What a new head of procurement should do in week one

Quick answer: In your first 30 days, do not reach for a savings target or an AI tool. Audit what spend data you actually have, find out who, if anyone, categorises it today, and name a single owner for data quality before you build a strategy on top of it. Skipping this step means every report you present later inherits the same untrustworthy numbers.

On this page: The baseline gap you inherit · Why AI is the wrong first move · What to actually do in week one · Why data quality needs an owner · FAQ

The baseline gap new procurement leaders inherit

Most people who start a new head of procurement role ask for the same thing in their first week: a categorised view of spend. What does the business buy, from whom, at what price, and how does that break down by category. In hard FM, construction, HVAC, manufacturing and infrastructure, this view almost never exists in usable form. It exists as an ERP export with inconsistent supplier names, free-text descriptions, and a category field that was filled in once, years ago, and never touched again.

This is not a rare gap. Industry benchmarking on spend under management puts the average enterprise at around 71% visibility into total spend, and that figure covers established procurement functions with people, process and tooling already in place. A new head of procurement walking into a role with no prior category structure is typically starting well below that average, not at it. Separate research on how organisations actually collect spend data finds that roughly six in ten still rely on manual collection and analysis, which is a direct explanation for why the baseline is missing rather than just outdated.

The person who asks for a category breakdown in week one usually gets one of two things back: a spreadsheet someone has to rebuild by hand, or a report the ERP produces that only makes sense to the person who built it. Neither is a baseline. Both take days to produce and neither survives a follow-up question from the board.

None of this happened through neglect. Nobody sat down and decided to let the category structure rot. It happened because classifying spend accurately was never anyone's actual job. It was a task bolted onto procurement, finance or IT, done well for a while by whoever cared most, and then quietly abandoned when that person moved on or got busier. The data did not break. It was simply never owned.

Why AI is the wrong first move

Once a new head of procurement confirms there is no baseline, the instinct is almost always the same: point an AI tool at the raw export and let it sort the mess out. Microsoft Copilot or a general-purpose model can classify a spreadsheet in minutes, and against a totally uncategorised dataset, minutes feels like a win compared to weeks of manual sorting.

The problem shows up at scale, not in the demo. Enterprise benchmarking on generative AI puts hallucination rates for commercial models anywhere from roughly 15% to over 50% depending on the task, and open-ended classification against ambiguous, company-specific categories is exactly the kind of task where that range sits at the higher end. A general model has no memory of the rule it applied to line 40 by the time it reaches line 4,000, and no audit trail explaining why "3/4 inch ball valve" landed in one category on one export and a different one on the next.

This is not a case against AI. It is a case against AI as the first move, before there is a category structure and a rule set for it to apply consistently. Point a general model at a mess and it gives you a confident-sounding, differently-shaped mess back, faster than a human would have. Showing a leadership team one report built on hallucinated categories does more damage to trust in the function than showing them a slower, honest "we don't know yet."

What to actually do in week one

Resist the pull toward a savings target, a technology roadmap, or an AI pilot in the first month. None of those can be built reliably on top of data nobody has verified. Instead:

• Find out what already exists. Ask finance and IT for the raw spend export, not a cleaned-up summary. The mess is the useful part, because it tells you what you are actually starting from.
• Check categorisation consistency, not just completeness. Pull a sample of 50 to 100 similar line items and see how many were classified the same way. A category field being populated is not the same as it being reliable.
• Identify who owns data quality today. In most organisations the honest answer is nobody, or several people partially. That is a diagnosis, not a personal failing, and it is the single most useful fact you will learn in your first month.
• Delay any public number. A savings figure or a spend-under-management claim announced before the baseline is verified becomes the number you get held to, correctly or not.

Pearstop builds a categorised, audit-ready spend baseline in four to six weeks for procurement leaders starting a role with none, using a classification model trained on the company's own purchasing descriptions rather than a general-purpose AI tool with no memory of the last decision it made.

Why data quality needs a single owner

The reason this gap keeps reappearing, even at companies that have cleaned their data once before, is that the cleanup was treated as a project rather than a function. An ERP migration or a one-off classification exercise fixes the data on the day it finishes. Eighteen months later, new suppliers have been onboarded, new product lines added, and nobody was accountable for keeping the category structure current, so it has decayed again.

The fix is not a better one-off project. It is a named, permanent owner for data quality, with ongoing classification treated the same way an organisation treats financial controls: as something that runs continuously, not something that gets done once and left alone. That ownership can sit with a category manager internally, or be delivered as an ongoing managed service, but it has to sit somewhere specific enough that when a new head of procurement starts in three years, they inherit a maintained baseline rather than this same problem again.

Framed the right way, this is not really about AI, or savings targets, or dashboards. It is about whether the numbers a new head of procurement presents in month three can be trusted by the people making decisions on top of them. That trust is what should be built first.

Frequently asked questions

What should a new head of procurement do in the first 30 days?

In the first 30 days, a new head of procurement should map what spend data already exists, check whether it is categorised consistently, and identify who currently owns data quality, if anyone. Building a savings plan or a technology roadmap before this diagnosis is complete means building on numbers nobody has actually verified.

Why is there no spend data baseline when I start a new procurement role?

Most organisations never assign permanent ownership of spend classification, so it decays as new suppliers, sites and product lines are added without anyone maintaining the category structure. By the time a new head of procurement arrives, the data reflects years of ad hoc fixes rather than a maintained standard, which is why no usable baseline exists.

Can AI tools like Copilot classify procurement spend data accurately?

General-purpose AI tools can produce a first pass at classification, but they are not built to hold consistency across thousands of line items or to apply a company's specific category rules. In practice they hallucinate categories, drift on similar items, and often require as much manual correction as they save, especially without human oversight built in.

How does Pearstop help a new head of procurement build a spend baseline?

Pearstop builds a categorised, audit-ready spend baseline in four to six weeks for procurement leaders starting without one, using a classification model trained on the company's own purchasing descriptions rather than generic AI. The output gives a new head of procurement a verified starting point to report from and build strategy on.

Who should own data quality in a procurement team?

Data quality needs a named, permanent owner, not a responsibility split informally across procurement, IT and finance. Without one accountable owner, categorisation standards drift the moment the person who understood the data leaves or gets reassigned, and the next leader inherits the same unreliable numbers all over again.

Neharika Kishore

Content & Visibility, Pearstop

Neha works on content and visibility at Pearstop. She writes articles on procurement data quality for facilities management, construction and infrastructure teams, and supports The Data Edge podcast. She also runs outreach to procurement and finance leaders, which keeps her writing close to the problems those teams are actually raising.

LinkedIn →