
Memory doesn't just make your AI more accurate. It makes it read less — and reading is what you pay for
Here is the part almost nobody explains to business owners. Every time an AI answers a question, it has to read something first — and you are billed, and you wait, for every page it reads. Search-based AI reads a fresh pile on every question. Memory hands it a one-page briefing instead. Same question; a fraction of the reading; a better answer.
Searching your tools
Every single time the question is asked
The whole pile: dozens of threads and files — including out-of-date versions the AI must weigh up itself. More of it degrades the answer.
Organisational memory
After a person has already decided
One or two pages: the current rule, its reasoning, and a pointer to live numbers. Curation cost paid once, then reused by every person and every AI task.
Illustrative, not a benchmark. We are measuring this on our own sales and marketing memory right now and will show the real figures on the day.
The bigger picture
Why this is where the whole industry is heading
The first wave of business AI worked by "search, then guess": fetch whatever looks similar, hope it's right. The consensus now — from the labs building these models to the firms deploying them — is that the quality of an answer depends on the quality of what the AI is given to read, not on how much. The discipline has a name in the trade: context engineering. Organisational memory is that discipline applied to your company.
Why more isn't better
Independent research across eighteen leading AI models found the same thing: the more you give them to read at once, and the more near-duplicates and stale versions are in the pile, the worse their answers get. Curating what goes in isn't tidiness. It is the accuracy mechanism.
Why it makes AI assistants possible at all
An AI that drafts your proposals, briefs your sales team before a call, or watches for regulation that affects you is only as good as what it knows about you. Without memory it produces generic work and gets switched off. With memory, the same assistant becomes specific, consistent, and affordable enough to run every day.
Plain-English glossary for the day
- Organisational memory
- The written, current, reviewed record of what your company knows and has decided.
- Context
- Whatever the AI is given to read before it answers.
- Context engineering
- Choosing that reading carefully instead of dumping everything in.
- AI assistant (agent)
- An AI that carries out a task, not just answers a question.
- Human-in-the-loop
- The points where a person, not the AI, makes the decision.
Live, not slides
Four things you'll watch happen on our own company's memory
We're running this on Emvigo's own sales and marketing first — real pipeline, real decisions, real mistakes — so what you see is a working system, not a mock-up. Client material never appears.
Two versions of the truth collide — and a person settles it
A pricing rule changed in a chat thread contradicts the deck everyone still uses. Watch the pipeline catch it, a human decide, and the reason get written down so it never has to be re-argued.
An AI assistant does real sales work from memory
A first-draft proposal for a new prospect, built from our current positioning, pricing rules and the reasons we lost the last three similar deals — then the human edit that makes it sendable.
The assistant learns something — and asks permission to remember it
A new objection from a prospect call becomes a proposed addition to memory. It waits at the gate until someone approves it. Nothing gets in on its own.
"What did the company learn this week?"
One screen showing every change to the company's knowledge in the last seven days, who approved each, and what it replaced. The founder's Monday-morning view.
Where humans stay in charge
The AI drafts and fetches. People decide. Here's the line, drawn in advance
This is not a caveat we add at the end; it's the design. The reason memory can be trusted is that every consequential entry passed a person. We'll show you where those points sit, who owns them, and how to keep them from turning into a queue nobody clears.
- Which version of a fact is current, when sources disagree.
- Anything about pricing, contracts, legal or financial commitments.
- Anything a customer will see.
- What the AI is allowed to add to memory on its own (our answer: nothing).
- Who — and which AI tasks — may read the sensitive parts.
You'll also hear where we think it doesn't pay off yet, and when you should simply use the search feature you already have.
Where Emvigo comes in
Three minutes, at the end, and only once
The software behind this is largely open source and runs inside your own environment — your knowledge never leaves your walls. We'll tell you exactly which pieces, so you can try it yourself. What can't be downloaded is the work of sitting with you to decide what's true, capture why, and keep it alive as the company changes. That's the part we do, alongside your team, the way we already co-build software with customers.
Whether you do that with us or on your own, you'll leave knowing how it works and where it gets hard.
Missed Session 1?
Watch the previous session in full
Session 1 — Watch AI Build on a Real Product, and Get It Wrong, Live — is on YouTube in full: the same format, on a real product, mistakes included. Every session is self-contained, so you've missed nothing by starting here; but if you want a feel for how these run before you register, this is it.
Watch AI Build on a Real Product — and Get It Wrong, Live · Build Smart with AI, Session 1 (full recording) · Open on YouTube
Register
Register, and bring the one question you're tired of answering
We'll ask for it when you register — it shapes what we show on the day. Registration is free and needs a business email. Your questions during the session are private to you and the presenters.