ProcureVue

10/22/2026

If your fine-tuning comparison rests on one training run, your confidence interval is answering the wrong question.

That's the main finding from Part 2 of my Laya field report, now on Medium.

I fine-tuned a self-hosted Laya model on 11,010 human-reviewed record-linkage pairs. On a second client's real data, it reached a hosted model's precision band. Then I pre-registered a retrain that added six domain rules to the training question. At one seed, recall fell 21.9 points. At a second seed of the identical recipe, it rose 4.9. Both intervals cleared zero in opposite directions.

Both were valid intervals. They just measure label sampling at a fixed checkpoint, not variation across retraining runs. In my runs, changing only the seed moved out-of-sample recall by 9 to 18 points, which is as large as most of the training effects I had been reporting.

What holds up: fine-tuning made the local model competitive on this client's data.
What doesn't: any single-seed claim about the recipe, including the rules.
Next up are seed budgets, calibration, and routing at matched review load.

https://medium.com//i-fine-tuned-a-local-model-to-match-a-hosted-model-35a4a8a650db

Fine-tuned on the same 11,010 human-reviewed pairs my production auditors learned from, a self-hosted Laya model reached the hosted model’s…

10/15/2026

I spent the last couple of days putting a new AI decision model through its paces against the locally hosted LLM auditors we use at ProcureVue.

The initial results looked impressive: adding one domain-specific rule took recall from 15% to 89%, eventually reaching 92% at 98% precision. But digging into the results told a very different story. That dramatic improvement was driven largely by one product family dominating the test data. When I moved to a second client's actual record-linkage pipeline output, our existing local auditors proved more effective at what matters most to us: finding legitimate matches without overwhelming human reviewers.

Along the way, I uncovered some interesting lessons about domain-specific rules, fine-tuned versus zero-shot models, evaluation bias, and why precision and recall alone don't tell you which system belongs in production.

The entire experiment cost $16.16 in API charges. I wrote up the methodology, results, failures, and a few surprises in a new Medium field report.

https://pub.towardsai.net/can-a-local-auditor-ensemble-match-a-hosted-decision-model-48faac0f5111

I wrote five domain rules for a hosted decision model and tested it against my locally hosted LLM auditors. On my development data, where…

10/06/2026

Our dream team was back in action at , and the conversations are still going.

Between the kickoff event, the after-parties, and (if we're honest) very little sleep, it was a high-energy few days from start to finish. But some of the best conversations happened in exactly those moments, over a drink or waiting in line for coffee the next morning. Turns out some of the most useful conversations happen when nobody's trying to pitch anything.

The practitioners who stopped by Booth S21 were the real highlight. They came to talk through what's actually keeping them up at night: spend visibility, supplier rationalization, AI-driven insight, and a new P2P platform they're evaluating. If there was one thing we couldn't stop talking about, in sessions and hallways and everywhere in between, it was clean data. None of those goals hold up if the data underneath is a mess, and we heard some version of that from more people than we could count.

There were too many people, conversations, and moments to name, but one line from the week stuck with our team: we all have something valuable to add, so take that minute to reflect, reach out, listen, and shape the journey. It's a brilliant reminder of what can happen when you follow your passion and have the courage to back yourself.

We're already looking forward to picking up some of these conversations before next year's event, and we can't wait for the next one.

Thank you to the DPW team for pulling off such a great event, and to everyone who stopped by Booth S21, grabbed a coffee with us, or just said hello.

Until next time, Amsterdam.

10/01/2026

Data-driven decision-making has been a boardroom priority for years. In procurement, it's often still aspirational.

The reason isn't a lack of ambition; it's the data itself.

Spend is scattered across ERPs, supplier portals, invoices, and contracts, each with its own formats and blind spots. That fragmentation doesn't just slow teams down; it hides the answers leaders actually need: what's being bought, at what price, from whom, and what that means for cost, risk, and compliance. In a market this volatile, that gap stops being an inconvenience and starts being a competitive liability.

That's the conversation behind our newly refreshed white paper, Unlocking Value with Spend Analytics. Smarter look. Same trusted ProcureVue. Now updated with a sharper look to match its insights. Same business case we've always made: procurement data doesn't have to stay garbage. It can become diamonds. We've seen it turn a tangled tariff-exemption question for one manufacturer into $3.1M in verified savings, not because the product changed, but because someone could finally see the full cost picture behind it.

Read the white paper → HERE

And if there's a topic you think we should be digging into next, something you're wrestling with, or think more leaders should be talking about, write us and let us know. We're always looking for what's next through the ProcureVue lens.

10/01/2026

80% of your suppliers. 20% of your spend. Almost none of your attention.

That's the tail, and it's where most procurement teams are still leaving money on the table.

Our new white paper, Taming the Tail, breaks down why tail spend is procurement's biggest untapped opportunity and what it takes to finally manage it:

🔹 5–10% of tail spend recoverable in the first 12–18 months of active management
🔹 99% classification accuracy vs. 24% for a Big Three incumbent, on the same data set


Heading to DPW Amsterdam? Let's talk about what's hiding in your tail. Find us at Booth S21 or book time with our team: https://hubs.ly/Q04yNXPT0

📄 Download the white paper and check out our new blog post: https://hubs.ly/Q04yNX3k0

09/28/2026

⏱️Three days until Amsterdam.

One thing we love about DPW is the new surprises year after year. This year, they're leaning into mentorship across categories in the procurement tech stack, and we're glad to be one of the vendors in the spend space showing up for that. If you're navigating this space and have questions, come find us. We're happy to talk it through.

At ProcureVue, we've always thought of ourselves as more than a dashboard. We're people who want procurement teams to build real confidence in their data, not just look at it. One of our goals is to empower leaders with the information they need to advance in whatever space they dominate.

So find us at booth S21. Talk to our team about the RECODE vision, or grab one of us for a coffee to equally happily discuss spend data, or any of the in-between topics. See you in a few days!👏

09/23/2026

DPW Amsterdam's theme this year, RECODE, asks a hard question: what should you automate, what should you augment, and what has to stay human?

Our answer starts with the model itself. Our Procurement Language Model isn't a generic LLM. It's purpose-built and trained specifically on over a decade of procurement data, so the automation we trust is pattern recognition and cleansing at scale, not guesswork. What stays human is the judgment: every output runs through multi-model audit loops, and analysts review what gets flagged before it reaches a decision-maker.

That balance is what turns blind spots into leverage: it's how dozens of fragmented vendor records once became a single $20M supplier relationship, with numbers that held up under audit and board-level scrutiny.

We'll be at DPW Amsterdam, September 30 to October 1, at the RAI. Come find us, we'd love to show you how we're already living RECODE.

08/11/2026

We hope you've marked your calendars 📅, we sure know we have, as returns September 30 to October 1 at the RAI Amsterdam, and this year's theme couldn't be more on-brand for us: , Building the AI Blueprint for the Next Decade.

AI in itself is no longer the story. Ex*****on is. It's exactly why we built our business model around more than another dashboard: pairing the platform with direct consulting, so teams get the human touch needed to turn insight into faster ex*****on.

The ProcureVue team will be there at booth S21! Let's find time to connect, swap real-world examples, and maybe even walk through a live demo of how we're bringing the RECODE vision to life. See you in Amsterdam. 🚀

05/22/2026

T-Minus 🕝 5 days until

The ProcureVue team will be in New York June 3-4, and if you'll be there too, we'd love to find time to swap notes on spend, savings, and where procurement is heading next. Can't make the trip?

Our Virtual booth will be accessible in the DPW NY app, and it will have some of the same information we'll be sharing on the floor. What we do, what we've been working on, and the resources our team brings into procurement conversations.

Either way this is one of our favourite rooms of the year. Drop a comment or DM us if you'd like to connect.

05/11/2026

Notice anything different about us? 👀
The upgrade is here. The reviews are in.

The new ProcureVue site is live, the reviews are coming in, and yours is the next one we want to read. Step inside for a clear, honest walkthrough of how we help procurement and finance leaders see their full spend picture, surface savings hiding in plain sight, and make decisions backed by real evidence.

➡️https://hubs.la/Q04fKnF10