SMD Social media judged by the work it wins.
SMD is one system that creates, distributes and measures social media for every project I run. It judges each post by the leads, pipeline and won work it produced, read from each brand’s own CRM, not by the likes.
- Role
- Sole builder
- Built as
- Web app and a local worker
- First networks
- Instagram and YouTube
- Status
- Running locally
- Receipts
- 19 commits · 43 test files · since 30 September
Performance
| Post | Pillar · hook | Clicks | Enquiries | Leads | Pipeline | Won | Impr. | Reactions |
|---|---|---|---|---|---|---|---|---|
The close nobody times | Month-end · number | 96 | 4 | 3 | £22k | £18k | 3,410 | 58 |
Close: 12 days to 4 | Measured · number | 71 | 3 | 2 | £19k | — | 2,880 | 44 |
A worktop, measured twice | Fit-outs · story | 52 | 1 | 1 | £17k | — | 8,400 | 162 |
Your brand guide is a PDF nobody opens | Audits · contrast | 38 | 1 | 1 | £6k | — | 1,120 | 19 |
Is your team ready for AI? | AI in practice · question | 12 | 0 | 0 | £0 | — | 6,950 | 211 |
| First touch | Leads | +1 mo | +2 mo | +3 mo | +4 mo |
|---|---|---|---|---|---|
| June | 2 | £0 | £6k | £18k | £18k |
| July | 1 | £0 | £0 | £0 | — |
| August | 3 | £0 | £12k | — | — |
| September | 1 | £0 | — | — | — |
Likes are easy to count and say nothing.
I run several brands, and each one’s socials lived somewhere different, judged by reach. Reach does not pay the bills. The question that matters is which post brought the enquiry that became the work, and answering it needs the post, the link and the CRM in one place.
Every post carries its own link.
Each brand is a record with its own voice, look and posting rules, and a brain of offers, proof and objections. A post is drafted from those records, goes out with a short link that carries its UTM, and is judged by what the brand’s CRM says happened next. An AI console works over MCP, reading the same brain a person does: it can draft, but it cannot schedule or publish.
Studio draft
Twelve days. That is how long one finance team we worked with took to close a month.
Every invoice was matched by hand, every accrual rebuilt from last month's sheet.
After: four days, and nobody rebuilding anything.
Most teams have never timed this. We will time yours in 15 minutes, free: halden.example/go/c7k2
Halden: analytics
| Question | Progress | So far | Stops at |
|---|---|---|---|
| Number hook or story hook? alternate posts | 3 leads to 1 | 10 posts each | |
| Carousel or text, same idea? pairs, a week apart | too early | 6 pairs | |
| Review page or home page? half the links each | proposed | — | 200 clicks |
Calendar
Bots
| Bot | May | State |
|---|---|---|
| Instagram replier | approved answers only | on |
| LinkedIn triage | draft; you send | on |
| Instagram replier | approved answers only | on |
| LinkedIn poster | post what you approved | paused |
Screens drawn with dummy data. No real customer, client or account appears.
What it does
A brain per brand
Offers, pillars, lead magnets, proof and objections, kept as records rather than prompts, so a person, the console and the studio all read the same thing. An offer with no lead magnet is flagged.
One idea, many shapes
The studio turns one idea into a post, a carousel, a reel script or a short. Cards are drawn from code in the brand’s palette, with no stock imagery, and the author’s Publish is the approval.
Performance from the CRM
Clicks come from SMD’s own short links; enquiries, leads, pipeline and won work come from each brand’s CRM. Cohorts follow leads by the month they arrived, and what people say on the form stays in its own column.
Analytics that stay honest
What moves leads, where the funnel leaks, and experiments that change one thing at a time. Every finding says how many posts it rests on, and says when that is too few to tell.
Learnings, with sources
Videos are transcribed on this machine and distilled into dated claims, each with its source and the moment in it. Claims that disagree sit side by side, searchable by words and by meaning.
Bots that cannot be steered
A bot posts what was approved and answers comments from replies a person wrote in advance. A comment can choose an answer, never write one, and one that reads as instructions is held.
More of the work
Sentinel
A Mac companion for AI agents
One window for what an AI agent needs from you, what it's doing and in what order, what it has learned, and what it left running. It keeps the Mac inside a budget, and every check is broken on purpose to prove it works.
Roomi
Pitch the product, not the deck · getroomi.com
A platform for deploying working demos into a pitch room: the demo, the narrative and the materials behind a link per viewer, with analytics and feedback coming back.
Tengen
AI algorithmic trading, built data-first
A trading platform whose edge is connecting cause to effect as it happens: which event moved which price, and why. Underneath, a data foundation where every source, run and figure is recorded and can be traced.
Daedalus
Co-founder and CTO · since 2022
Data engineering, data science and software architecture for B2B financial-services ventures. The client work is private; the approach is everything on this site.
Alfred
A chess coach that runs on your Mac
Paste a game or drop in a screenshot and it shows the best moves in order, explains them in plain words, plays them out on the board, and learns your patterns across games.