"A negotiated discount is a promise. A realized saving is a smaller number in somebody's budget. Most companies celebrate the first and never audit the second."
8 min read · August 2026
Good Morning, Good Evening, and Good Night — wherever you're reading this. Last month we followed a molecule. This month we follow something harder to see: the savings number. The one announced in a quarterly review, celebrated on a slide, and then — with remarkable frequency — never found again. Almost every company can tell you what it negotiated. Far fewer can point to the line in the financials where that money actually landed. This issue is about the distance between those two facts, and why closing it has very little to do with negotiating harder.
Start with two sets of findings that cannot both be comfortable at once.
Deloitte's 2025 Global Chief Procurement Officer Survey, drawn from more than 260 CPOs, found cost reduction ranked the number one priority — cited by roughly 72% — with operational efficiency close behind at 68%. And the scorecards look good: around 88% of surveyed organizations reported meeting their cost-savings targets.
Now the other set. Research from World Commerce & Contracting has for years put the cost of poor contract management at roughly 9% of annual revenue, with more recent work suggesting something closer to 11% of contract value evaporates once a deal leaves the negotiating room and enters delivery. For a billion-dollar business, 9% isn't a rounding error. In plenty of industries it's the entire margin, twice.
Hold both up together and you get the central puzzle of corporate cost management: nearly everyone reports hitting their savings target, and roughly a tenth of the money leaks out anyway. Those statements are only compatible if "savings" and "money" have quietly become two different things. That's exactly what has happened. Procurement is measured on an event — a signature. Finance is measured on a balance — what's left at period end. Nobody owns the road in between.
A saving isn't a moment. It's a relay race, and the baton gets dropped at predictable handoffs.
1. Negotiated. A better price exists in a document. Most scorecards begin here and, tellingly, end here too.
2. Implemented. The price is loaded into the ERP, the catalog, the rate card, and the supplier's own billing system. Miss any one of those four and the discount lives only in the PDF.
3. Realized. Invoices actually arrive at the lower price, for what was actually consumed — verified against paid invoice data, not against the contract.
4. Validated. Finance agrees. It shows up as a smaller budget or a lower unit cost, signed off by someone whose bonus depends on the financials rather than on the savings report.
Estimates of where the money dies cluster in an uncomfortable band: roughly a third of a savings pipeline is lost between identification and planning, and something like another fifth disappears in execution — meaning close to half of projected savings never reach the P&L. Organizations that formally track all four stages tend to land far higher, in the 75–85% range. The difference isn't negotiating skill. It's bookkeeping discipline.
And note the incentive this creates. The further down the funnel you measure, the worse your numbers look — so the honest teams report smaller savings than the sloppy ones. Measure at Stage 1 and you're a hero. Measure at Stage 4 and you're explaining yourself.
Every savings claim is the same equation: (baseline price − new price) × volume. Three variables, and all three are arguable.
Take the baseline. Last price paid? List price? The supplier's opening quote? The budgeted rate? A market index? Each is defensible, and each produces a wildly different figure from the identical negotiation. Anchor to the opening quote and you can manufacture savings out of a price increase — "they asked for 15%, we settled at 8%, that's 7% saved" — while next month's invoice is unambiguously larger than last month's. Whoever controls the baseline controls the story.
Then volume. Drop the unit price 12% and let consumption rise 20%, and total spend goes up while the savings report stays green. Both true. Only one is cash.
Then the distinction that swallows the most credibility: cost avoidance versus cost reduction. Avoidance is the value of an increase you prevented. It's real work — and it can never appear in the P&L, because it's measured against a hypothetical. There is no line item for a bill you didn't receive. Report avoidance and reduction in one total and you've built a number no CFO can act on, because half of it is spendable and half is counterfactual.
"If no one's budget got smaller, you didn't save money — you bought permission to spend it somewhere else."
— The only savings test that survives contact with a CFONegotiation gets the attention because it's the dramatic part. The leak is downstream, and it's mundane.
Rate-card drift. Contracted rates go stale. Annual escalators fire automatically while the savings baseline stays frozen at year-one pricing. Labor categories creep upward — work scoped for an analyst gets billed at the senior architect rate, one of the most common and least detected patterns in professional services billing. Nobody breached anything; the rate card simply drifted away from the deal.
Off-contract buying. The discount only applies to spend that flows through the agreement. Every purchase routed around it silently invalidates a savings claim — rarely malicious, usually just faster.
Unclaimed entitlements. Volume rebate tiers nobody tracks and therefore nobody invoices for. Service-level credits earned and never claimed. Savings you already negotiated and already earned, left on the table because claiming them required someone to be watching.
Auto-renewal. The most expensive clause in most portfolios, because it converts inattention directly into cost.
Underneath all of it sits a structural problem: services spend cannot be three-way matched. For goods the control is clean — purchase order, goods receipt, invoice; three documents, and a discrepancy is arithmetic. For services there is no receipt. There's a purchase order, a timesheet, and a degree of trust. Which is why the same company that catches a $400 overcharge on a materials invoice will pay a six-figure services invoice on the strength of a familiar name in the "from" field. The control that protects the physical supply chain doesn't exist for the half of spend that has grown fastest.
The cash line nobody scores: payment terms are the purest version of this disconnect. Moving from 30 to 60 days changes no unit price and produces no P&L savings — it's free cash flow, which is what treasury and the CFO care most about. Because it doesn't show up on a savings scorecard, it gets traded away at the table for a discount that then leaks. The company gives up cash it would have kept, for savings it never realizes — and the trade is invisible because the two things are scored by different departments in different units.
The market wants to tell you AI will negotiate better deals. I don't think that's where the value is, and the reason is instructive.
The savings gap isn't an intelligence problem — it's a systems-of-record problem. Commercial terms live in a PDF in a contract repository. Prices live in an ERP. Consumption lives in invoices and timesheets. The budget lives in an FP&A model. The rebate tiers live in an appendix nobody has opened since signature. Five systems that don't speak, plus one document format that was never machine-readable. The money isn't lost because someone negotiated badly. It's lost because no single system can answer "are we being billed what we agreed to?"
That shape — unstructured documents on one side, transactional data on the other, and a reconciliation too tedious to do continuously by hand — is exactly what language models are good at. The valuable applications are unglamorous: pull obligations, rate cards, escalators, tiers, and credit triggers out of executed contracts into structured data; then test every invoice line against that structure, every cycle, and flag drift the month it starts instead of at renewal. One caution, though, and it's the same one I'd give any analytics program: automating a bad baseline produces confident nonsense at higher volume. AI won't clarify a fuzzy savings definition — it will scale it.
First, who owns the savings scorecard. I think there's a real chance the CFO takes it, and savings claims start requiring a finance countersignature the way capital requests do — one shared record where procurement's claim and finance's budget line are literally the same row. That change is organizational, not technological, and it would do more than any tool.
Second, the shift from measuring events to measuring balances. Reporting realized savings against paid invoice data is a harder, smaller, less flattering number. The teams that volunteer for it will look worse for two quarters and be trusted for a decade.
Third, and genuinely unresolved: what happens to services pricing when the work is done by software. The entire outsourcing and staff-augmentation model is priced in units of human time — rate cards, FTEs, billable hours, labor categories. As vendors deliver more of that work with agents, "a senior developer hour" stops describing anything real. Outcome-based pricing becomes inevitable, and it opens a brand-new leakage surface, because verifying an outcome is far harder than verifying a timesheet — and the timesheet was already the weak control.
The through-line is the same throughout. Cost management fails at the handoffs, not at the negotiation. The savings that never show up were never stolen. They fell into the gap between two systems and two scorecards, and nobody was standing there to catch them.
"Procurement is scored on the discount. Finance is scored on the outcome. Until those two live on the same record, the savings will keep dying in the gap between them."
— Daivik Suresh, August 2026-DAIVIK SURESH-
Supply Chain + Business Analytics Enthusiast · August 2026Not financial advice. All opinions are personal and do not represent any employer. Investing involves risk including potential loss of principal.