Good morning.

Your next sale is more likely to come from someone already paying you than from your next batch of leads, and so is your next referral.

The clients who are ready have usually said so already, on a call or in an email, and the one who hears about your next offer is still whoever you happened to speak to last.

In today’s isssue you get the agent that reads every client once a month, ranks who's ready to buy more and who's ready to refer you, and tells you who gets the conversation.

— Sam

Let’s get into it.

Take a coach with a dozen clients on monthly retainer, sitting down on the first Monday of the month.

In the last six weeks, one client asked on a call whether she also works with sales teams, because his has just doubled.

Another emailed, without being asked, to say the program had changed how she runs her week. A third hit the goal she signed up for three weeks early.

None of those became a conversation. The month filled up, the calls ran to their agendas, and the client who heard about the new program was the one she happened to speak to on Thursday.

When I ask consultants and coaches how they decide who hears about the next offer, the honest answer is nearly always whoever they spoke to last.

Referral asks go to nobody at all, because there's never a week when asking a client for an introduction feels like the most important thing on the list.

The first version of this agent takes about forty minutes on the Claude or ChatGPT plan you already pay for: one sheet, ninety days of notes and email, and one prompt. The version that runs itself every month is an afternoon on top of that, in Claude Cowork or Codex.

Somebody already paying you has decided you're worth paying, which is the hardest part of any sale, and a client you've just helped is the most believable person who will ever describe you to a friend.

Both conversations cost less than any lead you can buy, and both depend on timing: the week after a result, or the month a new problem shows up.

The evidence for who's ready already exists. It's in your session notes, in call summaries from Granola, Fathom or Zoom, in the Gmail thread, and in the payment history in Stripe.

Reading twelve clients' last ninety days before every month starts is the job that never gets done, and it's reading, which Claude and ChatGPT do well.

Once a month, the agent:

It also lists every client it couldn't score, with the reason, so a client whose notes sit in the wrong place shows up as a gap you can fix.

It helps to manage it like a hire with one job:

An account manager who reads every client file before the month starts and brings you a short list.

You'd give that person a job description (the system prompt), a definition of what good looks like (the context file), a line they don't cross, and a monthly look at the calls they got wrong.

The line here is that it never contacts a client and never sets a price (unless you want it to). It can tell you which of your offers a client's new problem touches, and the offer, the price and the timing stay with you.

Set one boundary before the build. This agent looks for clients moving toward a yes, so a client who's drifting away shows up here only as a reason to wait, and keeping them is a separate job with its own playbook.

There are two paths, and the choice comes down to where your client records live.

Claude Code in Cowork mode suits you if your notes are files (a Drive folder, Granola exports, a notes app synced to your machine) and you want the scores kept as documents you can open.

Each client gets a file, each month's board is a file, and you can read the history without asking the agent anything. Gmail, Google Drive, HubSpot and Stripe connect through Claude's connectors.

The Codex app suits you if your client record sits in Notion, Slack or a Microsoft 365 workspace, and you'd rather the agent run next to your other automations, with its memory keeping last month's scores.

Codex has no Skill primitive, so the setup is a system prompt plus a scheduled run.

The quick decision rule:

Both produce the same board, and the architecture below doesn't change between them.

The architecture is a health score pointed at growth: signals collapsed into a number, then a decision based on where the number sits and which way it's moving. It has four parts.

The agent scores every client on four signals, each 0, 1 or 2, and every point needs a dated quote or event behind it.

Momentum compares each client with themselves. A client who took three days to reply in the summer and now replies the same morning has moved, whatever your other clients do.

Two kinds of evidence never score: your own activity (you sent three emails) and anything inferred about the client (their company looks like it's growing). If there's no quote or event, the signal scores zero, which keeps the board honest and keeps it short.

The same four signals answer two different questions, so the agent reads them separately.

Result sits in both readings, and nobody qualifies for either with Result at zero. A client who hasn't had anything from you yet won't buy more and won't send a friend, however much they like you. A reading qualifies at four or more, and you can move that line once you've seen a few boards.

Before anyone is ranked, the agent checks for reasons to wait, and any one of them moves the client to Hold however high they score:

The agent names the reason and stops. It doesn't work out why something is open, because that's a different job, and an offer made in the middle of an unresolved problem tells the client you weren't listening.

The cooldown does the most work of the three. Without it, the same three enthusiastic clients top the board every month and hear a pitch from you every month, which is a fast way to make your best clients dread your calls.

Each month the agent compares every reading with last month's and shows the movement next to the number.

A client climbing from two to five is usually a better conversation than one parked at five, because something changed recently and they'll remember what it was.

A client who sits at the top three months running with no conversation logged is telling you something about your calendar.

Then it cuts the list at your capacity. You tell it how many conversations you have room for this month, and for a solo coach two offers and two referral asks is plenty. It gives you that many, in order, and everyone below the line stays on the board for next month, so nobody drops out of sight because the month was short.

It can also report that nobody's ready. Some months the honest board says no offers, one referral ask, deliver for everyone else, and that's a correct output. An agent that finds three upsells every month is promoting pleasant emails to fill the list.

The board accounts for every client on your roster, and each one ends in exactly one lane:

A software founder I work with found $38,400 in that last lane. His business does about $980,000 a year in subscriptions across 740 accounts, and his upgrade agent has to record one of three outcomes on every account, every run: drafted, skipped with a reason code, or not eligible. In its first ten weeks it drafted 88 and skipped 151.

His Monday review of the skipped list takes about fifteen minutes, and it found that 64 of those skips had the same reason:

A usage field that stopped filling in for customers who connected through the API instead of the dashboard.

Once the field was fixed, those 64 accounts went back into the pool and produced 19 conversations and 7 upgrades, $38,400 in new annual revenue from accounts the agent had been unable to see.

That figure assumes the seven wouldn't have upgraded anyway, which is probable but unproven, and three of them were accounts he knew well.

You won't have 740 accounts, and the lesson holds at twelve: the client the agent can't score is usually the one whose notes live somewhere it can't read, and you only find that out when the agent has to say so.

Every step below runs from the prompts in this issue. If you'd rather have it installed, Cortex members get the whole agent as a download:

A Claude Skill that sets it up and runs it, the Codex setup, and four filled-in context files to copy from. Get the Customer Intelligence Agent Build Pack →

Once you've upgraded, come back to this issue on the web and refresh, and you'll see the download link near the bottom.

Before any connectors or schedules, run the whole idea once by hand. It takes about forty minutes in Claude or ChatGPT, and it tells you whether your notes hold enough to score, which is the thing worth knowing before you build anything.

Make the sheet (ten minutes). One row per client: name, what they bought, start date, monthly or total value, and the date you last offered them anything or asked for an introduction.

If you don't know that last date, write unknown, and the prompt treats it as a long time ago.

Gather ninety days (twenty minutes). For each client, pull the last ninety days of session notes or call summaries, plus the email thread.

With the Gmail or Outlook connector turned on, Claude or ChatGPT can pull the threads for every name on the sheet, and Granola, Fathom and Zoom all export call summaries.

Run the prompt (ten minutes). Paste the sheet and the notes in, then:

Read the evidence lines before you read the ranking. If the quotes are right, the ranking usually is too, and if the agent is scoring things nobody said, you'll see it in the first client.

If the board gives you two names worth a call, you have this month's result already. The protocol below turns it into something that runs on its own every month.

The build has seven steps, and every prompt is usable as written.

The agent needs to reach four sources:

Then make one folder for the agent's own files:

The evidence folder is what makes the monthly run fast and checkable. A weekly sweep adds dated entries to each client's file as things happen, so the monthly board reads one short file per client and every score points back to a line you can open.

If you already keep per-client files, start from those. One executive coach I've worked with keeps three for each of her fourteen retainer clients:

The client's goals, every session note, and what the client agreed to do and whether they did it.

She built them for her session follow-ups, and between them they hold every signal this agent scores.

This is the file the agent reads on every run, and it decides the quality of every board. Budget twenty minutes for it.

Spend most of the twenty minutes on Signal Anchors and The Referral I Want. Generic anchors produce a board where every client who sent a pleasant email scores, and a referral ask without a named kind of person gets you a polite yes and no introduction.

You already know what a ready client sounds like on a call. The anchors put that on the page, where the agent applies it to every client, including the ones you haven't spoken to in a month.

Put this in a file called customer-intelligence-system-prompt.md in the same folder, or paste it as the agent's instructions in Codex.

The prompt does two things worth knowing about:

The weekly evidence sweep keeps each client's file current, so the monthly board has something to read:

The monthly board is the run you read:

The conversation log closes the loop after you've talked to them:

The log is what makes the cooldowns work. If the asks never get recorded, the Hold rule has no dates to check, and the same clients reappear at the top every month.

Don't start live. Start on the last twelve months, where you already know who bought more and who sent you someone, and run the board as it would have looked in the month before each of those happened.

The misses tell you which anchor to fix.

The agent drifts in predictable directions, and these four cover most of it:

The agent runs on three rhythms:

In Cowork, the sweep and the board are scheduled tasks that run the Skill against your folder. In Codex, they're scheduled runs with the same prompts. Either way, give the agent write access to its own folder and read-only access to everything else, because nothing about this job needs it to change your email, your CRM or your invoices.

The four signals hold across business types, and the table shows what moves with the business: what counts as "more", where each signal turns up, and which introduction you want.

For a consultant or coach, New Need is the signal that pays, and it almost always arrives on a call as a passing question.

When call summaries reach the evidence folder every week, the board shows it within days, while it's still a conversation and before the client hires someone else to solve it.

An online store or a software product uses the same four signals from order history, reviews and usage data, and the board works at the level of segments with your top customers named.

Solo, this costs you about thirty minutes a month: reading the board, having the conversations, and logging them. Everything else runs on a schedule.

With a team, the split is simple:

You should have an agent like this running for you, so the clients ready to buy more or send you someone hear from you in the month they're ready.

You can build it from the prompts above. If you'd rather skip the setup work, Cortex subscribers get the full Customer Intelligence Agent Build Pack as a download:

This is only available to paying Cortex subscribers.

Cortex is the premium monthly membership version of Bionic Business, helping entrepreneurs and business owners grow and scale with AI and Agents.

In addition to 3-4 full, unrestricted regular issues per month, you also get:

So, if you’d like to receive full issues and the monthly special Signals and Circuits issues, then you should sign up for Cortex, which is $50/month (and you have full archives of previous regular issues for as long as you’re a paying member).

One note on Signals and Circuits: You only get these special issues for the months you’re subscribed—which means if you skip a ~30 day period, about a month, you won’t get them for that month. There are no back-issues for past months, so now is your chance to lock-in these upcoming issues while you can.

After you’ve upgraded to Cortex, log in to the website and come back to this issue, and the Build Pack will be unlocked below.

Ready? Get the Build Pack, full regular issues, and the upcoming Signals and Circuits issues:

Cortex is a monthly premium version of Bionic Business, helping entrepreneurs grow and scale with AI and Agents. You can cancel at any time.

One tap tells me how most of you handle this today. If you want to say more, the box after your vote comes straight to me.

The coach at the top of this issue had three clients ready for a conversation and talked to none of them about it.

Each of them had said something on a call or in an email, and the month filled up before she read any of it back.

Run the forty-minute version before the first of the month.

Have two conversations from it, ask one client for one introduction, and put both dates on the sheet, so next month's run knows which clients to leave alone for a while.

Talk soon,
Sam Woods
The Editor

.