The asymmetry runs against the buyer
A manufacturer negotiating a machined part, a casting, or a molded component negotiates blind.
The supplier on the other side of the table quotes hundreds of buyers and knows exactly where the market sits. The buyer knows one number: what they paid last time. Every subsequent number in the conversation is anchored to that one.
Raw materials have public indices. The converted part — the bracket, the casting, the molded housing — has none. Spend analytics reports what you paid, in more detail every year. It cannot report what the market is paying, because your own records do not contain the market.
That gap is the only thing this bureau exists to close, and it closes it in one direction: toward the side of the market that cannot see.
What the bureau publishes
- A band, never a point
- P25–P75 with P10 and P90 whiskers. No single price, and no mean — a mean leaks dominance and invites false precision.
- Trend
- Trailing eight quarters of the band, shifted by the six-month lag.
- Contributor count, as a floor
- Shown as "3+", never as an exact number above the minimum. Exact counts and their period-to-period deltas can reveal who entered or left a cell.
- A confidence tier
- Every cell carries a tier derived from how much of the published normalization spec its underlying observations satisfied.
- Your position against the band
- Your own normalized price beside the market band, per covered category. If you sit above P75, the bureau says so plainly.
- Coverage
- The share of your addressable spend for which a benchmark clears the floor — and what is locked, with the honest reason and what it is waiting on.
- Labeled public reference bands
- Where they exist: quartiles over public buyer records, such as U.S. government award prices and municipal purchase-order lines. Drawn as their own series with their own label and explicit n, shown at a minimum of 5 distinct negotiations, never blended into a contributor band and never counted toward the floor.
What the bureau refuses to publish
Benchmark data flows only to the side of the market suffering the information asymmetry. Suppliers cannot buy access, cannot subscribe, and cannot see a member's position. This halves the theoretical market and was decided knowingly; it is published in our methodology and written into the member agreement.
Below the contributor floor the answer is "insufficient data" — said without apology, and not waived for a large member or an urgent negotiation. The floor rises as the pool grows, and tightens further where a cell's contributors compete with one another.
Ranges, quartiles and trend only. Never a row, never a point, never a mean, and no export granular enough to rebuild an observation.
No supplier is identified in any price view. A benchmark is a market range, not a target list.
Nothing under 6 months old is published. There is no premium tier that buys it sooner and there never will be — the lag is structural, which is also why there are no price alerts.
No contributor is identifiable in any output. There is no participant list, no member directory, and no member-to-member channel of any kind. We do not confirm anyone's membership, including to other members.
Cells whose change in composition between periods would reveal a contributor are suppressed. Queries are rate-limited and logged: a member fishing to isolate one company's cells is detectable, and must be.
No forecasts, no recommended prices, no target prices, no "should-cost" verdicts. The bureau reports what the market did, lagged and banded, and stops there.
Published bands are compared against public price series to catch drift in thin cells; a flag sends a cell to human review before publication, and never edits it. An adjusted number is an unexplainable number.
Lines whose unit of measure will not resolve are rejected rather than guessed. Tooling and surcharge lines, related-party pricing and blanket-release duplicates are detected and handled before anything reaches a band.
The floor, the dominance cap, the lag and the window are fixed and versioned. Changes are published and apply forward only, never retroactively to make a past number look better.
How it works
Contribute your purchase record
Purchase-order lines reach the bureau through a direct connection to your ERP. No upload, no spreadsheet, no implementation project. Contribution is automatic once the connection and the data-rights agreement are in place.
Lines are normalized before they are compared
An ERP unit price is not a market price. Units of measure are resolved and currency converted at the purchase-order-date rate; tooling amortized into a piece price and separately-listed surcharges are detected; related-party pricing is excluded; and blanket-purchase-order releases are deduplicated to negotiations, so one agreement cannot enter a cell as many observations. Parts are matched on keys ranked by how hard they are to fake — a national distributor's catalog number outranks a manufacturer part number typed into an ERP field. What will not resolve is rejected. This is the hard part of the product, and the specification for it is published to members.
The structural rules decide what publishes
A cell publishes only when, inside the lagged window, it clears the contributor floor and no single contributor exceeds 40% of its negotiations. There is no query path that can override this and no per-member exception.
You see the cells that cleared, and what the rest are waiting on
Covered categories show the band, your position against it, and the trend. Locked cells state the reason — below the floor, or too concentrated — and how many more contributors would open them. Coverage is the number that matters, and the bureau reports it against your whole addressable spend rather than only the part it can answer.
| Rule | Value | Why it is set this way |
|---|---|---|
| Contributor floor | Set with counsel before launch | Below it, a cell is withheld. Rises as the pool grows and tightens where a cell's contributors compete. Published in the methodology, versioned, never waived. |
| Dominance cap | 40% | No contributor may account for more than this share of a cell's negotiations, so no member reads back their own prices. |
| Lag | 6 months | Nothing fresher is published, to any member, at any price. Lagged data cannot function as a live price signal. |
| Window | 24 months | The trailing period whose negotiations feed a cell. |
| Cell | category × material × volume band × period | Cells are keyed by what is bought, not by industry, so the pool runs horizontally across unrelated buyers. |
| Output | quartiles only | P25–P75 with P10/P90 whiskers and a median. No means, no rows, no exact contributor counts above the floor. |
| Public reference data | minimum 5 negotiations | Public buyer records appear only as separately labeled reference series with their own n, and never count toward the floor. |
| Methodology | v0 | Published to members and open to member audit. Versioned; changes are prospective only. |
Membership
Manufacturers and other volume buyers of direct materials and manufactured parts, whose ERP already holds the purchase record.
Contributing members
Contribute automatically through the connection and receive benchmarks in the categories they contribute to. Contributing is structurally the better arrangement, and is meant to stay that way.
Non-contributing subscribers
A buyer who cannot contribute yet can subscribe at a premium. The premium exists to make contribution the rational choice, not to punish caution.
Members are not competitors
A furniture maker and a pump maker both buy machined brackets and compete in no output market. Cells are keyed by category, so the pool is horizontal across industries; where contributors in a cell do compete, the protections tighten.
Write with the categories you buy and the next negotiation on your calendar. A useful first conversation is about one category you know cold — either the band confirms what you suspected, or it surprises you and you can check it. There is no form, no demo funnel, and no free tier.
The bureau does not buy, sell, broker, or quote parts, and takes no compensation contingent on any price outcome. It is operated by Didero and built so that contributors on any ERP can join. We will not confirm your membership to anyone.