AI in quantity surveying: where automation helps and where QS judgment remains critical

A quantity surveyor is a cost professional: they measure construction work, prepare bills of quantities, estimate project costs, manage procurement, certify payments, and advise on contract risk. On a development project in Kenya, the QS sits at the intersection of design and construction — translating drawings into numbers that contractors can price and clients can commit to.

The QS role is fundamentally data-intensive. A bill of quantities for a medium-sized residential block may contain thousands of line items, each requiring accurate measurement from drawings, appropriate rate application, and correct coding. Manual errors in measurement or rate application compound across the bill and can produce a cost plan that is significantly off.

This is the area where AI and automation are genuinely useful. Pattern recognition, cross-checking, benchmark comparison, and data entry are tasks that machines do well and that consume a disproportionate share of QS time. The question is not whether automation is relevant to quantity surveying — it clearly is — but where the automation adds value and where it falls short.

The most immediate application of AI in QS practice is preliminary cost estimation. Before drawings are complete, a client or developer needs a cost order of magnitude: is this project in the range of KES 50 million or KES 150 million? What is the rough split between structure, envelope, fit-out, and MEP? How do the numbers change if the specification is upgraded from mid-range to high-end finishes?

Traditional preliminary estimates rely on the QS's experience — mental benchmarks from past projects, adjusted for current market conditions. This is not a bad process, but it is slow, undocumented, and difficult to validate. Two experienced QSs on the same project may produce estimates that differ by 20-30% without either being wrong in principle.

AI-assisted cost estimation uses documented benchmarks derived from real project data: measured rates by construction type, building class, and location, updated regularly as market data comes in. A preliminary estimate generated from these benchmarks is traceable — the client can see which benchmarks were used and on what assumptions the estimate rests. Disagreements become technical discussions rather than credibility contests.

In the REDM system, the cost benchmarking engine supports the feasibility stage of every project. The preliminary estimate is generated from the project's defined floor area, use type, specification level, and location, and is updated as the design develops. The QS reviews and signs off the estimate before it goes to the client.

A completed bill of quantities contains hundreds or thousands of measured items. Errors can appear as transcription mistakes (wrong unit, wrong quantity), measurement errors (incorrect dimension or incorrect calculation), rate errors (outdated rate or wrong category), or omissions (items missing from the bill that are visible in the drawings).

Traditional QS practice uses a second checker — another QS reviews the bill for errors before it goes to tender. This is good practice but expensive: it doubles the QS time on the checking stage, and checkers are not infallible.

AI-assisted bill checking uses pattern recognition to flag statistical outliers: items priced significantly above or below the benchmark for that category, items with unusual quantities given the overall project size, units that are inconsistent with the item description, and cross-references between the bill and the drawing schedule that do not match. This does not replace the second checker, but it focuses their attention on the items most likely to contain errors.

For procurement, AI can also assist with contractor shortlisting and tender evaluation: flagging unusual pricing patterns in bids, identifying items where a contractor has significantly underpriced (which often signals a claim strategy), and benchmarking the bid against the cost plan.

There is a clear limit to what AI can do in quantity surveying, and it is important to be precise about where that limit lies.

Market intelligence is not automatable. A QS who knows the Mombasa subcontractor market knows which MEP contractors are currently busy and likely to price high, which concrete subcontractors have had quality issues in the past year, and what the current premium is for steel given recent import tariff changes. This intelligence affects procurement strategy in ways that benchmark data cannot capture.

Contract risk assessment requires professional judgment. A QS reading a contractor's tender can assess whether the programme is realistic, whether the preliminaries cover the likely site establishment costs, whether the contractor has understood the scope, and whether there are signs of a front-loaded pricing strategy. These assessments draw on experience and professional knowledge that cannot be automated.

Variation management during construction is also judgment-intensive. When a contractor submits a variation claim, the QS must assess whether the varied work falls within the original contract scope, whether the pricing method is appropriate for the type of change, and whether the claim value is consistent with the market. Automation can assist with cross-referencing the claim against contract documents, but the substantive evaluation is professional work.

Finally, final account negotiation is a professional skill — not a data process. The QS represents the client's or contractor's interests in resolving the final account and requires understanding of contract law, construction practice, and negotiation, not just arithmetic.

The right model for QS practice in 2026 is not 'AI does it' or 'QS does it' — it is AI handles the data-intensive, repeatable tasks, and the QS focuses their time on judgment, risk, and professional advice.

This is the model that Architect Darani's QS service uses. The REDM system handles preliminary cost benchmarking, rate database maintenance, bill cross-checking against drawing schedules, and payment certificate tracking. The QS's time is concentrated on specification review, procurement strategy, tender evaluation, contractor risk assessment, and variation management.

For clients, this means faster turnaround on preliminary estimates (measured in hours, not days), better-documented cost plans with traceable assumptions, and a QS whose attention is focused on the parts of the project where professional judgment has the highest value.

For developers managing multiple projects, it also means consistency: cost benchmarks and rate databases are maintained centrally, so the preliminary estimate for a new project draws on the same validated data as the estimates for the current portfolio, not on a different QS's recollection.

When you commission a QS to prepare a cost plan, it is worth asking how they produce their preliminary estimates and what their benchmark data is based on. A QS who can show you documented rate tables and explain the assumptions in the estimate is giving you something more useful than a verbal number.

It is also worth asking how errors in the bill are checked. A firm that uses AI-assisted checking alongside a human review has a more robust quality process than one that relies solely on a manual second check.

The REDM project check produces a preliminary cost estimate before the QS is formally appointed. This gives the client a documented baseline for the design feasibility conversation, and means the QS's formal cost plan is a refinement of an established estimate rather than a first opinion.