On a large GCC construction build or capital project, the thing that quietly stalls progress is rarely a system going down. It is a number no one can trust.
Clean procurement data is not a nice-to-have on a capital project. One wrong cell in a Bill of Quantities (BOQ) is all it takes to stall a contract approval, slip a handover date, and trigger delay penalties. This piece breaks down exactly how bad supplier master data and spend data create project delays, and what a structured data-cleansing program looks like as an executable fix.
A new tower is going up. Procurement is running dozens of contracts at once, and a single bill of quantities pulls together pricing from several contractors, revised more times than anyone has counted.
One morning a line in the BOQ stops calculating. A supplier code was renamed upstream; a lookup can no longer find its match, and the total it feeds back returns an error. The approval that depended on that total is not rejected. It just sits, while everyone assumes someone else is holding it.
By the time the broken reference is traced and fixed, three working days have gone. There was no incident and no outage. Nothing that would show up in a report. In a program where the handover date incurs delay penalties, three quiet days are not a rounding error.
This isn’t a spreadsheet problem. It’s a trust problem.
The broken BOQ cell is the symptom. The real issue is that the team has quietly stopped trusting its own supplier master data.
Most construction procurement teams already know the master is not fully reliable. So they check it by hand before any award that matters and keep a private spreadsheet they trust more than the system. None of it is logged. It becomes habit, and habit is invisible, which is exactly why it never gets fixed.
The better a team gets at working around bad procurement data, the longer the bad data survives.
Where the cost of bad procurement data actually hides
When the supplier master can’t be trusted, the cost shows up in three places that never make it into a report:
- Approval cycle time –Every reviewer quietly re-checks what they’ve been handed before a contract goes out, so each award takes longer than the workflow says it should.
- Supplier onboarding –A new subcontractor who should be set up once gets entered, corrected, and chased across teams.
- Spend categorization –The cost picture a CPO carries into a project review is wrong before the conversation begins, because the records underneath it are wrong.
It doesn’t fail loudly. It surfaces as time, usually a slipped date on a program where the schedule carries penalties. The better a team gets at working around bad data, the longer the bad data survives.
Data does not replace judgement, it strengthens it.
— Ahmad Raffat , Head of Procurement & Supply Chain, A.R.M. Holding, in The Procurement Ledger
How to Fix It: A structured construction procurement data cleansing program
A one-off clean-up doesn’t hold. It decays back within a quarter, because new bad data arrives on the next load. So, we don’t run a clean-up and leave. We run a structured approach that makes clean data the default state, and it doesn’t start with you handing over a tidy file. The messy export is the starting point.
Here is how the work runs across six stages, most of it carried out by the data team, not another project for your procurement or IT staff to resource:
Stage 1: Discovery and Data Mapping
Short sessions with procurement, IT, and finance to map where data is created, where it breaks, and what “clean” must mean for your project reporting.
A typical finding: the supplier list lives in three places simultaneously, the ERP, the procurement team’s spreadsheet, and the quantity surveyor’s own file, and they do not agree. No data engineering begins until the target state is agreed in writing.
Stage 2: Field-Level Data Profiling
Every field in the spend export is checked against what is actually there: duplicate vendor records, blank or vague spend categories, BOQ lines returning #ERROR, rates keyed in the wrong currency, negative quantities, and two purchase orders sharing one reference number. Format errors are logged with proposed fixes before any transformation runs.
Stage 3: AI-Powered Feature Extraction
Unstructured procurement descriptions are parsed at scale using LLM-based extraction tuned to construction and capital project vocabulary. A single line reading “ready-mix C40, supplier ref and delivery note combined” comes back as separate structured fields for material grade, supplier, and quantity, in whatever language the original was written.
Stage 4: Vendor Master Deduplication
“ABC Steel”, “ABC Steel LLC”, and “A.B.C. Steel Trdg” resolve to one supplier golden record. The same rebar grade entered under three material codes becomes one item. Every merge retains a full audit trail, source IDs, the matching rule that triggered it, the date, and confirmation from a data steward. Silent merges that destroy audit trails are not part of the methodology.
Stage 5: Spend Taxonomy Classification
Everything is sorted into one consistent category structure that matches how the project reports: concrete, structural steel, mechanical and electrical, finishes, and civil works each roll up the same way across packages and across projects. UNSPSC or a custom construction hierarchy is applied depending on reporting requirements.
Stage 6: Ongoing Data Governance
Agreed naming rules and automated quality checks run on every new data load, so the clean-up holds. A new subcontractor can only be added through one approved route instead of being typed fresh into a BOQ each time. The governance layer sits on top of whatever system you already run, SAP, Coupa, Ariba, or a plain spreadsheet export. The ERP stays the system of records. Nothing is ripped out.
What a first pass data audit usually finds
You don’t have to take this on trust. A proof of concept runs the whole approach on a sample of your real purchase-order data, the messier the better, and needs one category or data lead for about four hours. Within a few weeks you get a quality scorecard for your own data and a list of the surprises that always turn up. On first-time client data, these are typical, and they vary by sector and data maturity:
- 8–15% of rows carrying hidden duplicate spend
- Currency and formula errors sitting inside exports everyone assumed were clean
- 85–95% field-extraction accuracy on the records processed in that first pass
These aren’t guarantees. They’re what the first look are reliable surfaces, and they’re usually enough to size the problem in numbers a CFO will accept.
What clean procurement data unlocks for leaders and stakeholders
- CFO –A spend picture that holds up under scrutiny, duplicate and leaked spend surfaced in numbers you can defend, and no disruption to the finance system of record.
- CPO –Project reviews built on records that stand up in a steering committee, and a clean data foundation in weeks rather than a multi-year program.
- Buyer –A supplier master you can act on without a manual re-check, and onboarding that happens once instead of three times.
Where to start
You don’t need a transformation program to find out whether this applies to you. You need to look at twenty rows.
Pull just a 20-row sample from your supplier master and read it carefully. Count how many records are duplicated, wrongly coded, or out of date. Twenty rows will tell you more than a dashboard will, because the dashboard is built on the same records.
If you want to understand:
- What your supplier master and spend data actually look like at the first audit
- Where the duplication and miscoding concentrate in your spend
- How a fix maps onto the ERP you already run, without replacing it
That’s a useful conversation to have.
Ready to scope what fixing your construction procurement data looks like?
We’ll walk you through exactly what your supplier master and spend data look like at first audit, and how to clean it without replacing your ERP.
