A one-partner valuation firm in Constantia, twenty past nine on a Sunday. Six residential bond valuations sit on the desk. Nedbank wants three by Tuesday. FNB wants two by Wednesday. Investec, one, no deadline stated, which in practice means yesterday. Each report needs a comparable-sales pull from PropStats or Lightstone, a Deeds Office title check on WinDeed, a photo set from Wednesday's site visit that never got sorted, a physical inspection reconciliation against the municipal valuation roll, and a written narrative that reads like a professional does not want to be sued by anyone.
A block down in Kenilworth on the same Sunday evening, an independent quantity surveyor is working through the second draft of a Sandton mixed-use refurb bill of quantities. Ten trades, a mid-rise structural steel frame, six variation instructions logged in the last week, and QS-side commentary the client's architect wants on his desk by Wednesday morning. The QS is not billing for Sunday. He is catching up on Sunday because Monday to Friday was site visits.
That is the practical shape of the property professions in South Africa in 2026. Both are governed by real regulators. The SA Council for the Property Valuers Profession on the valuer side. The SA Council for the Quantity Surveying Profession, backed by ASAQS, on the QS side. Both do work that is materially technical, materially quantitative, and shot through with dozens of small repetitive admin steps no client wants to pay for at professional rates. Both are exactly the kind of practice where AI, used well, changes the ratio of billable to non-billable time. Neither is a practice where AI signs the report.
Where the hours actually go
A valuer's week is roughly this: two days on the road (drive to Bellville, drive to Somerset West, drive to Milnerton, three inspections, back), one day pulling comparables and cross-checking title deeds, one and a half days writing reports, and half a day chasing the bond originator's admin team for missing instructions. The billable part of that is the inspection and the professional opinion. Everything else is the price of doing business.
A QS's week has the same shape and different content. Fewer client site visits, more contractor site meetings. More email. More time in Cubit or WinQS or CCS Candy pricing a bill against updated rates. And a growing volume of RFI and variation correspondence on active jobs, every subcontractor asking about a spec change a small clock the QS is running against a fixed fee.
On both sides, the arithmetic is the same. A single-practitioner or three-partner firm cannot buy its way out of the admin without either raising fees or losing the personal client relationships that keep them competitive against the corporate-side outfits. What they can do is push the admin down onto software that runs quietly.
The valuer's stack: comparables, title deeds, first-draft narrative
Three specific places AI earns its keep in a valuation practice.
Comparable-sales tidying. PropStats, Lightstone, and the Deeds Office data all get pulled routinely for every residential valuation. The pull itself is fast. The tidying is not. Pulling twelve raw comparables in Hout Bay, filtering out the sectional-title anomalies, the auction sales, and the family transfers, and reducing to the four cleanest arms-length comparables is fifteen to twenty minutes of a valuer's attention per report. A well-set-up AI-assisted flow, one that reads the raw pull, applies your firm's standard exclusion rules, and presents a shortlist for the valuer to confirm, makes that closer to three minutes.
Title-deed and endowment lookups. WinDeed and similar services return machine-readable data. The valuer still needs to read it. But an AI layer that extracts registered restrictions, servitudes, and the property's ownership history into a standard report subsection, ready for the valuer to review and sign off, removes a lot of typing and reduces the small copy-paste errors that show up in a Friday-evening report.
First-draft report narrative. This is the one people are most nervous about, and reasonably so. What works: a template-driven first draft that inserts the valuer's own findings (physical description, location commentary, market context) into a house-style narrative structure. What does not work: asking a general-purpose model to write the entire report from a few bullet points. The valuer is on the hook for every word. AI shortens the typing. It does not replace the reading.
The QS stack: BoQ pre-processing, RFI logs, variation letters
The QS side has a slightly different shape. Three places worth naming.
BoQ pre-processing. When an architect sends drawings for pricing, someone has to read them, extract quantities, and map each line to an SANS 1200 or POMI-standard item description. Tools like Cubit and PlanSwift do the take-off; the mapping to a priced BoQ line is the QS's job. AI-assisted mapping, pattern-matching take-off items against your firm's rate library and flagging anything unusual for the QS to inspect, cuts the routine part of the drafting time without touching the professional judgement on rate build-ups.
RFI tracking. On a live construction contract, the QS is often the person tracking RFIs and variation instructions across the principal contractor, subcontractors, architect, and client. The traditional method is a spreadsheet that eventually stops being reliable. A simple AI-assisted intake, where the site engineer photographs an instruction on WhatsApp, the system extracts the RFI text, drops it into a structured log with the responsible parties, and chases the responder after five working days, is a low-effort change with a real quality impact on the firm's admin file.
Variation and payment-certificate letters. A JBCC principal building agreement generates a lot of standard-format correspondence. Payment certificates, extension-of-time letters, notices of dissatisfaction, valuation of variations, notices of practical completion. Structured, high-volume, format-driven. AI produces a first-draft version from a structured input the QS provides, and the QS reviews and signs. In my experience this is where most QS firms get the fastest, cleanest win.
Site-visit coordination without the WhatsApp chaos
Both professions run on site visits, and both suffer from the same coordination pattern: a WhatsApp thread with the client, another with the bond originator or architect, another with the subject-property tenant, and a fourth with the office to confirm the diary. Something changes on the day, and the message that would have kept everyone in the loop lands in the wrong thread.
The lightweight fix for this is not a new practice management system. It is a booking flow that runs off one calendar, sends a confirmation and reminder over WhatsApp to whichever party the appointment involves, and routes the day-of message about a delay through the right people automatically. When a valuer is running forty minutes late because Rhodes Drive is at a standstill, the tenant gets a WhatsApp with the revised ETA, the office knows, and the valuer is not typing while stationary at a red light.
For a QS running a monthly progress inspection, the same infrastructure handles pre-visit RFI collection. The site agent uploads photos of the current works to a shared thread, the QS reviews the night before, and the visit itself gets spent on the questions that need the QS's eye rather than on catch-up.
Not glamorous. Real hours back.
What AI should not touch on either side of this fence
Worth naming clearly, because the market conversation tends to skip it.
For a valuer: the final valuation figure. The professional opinion. The narrative on encroachments, disputes, or anything where the physical inspection contradicts the paper record. Any bond-valuation report going to a lender under the Financial Sector Conduct Authority's oversight carries the valuer's professional liability, and the signature must reflect the valuer's own work. A first-draft assist is fine. Machine-generated conclusions are not.
For a QS: the professional opinion on rate build-ups, the certification of practical completion, the assessment of a contractor's variation claim on merit, and anything under the QS's fiduciary duty as principal agent. Same principle. The admin, the drafts, the coordination, that is where the software sits. The signed certificate is the QS's.
There is a specific failure mode I have seen twice in the last year and would rather not see again. A firm buys an AI tool that promises to generate the full report or the full BoQ. Junior staff use it to skip the reading step. The senior signs off in a rush. Six months later, an audit or a professional-liability query surfaces an obvious error nobody caught. The tool is not the problem. The workflow around it is.
SACPVP, SACQSP, ASAQS and the POPIA layer
Both professional bodies have made statements about AI use over the last eighteen months, and both boil down to the same requirement: the registered professional remains responsible for the work product regardless of what tools produced it. That is not a warning against AI. It is a clear framing of the correct division of labour.
POPIA sits over both practices with real weight. A valuer's report contains the owner's ID number, the property address, and often the sale history. A QS's file contains subcontractor rates, contractor payment schedules, and sometimes commercial information a listed client would rather not see leaked. Neither should be dropped into a consumer AI chat window. The processing should happen inside infrastructure your firm controls, or through a vetted processor with a signed operator agreement under Section 21 of POPIA.
Four things to check before turning on any AI in either practice:
- Where client data is stored, and whether it can leave South Africa. POPIA Section 72 has specific rules on this.
- Who at the vendor can see the data. Least-privilege access, logged.
- What happens on deletion. A POPIA erasure request must reach every AI system that touched the record, including any embeddings or model memory.
- What audit trail exists for AI-generated content. When the Information Regulator or the professional body asks how the report was produced, you must be able to answer without guessing.
Ask a prospective provider to walk you through their answers to these four in specifics. If they cannot, that is your answer.
A practical starting point
The instinct is to look at all of the above and try to fit it into one project. The firms that actually get value do the opposite.
For a valuation practice, the fastest first project is comparable-sales tidying paired with a first-draft report scaffold for one specific report type, usually residential bond valuations, because that is the highest-volume, most structured product. The valuer keeps their own methodology. AI shortens the typing.
For a QS practice, the fastest first project is variation and payment-certificate correspondence drafting, tied to a template library the firm already uses. Low-risk, produces visible time savings inside the first month, and does not touch anything that changes the shape of professional work.
Both firms end up with the same lesson after the first successful project. The software itself is quiet. What shifts noticeably is the weekly rhythm. A Sunday evening no longer disappears into report typing. A Friday afternoon reaches the practitioner instead of the admin backlog. Client relationships improve, not because AI did anything client-facing, but because the professional has time to make the phone call that used to slip.