Bristol Myers Cut Procurement RFPs From Nine Months to 30 Days. The Lesson Is Not ‘Perfect Data First.’

The most useful lesson from Bristol Myers Squibb's procurement AI overhaul is not that every company should automate sourcing tomorrow. It is sharper than that: waiting for perfect data can become its own operational risk.

Supply Chain Dive reported that Bristol Myers moved its request-for-proposal process from an average of six to nine months to less than 30 days after introducing AI into procurement workflows. The company did not get there by declaring that all procurement data had to be immaculate before work could begin. Rhonda Griscti, executive director of digital strategy and global process lead at Bristol Myers, described a more pragmatic path: centralize the data, start where the signal is clear, and improve quality as the process reveals what matters. Her blunt line deserves to be printed above every enterprise AI steering committee: "If you wait for perfect data, you'll never get started."

That does not mean data quality is optional. It means logistics and procurement leaders need a better sequence.

The data-readiness trap

The broader market still has a trust problem. Supply Chain Dive cited RGP research showing that only 10% of CFOs fully trust their data quality, while more than one-third identify data trust as their top barrier to AI return on investment. Those numbers explain why many organizations freeze at the starting line.

That caution is rational. Bad data can create bad awards. A carrier scorecard that misses accessorial patterns can make a cheap bid look safe. A supplier file with stale insurance, weak service history, or incomplete compliance documents can push risk downstream into operations. A lane history that blends one-time exceptions with repeatable freight can distort pricing assumptions. In logistics procurement, dirty data does not stay in a dashboard. It becomes late freight, invoice disputes, missed customer commitments, and awkward calls with executives.

But the opposite extreme is just as dangerous. A team that spends 18 months cleaning historical freight data before changing the RFP process may still be running bids through spreadsheets, inboxes, and disconnected portals while the market moves around it. By the time the data model is pristine, the supplier network, lane mix, cost structure, and business priorities may have changed.

That is the trap: perfect-data-first sounds disciplined, but it can quietly preserve the very manual process that needs replacement.

Bristol Myers chose signal before perfection

Bristol Myers' approach is useful because it separates data location from data perfection. The company emphasized the need for a data lake: a central place where raw information can be stored, accessed, and improved. That is different from pretending every field is already clean.

For procurement, the early value came from workflow compression. The company launched its AI procurement transformation after standardizing processes and moving away from fragmented email-based work. It then scaled AI-supported sourcing, with more than $1 billion flowing through the platform in the first year, according to the Supply Chain Dive report. Bristol Myers also reported roughly 10 times more RFPs while bringing previously outsourced work back in-house with about 50% fewer resources.

Those figures matter because they show what AI can do when it is attached to a redesigned process instead of bolted onto old chaos. The goal was not simply faster document handling. It was a procurement operating model that could solicit, evaluate, and manage supplier bids at a different tempo.

That distinction matters for logistics teams. Freight procurement is full of data that is useful before it is perfect: tender acceptance, on-time pickup, detention, claims, accessorial charges, invoice variance, rejected loads, mode conversion, and lane-specific service failures. None of those data sets is flawless in most organizations. They are still better than asking a buyer to reconstruct supplier performance from emails and memory.

AI adoption is becoming a supply chain design issue

This is not an isolated procurement story. Supply Chain Brain's coverage of Gartner's 2026 Global Supply Chain Top 25 noted that leading supply chain organizations are using AI not merely to automate tasks, but to redesign how work gets done between people and machines. Gartner highlighted autonomous workforce capabilities, network-centric strategies, and end-to-end orchestration across complex ecosystems.

That is exactly the lens logistics procurement needs. The question is not, "Can AI write an RFP faster?" The better question is, "Can the organization make supplier decisions with better context, tighter governance, and less manual drag?"

For freight teams, that means the sourcing event cannot sit apart from execution. A rate is only one input. Carrier decisions should also reflect lane reliability, equipment fit, documentation performance, claims history, payment behavior, accessorial exposure, compliance status, and how a provider performs when volume spikes or weather disrupts capacity. If AI can help summarize and compare those signals, it can improve the decision. If those signals remain trapped in disconnected systems, AI becomes a shinier version of spreadsheet theater.

What logistics procurement should copy

The practical takeaway is not "move fast and ignore governance." Supplier awards still need controls, especially in regulated industries, cross-border freight, cold chain, hazmat, pharma, and high-value cargo. AI should assist evaluation, surface exceptions, and compress administrative work; it should not quietly make supplier decisions with no audit trail.

The better model has four parts.

First, centralize the raw material. Freight bids, carrier contracts, shipment histories, invoices, claims, insurance documents, compliance records, and supplier communications need a common operating layer or at least a reliable integration path. If the data is scattered, the process will stay scattered.

Second, start with high-signal use cases. Do not begin by trying to cleanse every shipment ever moved. Start with active lanes, repeat carriers, current RFPs, high-spend categories, and service-sensitive customers. Third, let use cases expose the cleaning priorities. If accessorial history is changing awards, clean accessorial fields first. If tender rejection is the service risk, normalize rejection reason codes. If cross-border documentation causes delays, focus on document completeness and exception history. Data quality work should follow operational leverage.

Fourth, keep humans accountable for supplier governance. AI can summarize bids, compare terms, identify anomalies, and flag risk. Procurement and logistics leaders still own the award logic, supplier approval, exception escalation, and performance review.

Where CXTMS fits

Bristol Myers' result should make logistics leaders uncomfortable in a productive way. If a complex life-sciences procurement organization can cut RFP cycle time from months to less than 30 days, freight teams should be asking why their own carrier bids, supplier documents, lane histories, and performance scorecards still live across inboxes and spreadsheets.

CXTMS helps logistics teams connect procurement context to freight execution. Carrier qualification, rate visibility, shipment history, exception management, invoice behavior, and performance tracking belong in one workflow so teams can make supplier decisions with usable evidence—not perfect mythology.

The lesson is simple: clean data matters, but momentum matters too. Build the data foundation, start where the signal is strong, and improve the model as real procurement work exposes what needs fixing. Schedule a CXTMS demo to see how connected transportation management can turn freight procurement data into faster, better-controlled supplier decisions.