Polestar Solutions

Field Notes

What grounded AI means when the number goes into a negotiation

An AI that invents a benchmark is worse than no AI at all, because you will repeat its number to a vendor. What grounding actually means: citation tags, groundedness checks, refusing thin data, and how to audit an AI answer yourself.

Key points

  • It answers, fluently and confidently, that comparable companies pay about 18 percent less.
  • 2 Is the cohort actually comparable? A percentile against deals of a different size, region, or edition is a number wearing a costume.

Grounding is a pipeline, not a promise

On ISVCOSELL, "grounded" is a specific, checkable property with three mechanical parts.

Every figure carries a citation tag. When ISVCOSELL answers, each number in the answer is tagged back to its source: the clause in your stored contract, the line on your invoice, or the benchmark cohort it was computed from. The citation is not a footnote for decoration. It is a link you can click, and the source it opens is the evidence you would hand a skeptic.

Answers pass a groundedness check before you see them. After the answer is drafted, a separate check verifies that the claims in it are actually supported by the cited sources. An answer that asserts more than its evidence supports gets caught at this gate, not on your vendor call.

Thin data produces a refusal, not a guess. If your question runs past the edge of the data, a vendor with a small cohort, a metric we do not track, a contract you never uploaded, the honest answer is "the data does not support an answer to this," and that is the answer you get. A procurement AI earns trust faster by declining well than by answering everything.

app.isvcosell.com/ask

Every figure tagged to a clause, an invoice line, or a cohort. Click the tag, see the evidence.

THE SAME JOB, TWICE

TODAY, BY HAND

You ask a general purpose chatbot whether your price is fair, and it answers fluently with a discount figure it cannot source.

To check it, the analyst digs through old analyst reports, a peer's conference anecdote, and a two year old survey PDF.

None of the sources agree, so the number that goes on the vendor call is the one that sounded most confident.

The rep asks where it comes from, the honest answer is nowhere solid, and the vendor learns your numbers can be dismissed for the rest of the deal.

Hours of source-chasing, and the credibility risk is the real cost

WITH ISVCOSELL

Ask ISVCOSELL the pricing question in the Ask screen, scoped to your own contracts and the benchmark cohorts.

Read the answer with a citation tag on every figure: the clause, the invoice line, or the cohort each number came from.

Click through a citation before the call, so the evidence you would hand a skeptic is already open in front of you.

Where the data is thin, take the refusal at face value: no number leaves the building that the sources do not support.

Minutes to an answer you can defend across the table

What changes: hours of source-chasing become minutes, but the arithmetic that matters is credibility. One unsourced number read back to you by a rep can cost the rest of the negotiation; a cited answer drawn from modelled deal cohorts holds when the rep asks where it came from, which is the only test that matters.

"A procurement AI earns trust faster by declining well than by answering everything."

PART TWO

The weekly licensing brief

Want to be updated when major licensing and pricing changes land? One analyst brief a week: the price rises, metric changes and audit campaigns that move software costs. Work email only.

Get the brief

The bar rises when the answer becomes a document

A chat answer is read once. A report is forwarded, quoted in meetings, and pasted into board papers, so its errors travel further and live longer. That is why every report the platform generates, and there are 25+ types, from benchmark narratives to negotiation briefs to portfolio reports, is held to the same standard as the chat: every figure in the narrative is grounded in stored data and citation tagged, the whole document passes the groundedness check before it renders, and the visual standard matches what a human analyst desk would ship.

The discipline shapes the writing too. The reports interpret the data rather than restating it: where the leverage is, where the risk is, what a strong outcome looks like, and what to do first. But interpretation only stands on numbers that check out, which is the whole point of doing grounding first.

app.isvcosell.com/example-report

Report-grade output: the same grounding rules, applied to the document your CFO will forward.

PART THREE

Four checks before any AI number goes to a vendor

These apply to any AI tool, ours included. If a number is about to leave the building, run it through this list.

1 Can you click through to the source? A real citation opens the clause, the invoice line, or the cohort definition. "Based on industry data" is not a citation, it is an alibi.

2 Is the cohort actually comparable? A percentile against deals of a different size, region, or edition is a number wearing a costume. Check what the comparison set is before you lean on where you sit in it.

3 How fresh is the evidence? Enterprise pricing has moved fast for three years. A number from an old market is not conservative, it is wrong in an unknown direction.

4 Does the tool ever say "I do not know"? If you have never seen it refuse, you have no way to calibrate its confidence. An AI that answers everything is telling you something, and it is not good news.

One honest limit to close on. Grounding makes the numbers trustworthy, it does not make the decisions for you. Whether to push, when to accept, and what your relationship with the vendor can bear are judgment calls, and they stay yours. The platform's job is narrower and, we think, more valuable: when you finally say a number out loud, it holds.

FF

About the author

Fredrik Filipsson, Cofounder, ISVCOSELL

Fredrik has spent more than twenty years in enterprise software, with time at Oracle, IBM, SAP, and Salesforce before moving to the buy side. He structured and priced the kind of large agreements most buyers only see once or twice in a career, which taught him where the leverage sits and how far a vendor will actually move. He started ISVCOSELL to hand that knowledge to every sourcing team.

More posts by Fredrik Connect on LinkedIn →

See it in the product

How benchmarking works → Browse the use cases → Every feature → Calculate your time saved →

FREE TRIAL · FULL PLATFORM · NO CARD REQUIRED

Ask a question and click the citations.

The free trial opens the benchmarking database, 1,483 vendors deep, plus the negotiation guides, playbooks, and talking points for your own renewals. No card needed, a corporate email is all it takes.

Start your free trial → Or decode a contract free, no account

Free for 30 days, no card needed. Your data stays isolated at the database, and you can export or delete it any time.

Watch it in action

The ISVCOSELL Deal-Room Hand ISVCOSELL the negotiation The email that writes itself

Browse the full demo library →

THE ISVCOSELL AI BRIEF · WEEKLY

The week in enterprise software buying, in one email.

What shipped on the platform, and the pricing and licensing moves worth knowing before your next renewal. One email a week, to your work address. Unsubscribe any time.

Subscribe

More in Field Notes

1,483 vendors, one method: how the benchmark library is built

A benchmark is only as good as the deals behind it and the honesty of how it is compared. How the library is built from modelled deal cohorts, normalized, placed in the right peer cohort, and graded by confidence.

Read

300 vendors, 52 weeks, one team: the renewal calendar problem

The average enterprise runs 300+ software vendors and every one of them renews. Why notice windows are where budgets quietly die, and how a renewal desk with AI agents turns the calendar from a threat into leverage.

Read

A calmer desk, and Main Apps where the work starts

The platform now wears the desktop look: warm paper, one interactive colour, and Main Apps folded into Home so your instruments live where you start.

Read

A live analyst in your ear: inside the call copilot

The vendor call is where prepared positions meet improvisation, and the rep does this every day. The live call copilot runs a whisper rail beside the conversation: live transcript, grounded prompts, and the exact fact you need at the moment the claim is made.

Read

Adobe ETLA vs VIP: seat reclaim, right-profiling, and the walk away

An Adobe ETLA renewal is decided before you discuss price, by how many seats sit idle and how many are over-profiled. How to reclaim the waste, right-profile the rest, and build the VIP walk away Adobe respects.

Read

Agent to agent: how the Agent Negotiation Protocol works

When a buyer's AI agent negotiates with a vendor's AI agent, someone has to keep the record straight. How the open Agent Negotiation Protocol handles identity, mandate, and a ledger neither side can rewrite.

Read

Want help putting this into practice?

Contact us to discuss your project.

Get in Touch