AI Automation Specialist · application

Hi Daniel.
I read the post, and…

I thought, why not make a whole production of my thought process and approach?

See how I would start ↓See the builds I think matter
So this is me showing you how I would think about the job, not just telling you I can do it.
PWC · the standard I am thinking about built for real work

What I would be trying to get right

Not a giant automation. That is usually a careless way to build. I would want a system people can actually trust while the work is happening.

01
It knows enough before it acts.The right person, the useful context, the company rules, and a clear idea of what is still unknown.
02
Humans are involved where they actually add something.Not approving every tiny thing because we never taught the system enough to handle it.
03
If one tool is down, one tool is down.The rest of the company should not suddenly freeze because one connection is having a bad day.
04
The system can keep learning how the company really works.Useful corrections, exceptions and decisions should not disappear after one task.
01 · before building anything

I would need to understand how the work actually happens inside the company first.

SOPs are a great place to start. But I would not build from SOPs alone, and I would not build from only top management's version of how the day-to-day works either. The real workflow usually shows up once you sit with the people doing it.

I want to understand how everyone actually gets the job done, what each person wishes magically existed for them at the right moment, and what information they keep going hunting for.

Why I don't stop at the SOP

People have shortcuts. They check things that were never written down. Sometimes they use another tool because the official way is slower. I am not in the business of building something technically correct that the team quietly works around until it rots.

Once I understand those little realities, then we can sit together and say: this part could be automated, this is what it gives person A, this is what it gives person B, this should just become easier to see, and this should still stay with a person. I have built systems where the same underlying work has a different view depending on who is looking. That is usually a much better starting point than making everybody live inside one giant dashboard.

Click through the people below.

02 · one small example

If a new lead came in through WhatsApp, we could build this kind of workflow.

I obviously do not know your internal lead process yet, so this is not me pretending I have already designed the final Property Wealth workflow. It is just a sneak peek into what I would be thinking through: what does the message tell us, what do we already know, what are we still missing, and what should happen next?

Property Wealth
WhatsApp enquiry
New enquiry · just now
Hi, I saw Daniel's post. I'm looking at property investment but I'm not sure which route makes sense for me. I'm in Manchester and probably looking at around £40k to start.
Thanks — that helps. Let me get the right information and person for you.
The reply is not the interesting bit yet. The context is.
WHAT WE KNOW SO FAR

New WhatsApp enquiry

from this message
LocationManchestermessage
Interested inProperty investmentmessage
Available capitalAbout £40kmessage
IntentLooks seriousinference · reviewable
Still unclearTimescalemissing
Still unclearBest route / offerneeds context
Look for an existing person before creating another record. Check likely duplicates, other contact routes and any open opportunities already attached to them.
Bring in only the history that matters here: previous messages, notes, source, opportunity history, appointments and any company rule that changes how this enquiry should be handled.
Turn the conversation into useful CRM information rather than leaving the useful part buried inside 47 messages. What do they want? How serious are they? What is missing? Which facts came from them and which are our interpretation?
A fixed rule can handle the obvious things. AI can help where interpretation is useful. A person stays involved where the company wants judgment or the consequences are higher.
Update the useful fields, route the opportunity, assign the right owner, trigger the right follow-up or hold the action for review. GHL can remain the system of record and execution layer rather than asking it to be the whole brain.
I do not want the AI guessing too early. I want it gathering the right context first, and I want us to be able to see where each important conclusion came from.
03 · as the relationship develops

As that person keeps interacting with the company, the profile should get more useful.

Every new thing that happens should make the record a little more useful. A message, a call, a booking, somebody correcting the system — all of that can change what we know and what should happen next.

04 · once this is happening across the pipeline

The morning view can get a lot simpler.

I would not want leadership opening the CRM in the morning and getting bombarded with the same information everybody else sees. The person managing the work and the person doing the work are looking for different things.

Leadership usually has the least time to click around, but the whole system still has to report to them. Once I understand what Daniel actually needs to see, the first screen gets much clearer: what needs him, what is stuck, where money is sitting still, and what can wait.

PROPERTY WEALTH · TODAY

Good morning, Daniel.

updated from current activity
NEEDS YOU TODAY3

high-value enquiries need a person

MONEY TO LOOK AT£143k

across 12 opportunities that may be sitting still

WORTH NOTICINGReferrals

are producing fewer leads, but much better revenue this month

Amina · qualified, no meetingAsked about implementation yesterday. No appointment booked.
James · proposal viewed 3 timesNo next conversation is scheduled.
Sarah · moving normallyMeeting booked for tomorrow. Nothing needed from you.
Ask about the pipeline…
Try one of the questions above.
05 · the marketing side

The marketing side gets more useful when we can follow a lead all the way through.

If we only look at lead volume, we can very easily optimise the wrong thing. Once the CRM, conversations and revenue are connected, we can see which channels are bringing in people who actually become customers.

Instagram
300leads
34 qualified8 opportunities£12k revenue
vs.
LinkedIn
80leads
29 qualified17 opportunities£85k revenue
Fewer leads. Much better commercial outcome.
06 · things that should not quietly disappear

I would also want the system looking for things that should have happened but didn't.

A lot of revenue loss is boring. Nobody followed up. A proposal sat too long. A good lead quietly disappeared into nurture. I would want the system noticing those things before somebody remembers to check.

NEEDS ATTENTION5 things
Qualified 4 days agoNo meeting booked£18k est.
Proposal sent 11 days agoNo follow-up£24k est.
High-intent leadStill sitting in nurture£9k est.
Appointment missed yesterdayNo recovery message£8k est.
Open opportunity · 18 days quietNo meaningful activity£25k est.
These look like five separate admin problems until you add them up.
07 · when one piece is unavailable

If one thing is down, I want one thing to be down — not the whole company.

This is one of the reasons I like building on top of the stack a company already trusts. I want the custom layer to make everything feel joined up without becoming the only place the work can happen.

Why I care about building on top of what already works

If GHL is down, GHL is down. Email should still be email. Files should still be where the team expects them. If AI is down, the fixed rules should not suddenly forget how to work. I want the failure to stay as local as possible, and I want the system to know what can keep going, what should wait, and when a person needs to know.

Click a part of the stack and turn it off.The answer should not always be “everything stops.”
And then there are the less dramatic failures that still matter.
08 · human involvement

Not every action needs the same amount of human involvement.

I am very pro human-in-the-loop. I just do not think that means a person has to approve every tiny thing. Some actions can run on their own. Some should come to a person almost finished. Some decisions should stay with people completely. We decide that action by action: what could go wrong, can we undo it, are the rules clear, and does this actually need judgment?

MOSTLY SYSTEM

The system can do this.

ExampleUpdate a harmless CRM field after a confirmed event.
ExampleAttach an incoming message to the right existing contact.
SYSTEM + PERSON

The system can get this ready.

ExampleDraft a normal follow-up with the right context attached.
ExampleRecommend where an unusual lead belongs and show why.
MOSTLY PERSON

A person should make this call.

ExampleHandle an unusual objection with real commercial consequences.
ExampleMake a financial promise or exception outside an approved rule.
What could go wrong?Can we undo it?Are the rules clear?Does this need judgment?
09 · what the company learns

I also want the system to remember how the company actually works.

You can define a lot during setup. You are still not going to capture every useful rule before people start working. Some of the real method only shows up when somebody corrects the AI, makes an exception, or says: no, we do it this way because...

I built BlackBox because I kept paying the same context-reconstruction cost every time I moved between sessions and AI tools. At company scale that cost gets silly. If the team has already learned something useful, the next person — or the next AI — should not have to reconstruct it from scratch.

What I mean by company memory

You know the kind of memory AI assistants already have? Great. Useful. But for a company I want something much more inspectable. I want to see exactly what the system thinks it knows, where that came from, why a correction became a rule, who approved it, and whether the rule still applies. Not a broad memory like “Millicent likes humour” when the useful thing was the exact reason we changed sentence three.

Open BlackBox ↗
BlackBox governed memory
The system suggested the standard first-touch follow-up.
Sarah changed it because this person had already spoken to an adviser.
Returning investor leads should not be treated like first-time enquiries once adviser contact is confirmed.
Source: CRM activity + call note · Owner: Sarah · Updated today
THIS MAY BE A REUSABLE RULEWhen adviser contact is confirmed, check the returning-lead path before sending first-touch nurture.
CurrentCan expire.
SourcedShows where it came from.
PermissionedRight people only.
ReversibleCan be corrected.
Where I would take this for a team

For a team, I would push this much further.

I would want a shared company brain pulling from the CRM, team comms, documents, meetings, tasks and approved rules, so you can ask normal questions about what is stuck, what changed, what looks risky and what the team keeps correcting. The point is not one giant brain owning everything. It is the company not losing its own context between tools.

CRMteam commsdocumentsmeetingsapproved rules
What is actually stuck this week?
3 things need attention. One qualified lead has no owner. Two proposals have had no activity for 10+ days.
Why do we handle returning investors differently?
Because the team approved that rule after repeat cases. I can show the source and when it was last reviewed.
10 · things I have already built

These are the parts of my previous work I think are most relevant here.

They are different builds, but they keep coming back to the same thing: understanding what good work actually needs, then getting the system to carry more of it without making the humans rescue it all day.

01 / OUR PLACE CAMPAIGNS

I was tired of systems that gave people another place to record the work instead of doing more of the coordination around it.

Each campaign gets a room. Communication can route into it through its own email address. Links, files, approvals, creator updates, team context and current risks can come together without the team rebuilding the picture by hand.

Open the live build ↗
Our Place campaign roomOur Place memory and team contextOur Place context-first draftOur Place resilience and mirroring

Campaign room. The work has somewhere to belong.

What people know. Useful context arrives in the room instead of living in somebody’s head.

First drafts. The AI gathers the relevant context before it writes.

Resilience. The room helps coordinate the work without becoming the only place the work can happen.

Claude
ChatGPT
Perplexity
BLACKBOXone governed memory
02 / BLACKBOX

I built this because I was tired of reteaching one AI what another AI already knew.

For a team, that becomes shared memory you can actually inspect: where this came from, who changed it, who approved it, who can see it, and whether it is still true.

Why it matters here: relationship history, decisions and company rules that can be checked and corrected.
Open BlackBox ↗
LEAD QUALIFICATION
Opportunity score
84
✓ authority fits✓ timing signal✓ source verified✓ no duplicate trigger
03 / LEAD SOURCING + QUALIFICATION

“Find good leads” sounds simple until you have to teach a machine what good actually means.

So I built the standard around the evidence: who qualifies, why now, what is verified, how old a trigger is, what has already been used, and what the system should do when it cannot find enough good results. It is not allowed to lower the bar just to hit a number.

Why it matters here: qualification the team can inspect instead of a mysterious AI score.
COMPANY STANDARD · STAGED WORK
Do not write yet. First check the context, the standard and the exception rules that apply to this task.
04 / AI STANDARDS

A lot of my LLM work is basically this: what does good actually mean, and can I make the model hit that standard without somebody rescuing every first draft?

I use stages, rules, checks, examples of what not to do, and very specific output requirements. The subject can be editorial, company work, qualification or follow-up. The engineering problem is the same.

Why it matters here: follow-up, qualification and sales writing need a real standard, not “make it sound good.”
Open Cadence ↗
11 · a few recent posts

These are a few recent posts that probably give the quickest feel for how I have been thinking lately.

LINKEDIN · RECENTWhat you can do with Zapier's new updateOpen post ↗ LINKEDIN · RECENTMeta updateOpen post ↗ LINKEDIN · RECENTI feel like a genius sometimes when I figure…Open post ↗ LINKEDIN · RECENTAutomation inspo for teamsOpen post ↗ LINKEDIN · RECENTDecentering AIOpen post ↗
Open my LinkedIn profile ↗
12 · what I would not decide from the outside

There are some things I would not pretend to know from the outside.

I can show you the kind of system I would build. I cannot honestly tell you the final Property Wealth version until I know the answers below, because these are not small details. They change the architecture.

For me, they fall into three areas.

01THE WORK
What a genuinely good lead looks like here.That changes what we extract, what we score and what evidence the system needs before it calls somebody qualified.
Which customer journeys are actually different.Different offers, geographies or investor types may need different qualification, handoffs and follow-up instead of one universal funnel.
What the sales team really does inside WhatsApp.That decides where automation helps, where the conversation itself becomes qualification, and when a person should take over.
02THE PEOPLE
Which actions the company is comfortable automating.That determines approval gates, risk levels and which low-risk actions can eventually run without somebody clicking approve.
What managers keep checking or chasing.That tells me what should become obvious automatically and what the system should notice before a manager has to ask.
What leadership actually needs to see.That decides the first screen: what needs Daniel, what can wait, where money is stuck and which changes are worth interrupting him for.
03THE SYSTEM UNDERNEATH
Where the source of truth should live for each thing.Contacts, messages, files, calendars, payments and company rules do not all need to live in the same product.
What information is sensitive and who genuinely needs it.That changes permissions, what gets sent to AI, what can be exposed in a shared view and how small we can keep the blast radius.
What the team should still be able to do when something is down.That tells me what needs queueing, mirroring, a native-tool fallback or a manual recovery path before we call the workflow production-ready.
If I guessed these from the outside, I would be designing a demo, not your system. The machinery is the easy part to sketch. The useful part is learning enough about the company to make the machinery fit.
13 · if I joined

If I joined, I would probably start with one real journey and build from there.

Learn one real journey.The SOP, the real examples, and the people who actually handle it.
Decide what good looks like here.What the system needs to know, what it can do, what it should never do, and where a person still matters.
Build one complete slice.One source through to CRM, qualification, follow-up, handoff and recovery.
Run it next to real work.Normal cases, weird cases, bad inputs and things going down. Tighten it until we trust it. Then expand.