From paper BOQs to AI-powered estimates — a contractor's migration story

Every contractor in MENA who is now running AI-assisted estimates was, five years ago, running paper and Excel. The path between those two states is not a single switch flip. It is a five-stage migration that takes most contractors between six and eighteen months to complete.

This is the honest map of those stages, written from inside an Aletlala-style transition and the pilots we are now running with three more contractors across the GCC.

Stage 0 — paper and Excel

This is the starting state for most of the market. BOQ comes in as PDF or scanned paper from the consultant. The QS retypes it into Excel, prices line by line against last project's rates, sends back the priced BOQ as another PDF.

Pain points are well known:

• Rate inconsistency between QSs.
• No audit trail for why a rate was chosen.
• Re-typing errors that survive into the priced BOQ.
• No reuse across tenders.

Stage 0 is not stupid. It is the rational response to lacking better tools. A senior QS pricing 1,200 lines from memory and gut feel is more accurate than a junior QS pricing from an unfamiliar system.

Stage 1 — structured Excel with a rate library

The first migration step almost every contractor takes themselves. The QS team builds a "master rate library" in Excel and starts using VLOOKUPs instead of free-typing rates.

What improves:

• Rate consistency across QSs (somewhat).
• Faster pricing of repeat items.
• Some audit trail (the rate library version).

What breaks:

• The library drifts. By project 20 there are seven copies.
• Material price changes are not propagated.
• Cross-project history is still per-QS.

Most contractors hit the ceiling of stage 1 around project 30 or 40. The library is unmaintainable. That is when stage 2 becomes urgent.

Stage 2 — central BOQ database

Move the rate library out of Excel and into a structured database. Every BOQ item gets a code (CSI MasterFormat or local equivalent), every rate has a date and a source, every BOQ links to a project.

What improves:

• Rates are queryable. "What is our last-six-months average for 25mm PVC conduit?" becomes a one-line query.
• Material price changes propagate.
• The QS spends less time hunting and more time judging.

What contractors learn the hard way:

• BOQ coding is political. Two senior QSs will disagree on what category an item belongs in. Pick a coding scheme and enforce it.
• Historical Excel needs migration. Most contractors do not bother; they start clean from project N and live with two regimes for a year.

ORKSTRA implements stage 2 with a multi-region item catalog: 245 UAE-specific materials seeded, with extensions for KSA and Egypt. That seed gets a contractor to a working catalog on day one rather than week 20.

Stage 3 — AI-assisted pricing (one model)

The first contact with AI is usually a single-model price suggester. The QS sees a suggested rate next to the manual entry field. Accept or override.

What improves:

• Coverage on routine items is excellent. The QS focuses on the novel and the high-value.
• Suggestions ground newcomers. A junior QS can produce a senior-quality BOQ with senior oversight.

What goes wrong:

• Single-model AI hallucinates confidently. The QS catches the worst cases. Some get through.
• Trust erodes fast when the first wrong suggestion makes it to a consultant. Adoption stalls.

This is the stage where many contractors regress to "AI is a toy". The fix is not abandoning AI. It is moving to stage 4.

Stage 4 — multi-source AI with verification

The mature stage. AI suggestions are backed by four sources weighted behind a single number:

40% historical, 30% government rates, 20% crowd-sourced, 10% heuristic AI fallback.

The QS sees not just the number but the spread between sources, the confidence band, and the option to drill into any source. Critical items go through Triple-AI Verified Takeoffs — three independent vision models — with disagreement flagging.

What improves:

• Trust is rebuilt because the AI shows its work.
• The QS catches the disagreement cases (5-15% of items) and lets the rest flow.
• Audit trail is automatic and clause-defensible.

What contractors still need to invest in:

• Acceptance/rejection becomes training data. Closing the loop matters. Without it, the AI never adapts to your house style.
• Catalog hygiene. Garbage in, garbage suggested.

Stage 5 — closed learning loop

The endgame. Every accepted or rejected suggestion feeds back into a per-tenant learning loop. After 200-300 corrections, the AI is biased toward your style. After 500-1000, the acceptance rate plateaus high.

The per-tenant aspect matters. A KSA government infrastructure contractor and a UAE high-end villa builder should not share a model. Their pricing logic and material preferences are fundamentally different. ORKSTRA enforces per-tenant isolation on the learning loop by design.

Aletlala reached stage 5 around month six. Acceptance rate moved from 62% to 81% over 340 corrections. The compounding effect is what makes AI estimating economically real, not a demo.

The migration timeline

A realistic timeline for a mid-size MENA contractor:

| Stage | From | Effort | |---|---|---| | 0 → 1 | Paper | Weeks (Excel discipline) | | 1 → 2 | Excel library | 2-3 months (catalog build) | | 2 → 3 | Central DB | 1 month (AI integration) | | 3 → 4 | Single AI | 2-3 months (verification layer) | | 4 → 5 | Multi-source | 6+ months (loop maturity) |

Most contractors skip stages by joining ORKSTRA at stage 4 directly. The 245-material catalog, multi-source pricing engine, Triple-AI router and per-tenant learning loop are all in place from day one. What the contractor brings is their own project history — which the platform absorbs into the historical source and the learning loop over the first six months.

What does not change

Two things the migration does not touch.

The senior QS is still the bottleneck on judgement. AI accelerates the routine and surfaces the contentious. The judgement call is still human.

The supplier relationship still matters. The platform suggests a rate based on data. The negotiation happens human to human, and the last 4% of margin lives there.

Where to start

If you are anywhere between stage 0 and stage 3, the next step is not "more Excel". It is a structured catalog plus an AI-assisted suggester with verification. We can show you what stage 4 looks like on your own BOQ.

• /demo — bring a recent priced BOQ. We will show you the stage-4 view on it.
• /premium-stack — the technical map of the pricing engine and learning loop.

— Eng. Amr Shoieb