Methodology guide · 04
Including the awkward ones. If a question you have is not answered here, ask it and we will answer it directly rather than sending you a brochure.
Two sources build every discount curve, with a third opening up as customers opt in. Third party market data covering transacted software pricing at scale, our own data from advising on these negotiations, and, on an opt in basis only, anonymised customer contributed records. The sources page sets each out in full.
We do not disclose it. Not now, not under NDA, and not as part of a vendor security review.
We would rather give you that answer at the start than have you build a diligence step around a name that is never coming. What we will do instead is show you the methodology in full: how the curve is constructed, what it was calibrated against, where it is measured and where it is extrapolated, and what confidence we attach to each band. If that is not sufficient for your process, tell us early and we will work out what evidence would satisfy it.
No, and we are deliberate about saying so. A benchmark is a calibrated model of how a vendor prices, built from what our sources and our own negotiation work have observed. It is not a file of other people's signed agreements.
That is a design choice rather than a limitation. A benchmark assembled from real contracts is either too small to say anything about a deal your size, or large enough that somebody's confidentiality has paid for it. Modelling the curve lets us cover every deal size in your estate without trading on anyone's paper, and lets us re-cut a vendor the week its pricing model changes rather than waiting years for enough fresh contracts to accumulate.
Customers upload contracts to us for their own analysis, and those documents stay inside their own tenant. They are not pooled, not shared, and not visible to other customers. Nothing in the benchmark set is drawn from another customer's document unless that customer has explicitly opted in to contribute, and even then only as an anonymised aggregate with at least five contributing organisations behind it.
A research firm tells you what the market looks like. A data vendor sells you transaction records. We sit on the buyer's side of the table and have run the negotiations, which shows up in two places: the levers we model are the ones that actually move each specific vendor, and we tend to see a pricing regime change while it is happening rather than after it settles into aggregate data.
1,140 vendors, across 1,341 live benchmarks.
Because one vendor is often several different negotiations. An Oracle ERP Cloud renewal, an unlimited licence agreement and a Java subscription have different levers, different discount ceilings and different people on the vendor side. Collapsing them into a single Oracle number would be worse than useless.
Oracle carries 30 separate benchmarks on that basis. SAP and Salesforce carry 18 each, Microsoft 14, Cisco, IBM and ServiceNow 11 each. The other 1,122 vendors carry one each, which is the right number for a vendor you buy one thing from.
Ask and we will tell you straight away. If a vendor is not covered we will say so rather than produce a generic number, and we can tell you what it would take to build the curve.
Coverage runs from 2019 to 2026 and is weighted to recent years, because a 2021 discount tells you very little about a 2026 renewal. The large majority of our datasets carry a July 2026 calibration date, and every benchmark shows its own as of date on the face of the output rather than in a footnote.
When a vendor changes its commercial model we re-cut the curve rather than waiting for a scheduled refresh. Recent examples include SAP's 2026 licensing regime, where on premise discounting effectively stopped, and the post 2025 Microsoft enterprise agreement changes.
Three regions, North America at roughly 42% of coverage, Europe at 33%, and APAC and rest of world at 25%. Twelve industries, from banking and insurance through to public sector and logistics. Four deal size bands, from under $250k of annual value to above $5M. Your benchmark is cut to your band and your region before anything is compared.
It varies, and we tell you where, which is the part that matters. Every band the platform shows carries one of three labels. Measured means a real cohort of comparable deals sits behind it and we print how many. Curated means an analyst calibrated the band on a known date and we say so without attaching an invented count. Directional means neither, and you should treat it as a starting hypothesis.
Attaching a sample size to a band that does not have one is treated internally as a defect rather than a presentation choice.
Then you get no percentile. If fewer than ten comparable deals sit in your cohort we do not produce a placement, not even a wide one. A benchmark built on four deals is a rumour with a decimal point, and shipping it would be the fastest way to lose your trust in every other number we give you.
Three places, and we would rather tell you than have you find out. Above roughly $5M of annual value, deal counts thin for every vendor and the curve leans more on tier structure than on dense observation. For vendors that quote everything privately and publish no reference point, the curve rests on a narrower base. And immediately after a vendor changes its commercial model there is a window before the new behaviour is well observed.
Amounts are converted to a common currency before anything is ranked, so a euro deal is never compared against dollar deals at face value. Multi year deals are normalised to an annual value so a three year commitment does not look like three times the deal. Scope is handled by benchmarking at product and programme level rather than at vendor level, which is why one vendor can carry thirty benchmarks.
No, because there is no list of named deals to show. What we can show you, in full, is the construction: the curve for your vendor, the band structure, the cohort your deal was placed in, the size of that cohort, the confidence label, and the calibration date. If you want to interrogate a specific number, that is exactly the conversation to have and we will walk through it line by line.
Tell us, with your deal. Two things happen. You get a direct answer on why the model placed you where it did, and if you are right, the curve gets corrected. Buyers challenging our numbers with real deals is one of the ways the model improves, so it is welcome rather than awkward.
The benchmark itself is not a guarantee. It is the position the evidence supports you arguing from, not a price the vendor has agreed to. What you land depends on your leverage, your timing and how the conversation goes. We do offer engagement models where our commercial outcome is tied to your result, which is a separate conversation from the data.
No. Contracts you upload are processed for your use only, inside your own tenant. They do not enter the benchmark set and are not visible to any other customer.
Records are stripped of entity identifiers and enter a pool that only ever emits aggregates with at least five contributing organisations behind them. Below that floor the aggregate does not exist, not even as a count, so no output can be narrowed back to a single contributor, a counterparty or a specific deal.
The switch is off by default for every organisation, and it is reversible.
Your documents are not pooled into the benchmark set and are not shared with another customer. The specific commitments on processing and model training are set out in our data processing terms, which we will send with this guide rather than making you ask for them.
Access is scoped to your organisation, and staff access to customer material is restricted and logged. Views and downloads of contract files are recorded in an audit log you can inspect. If your security team wants the full control set, ask for the security review package.
Yes. You can benchmark a deal by entering its commercial terms directly, without a document ever leaving your side. Uploading gets you more, because the analysis can read the actual clauses, but it is a choice rather than a condition.
No. We do not take money from the software vendors we benchmark, we do not resell their licences, and we hold no reseller margin on anything you buy. Our revenue comes from the buyers we work for, which is the only arrangement under which a benchmark can mean anything.
No. Vendors have no access to the platform, to your position, or to anything you enter. What you choose to tell your vendor in a negotiation is entirely your decision.
Frequently, yes, which is generally the point. Vendors know their own discount curves better than anyone. A benchmark's job is to close the information gap between the two sides of the table, not to surprise anyone.
You get three things: where you sit against comparable buyers, the price that would put you in the top quartile of that same group, and what the gap between the two is worth per year. The third one gets you the meeting. The second one is what you ask for.
We use the top quartile rather than the best deal ever recorded because it is defensible. It is demonstrably achievable by buyers like you, which is precisely the argument that survives contact with a vendor.
They will, and the responses are predictable enough to prepare for.
| What the vendor says | What the method lets you say |
|---|---|
| "Every deal is different." | Agreed, which is why this compares you only against your own deal size band and region, not against the whole market. |
| "Those discounts are not real." | Then show us where your curve differs. The benchmark states its confidence and its date, so it is a discussion about evidence rather than assertion. |
| "That price required commitments you have not made." | Correct, and structure is modelled explicitly. Tell us which commitment and we will price it, which turns the objection into a negotiable item. |
| "Your data is out of date." | Every benchmark carries its calibration date on the face of it. If the market moved after that date, say when and how. |
Earlier than feels necessary. The benchmark tells you the size of the gap, but leverage comes from time, and time is the one input no model can give you back. Six to nine months before a material renewal is comfortable. Six weeks is a rescue.
Reply to whoever sent you this guide. Questions about the method get answered by someone who built it, not by a support queue.
One closing point. Nothing in this guide is written to be reassuring. It is written so that when a number in the platform looks surprising, you already know how it was built, what it rests on, and how hard to lean on it. That is the only basis on which a benchmark is worth anything.