Briefing
Twenty minutes. First, enough of the technology that we all use the same words. Then the numbers for a 1,500-unit wildlife estate.
Part one — the technology
Given a sequence of text, it predicts what comes next. Everything else is built on top of that.
The loop runs once per token. What people call "reasoning" is the model spending more tokens thinking before it answers.
Part one — the technology
A chunk of text, roughly four characters. Models read, price and limit work in tokens, not words.
How much text it can hold at once. Everything it knows about your task lives there, and is gone on the next call.
The trained parameters. Open-weight models can be downloaded and run on your own hardware.
How adventurously it picks. Low for data work and code. Higher for drafting and ideas.
Part one — the technology
The right-hand column is the specification for what an agent has to solve.
Part one — the technology
A chat window gives the model words.
An agent gives it hands, memory, and a job.
Part one — the technology
You send a request. Hermes gives the model tools, reads the real results back, and repeats until the work is actually done.
It never guesses what a script returned. It runs the script and reads the result.
Part one — the technology
01
Files, spreadsheets, databases, scripts, web, email. How the model touches anything real.
02
Durable notes loaded every session, so what you explained once does not need explaining again.
03
Your procedures on disk, loaded only when the task matches. Your SOPs, not the model's defaults.
04
Scheduled and background work, so levy reminders, statements and compliance checks run overnight and are waiting in the morning.
Part one — the technology
A skill is a written procedure with a trigger. Hermes loads it only when the work calls for it.
Your value is twenty years of knowing how this scheme wants its governance run. A skill is that knowledge written down once, applied consistently, and no longer trapped in one person's head or lost when they are on leave.
Part one — the technology
Protea is a privacy and memory layer, used as the example of the right architecture for regulated work. It sits between a client and any model backend, so you get the frontier model and keep the data.
Named entities — people, schemes, estates, addresses — are replaced with stable placeholders before the request leaves your hardware, and restored in the reply. The model never sees an owner’s name, ID number or bank details, so you can run frontier models without an exception.
A single SQLite file per user behind an OpenAI-compatible API. An embedded local model means it runs fully offline once the model weights are pulled. No third party ever holds your corpus.
Structured recall across conversations and months — spiral-indexed by relevance and recency — rather than starting from nothing every session. Twenty years of knowing how this scheme runs stops living in one person's head.
Conductor routing passes each task to the model best suited for it; a confidence cascade escalates only when needed: local model for the routine, frontier model for the hard call. Consensus fans a delicate question to several models and synthesises the answer. Teams decompose a brief, rate deliverables, and review every result. An Open WebUI front-end gives clean per-user separation, with identity forwarded through the API.
The pitch to the scheme: "we run AI-assisted administration and our owners' data never leaves our environment." Very few small vendors can say that sentence truthfully.
docs/README.md; redaction pipeline docs/redaction-algorithm-research.md (local: D:\Projects\Protea\docs)Part one — the technology
Asked for something it cannot know, a model produces a fluent, plausible, wrong answer rather than nothing.
An invented variable name. A confident number with no source. A mapping that looks right and is not.
Read before answering, run before claiming, show the source. Grounding in real output — and a person signing off. Never autonomy over a regulated deliverable.
Part one — the technology
The evidence is not "experts get faster" — it is that expertise is what lets AI be used on the hard stuff at all.
The model cannot produce the question that only years of knowing schemes and the STSM Act produces. The value is upstream of the answer.
AI is fluent and confidently wrong. Inside its capability boundary it gains ~40%; pushed past it, performance drops 19 points. Only experience says where that boundary is.
The model suggests; the domain expert decides. A 19-point penalty exists precisely because it is not obvious to a generalist which tasks AI can do and which it cannot.
A senior hand encodes their own judgement as a skill. Their expertise stops being a bottleneck and becomes repeatable.
AI compresses the skill gap on routine tasks — a beginner with AI reaches near-expert output on well-defined work. But a scheme does not win on routine. It turns on the messy, consequential, one-off judgement where a confident wrong answer costs the scheme.
The real gain for an expert is not doing the same task faster. It is having a tireless assistant who drafts and cross-checks, while the expert spends the saved hours on the resolutions, the AGM, and the relationships no model can hold.
Part two — the numbers
The question is not whether AI is impressive.
It is how much of a R93-million-a-year estate is still run by hand.
Part two — the numbers
A wildlife estate at 1,500 units with 45 employees is not a small operation. The trustees are volunteers; the staff and the administration are not.
59%
The share of a property team's time spent on administrative tasks. Across 45 staff that is levy statements, utility bills, reminders, minutes, quotes, compliance packs — and the same owner query answered for the hundredth time.
≈45,000 hrs
A year of administration across 45 employees — roughly 76,500 staff hours, of which the majority is the repetitive kind that never needed a person to sit through it.
That second box is the target. Not the 45 employees — the 45,000 hours of work nobody values and every owner funds.
Part two — the numbers
R93m
R45m in levies plus ~R48m of bulk water and electricity — R4m a month moving through the estate's books.
15,587
Applications to the Ombud — 62.5% of them linked to levy arrears.
R15,000
An unopposed Section 66 application, before the debt. Four adjudicators cover the country, 450+ cases each.
You do not need all of it. You need the estate that stops bleeding it — and can show 1,500 owners the numbers.
Part two — the numbers
Annual cost of administration, collections and compliance, ZAR. Modelled on this wildlife estate: 1,500 units, a R2,500 levy, and R4m a month of bulk utilities — about R93m administered a year.
Part two — the numbers
Annual contribution to the reserve fund, ZAR. For an estate this size, waiting is not neutral — arrears and utility-recovery failures keep eating the fund whether or not management acts.
Part two — the numbers
| Path | Admin cost | Admin hours | Reserve contribution | Special-levy risk | Cumulative saving 2026–31 |
|---|---|---|---|---|---|
| Start now (2026) | R4.25m | 17,000 | R4.5m | Low | R14.75m |
| Wait until 2029 | R5.65m | 27,000 | R2.4m | Moderate | R7.85m |
| Do not act | R9.12m | 45,000+ | R0 | High | R0 |
R6.9m
Extra cost carried by starting in 2029 instead of 2026 — money that belongs in the reserve fund of a 1,500-unit estate.
≈R150k
First-year programme: setup, the AI subscriptions for the admin and utility teams, and the time to write the estate's procedures down properly.
Part two — the numbers
Every scenario keeps the same 45 employees. The gain is governing 1,500 owners and R93m of their money well — not thinning the staff.
The dull half. Statements, utility reconciliation, reminders, minutes, the fourth version of the same letter, chasing the same arrear across 1,500 owners.
Fiduciary judgement. The AGM of 1,500 owners. Security decisions. Who signs off on a repudiation or a disconnect.
What the same 45 people can carry — more owners served well, more of the estate run properly, and a reason to pay them more.
An estate does not win by employing fewer people. It wins by being governed well enough that owners stop asking why levies and utilities went up.
Part three — where to point it
1,500 owners: statements, the reminder ladder, the letters, the s66 or CSOS pack. The agent approves and sends.
R4m a month of water and electricity — metering, recovery, exceptions. The estate's single biggest cash line.
Notice, agenda, minutes and resolutions for 1,500 owners — first draft from last year's pack and this year's numbers.
CSOS registration and governance documents, rule amendments, the reserve-fund plan. Rule-based and reviewable.
Roads, water reticulation, electrical. Route owner reports, chase quotes, track jobs to done. An exceptions list each morning.
Incident logs, access exceptions and the monthly security report — summarised for the trustees.
Common-property claims: incident note, quotes, photos, the policy reference. A person signs it off.
Circulars, rule reminders, FAQ answers for 1,500 households — in the estate's voice, not the model's defaults.
Part three — where to point it
The interesting part is not doing the same work cheaper. It is what becomes possible at this scale once the marginal hour is cheap.
Reconcile bulk water and electricity against 1,500 meters every month. At R48m a year, even 1% recovered is R480k.
Flag the owner at 30 days, not 120 — across 1,500 accounts. Cheaper than a s66, and the difference between a blip and a crisis.
Governance documents kept current and CSOS-ready all year, instead of a scramble before the AGM.
Statements, rules, requests and answers on demand for 1,500 households. Fewer calls; the staff do the judgement work.
A living maintenance and reserve plan for a large estate, updated from real spend — the STSM plan taken seriously at last.
Owner names, ID numbers, bank details. Local models mean the benefit without the data leaving the estate.
Part three — where to point it
Pick the single most repetitive task — levy reminders or the monthly utility reconciliation. Time it honestly for two weeks first.
Write that procedure down properly as a skill. This is the real work, and it pays off even if the AI part fails.
Run it alongside the team for a month. Compare outputs. Keep the human sign-off permanently.
Measure the hours recovered and the rands recovered on utilities. That is the whole thesis — prove it on one line.
1,500 units, R2,500 average monthly levy (~R45m a year), plus bulk water and electricity of ~R4m a month (~R48m a year) — about R93m administered annually. 45 employees, ~76,500 staff hours, of which 59% (≈45,000 hrs) is administration. Arrears handled at up to R15,000 per unopposed Section 66 matter.
Repetitive admin absorbed over ~4 years, releasing roughly 60% of those hours by 2031. Costs in the "do not act" path rise ~7%/yr, consistent with the survey finding that 91% of schemes face unexpected increases.
Sources: unit count, staff count, levy and utility figures are the estate's own, supplied by management — not published averages. CSOS disputes and arrears share (CSOS 2023/24 Annual Report via REI); adjudicator capacity and R15,000 s66 cost (ANGOR); admin-time share 59% and turnover 29.2% (NAA via RealPage); AI adoption 20%→58% (Buildium 2026 via NAR); reserve-fund minimum (STSM Act Reg 2); cost-increase survey (Foundation for Community Association Research).
Close
The technology is ordinary. It predicts text, and a framework gives it tools and memory.
The opportunity is not. A R93-million estate, run on 45,000 hours of unpaid administration a year.
Nothing here requires betting the estate. It requires writing down how you govern, and starting.