T

Trial Terms

Negotiation prep for clinical data & R&D deals

Decision matrix · negotiation tool · responsible-AI proof of concept, mapped to FDA principles

Walk into a vendor, CRO, or offer conversation already knowing where the deal lives.

A negotiation matrix built for the clinical research world. Score what matters to you and to the other side, find the trades, plan for their anchor, and get a prep brief. One person uses it, and it is built to protect your position while still landing a deal the other side will keep. Built on Getting to YES and the Shadow Negotiation.

See it work in one click

Load a worked example

Two real clinical-research deals, pre-filled. Open one, then jump straight to the matrix and read it.

Or start from scratch · step 1, your footing

Set the table

Pick what you are negotiating, then build. Add your side and footing if you want them, or go straight to the matrix.

◆ Your footing
How strong is that alternative?

🔒 Your figures stay in this browser session. Nothing is saved or shared.

Step 2 · interests, not positions

Build the matrix

🤝
Fairness lens
When on, it flags when your asks are so one-sided the other side has nowhere to land, the overreach that can sink a deal you would otherwise win.
Strategic matrix
Every issue plotted by how much it matters to each side. Where it lands tells you what to do with it: push, trade away, fight for, or let go.
Push, your wins Core negotiation Trade away, goodwill Low stakes

Intended use: advisory negotiation prep only. Not medical, legal, or financial advice, and not a regulated device. You make every decision. See the Governance tab for the full scope and limits.

What it runs on

The method

Three sources, applied to clinical-research deals. Summarized in plain terms; read the books for the full treatment.

Getting to YES by Fisher, Ury, and Patton, 2nd edition (Harvard Negotiation Project)
Separate the people from the problem

Be warm with the vendor or hiring manager and firm on the terms. Do not let friction with a person become part of the deal.

Interests, not positions

A position is what they ask for. An interest is why. A vendor protecting recurring revenue and you protecting data ownership can still find a structure that holds.

Options for mutual gain

Generate several possible deals before you choose. Phased scope, trial periods, and volume terms are options you invent, not concessions you lose.

Objective criteria

Anchor to a fair standard: market rate, CDISC conformance, validation requirements, precedent. That turns a contest of wills into a question of what is reasonable.

The Shadow Negotiation by Kolb and Williams

Every deal runs on two tracks. One is the substance. The other is the quiet contest over power and positioning, where the other side decides how much weight to give you before they engage with your points. This track is pervasive when you are underestimated or when timeline pressure is being used against you.

Power moves

Get a reluctant party to engage: raise the cost of no deal, make the value of yes concrete, bring in a voice they respect.

Process moves

Set the stage before substance: who is in the room, the agenda, the timing, the framing. Fix a bad frame first.

Appreciative moves

Keep them engaged: acknowledge their constraints, ask real questions, hold a collaborative tone.

Workshop playbook from Areen Shahbari, MBA, Harvard Ed Portal

Responsible AI, mapped to FDA thinking

How this is governed

Built to mirror the principles in the FDA's current AI guidances. Those guidances are listed at the foot of this page; several are still draft and non-binding as of June 2026.

Scope, stated plainly

This is an operational decision-support aid for negotiation prep. It is not a medical device or software as a medical device, and it does not generate data to support a regulatory decision about a drug's safety, effectiveness, or quality. The FDA's AI guidances therefore do not legally bind it. It follows their principles by choice. Calling a tool "FDA compliant" when it sits outside that scope would itself be the failure, so this page maps to the guidances without claiming regulated status.

Context of use

It organizes your inputs into a matrix and returns advisory prep: trades, interests, an anchor plan, and framing. Intended user: a person preparing for a business or career negotiation. It does not decide, send, or advise on medicine, law, or finance, and it never touches a regulatory submission.

Model risk

On the FDA's two axes: model influence is low, since the model only advises and a full answer runs with no model at all; decision consequence is low to moderate, since outcomes are business or career, not patient safety. Net model risk is low, so the evidence here is transparency and control, not clinical validation.

Model card

Model: Claude (Anthropic), Sonnet class, called only when AI is connected. With no model connected, nothing AI runs.

Inputs: the text and scores you type. Nothing is stored or used for training.

Outputs: estimates and suggested framing, labeled as such, with stated confidence and assumptions.

Performance: none claimed. This is a proof of concept with no validation study.

Failure modes: it can be confidently wrong; counterparty estimates are guesses; it can reflect bias in the model's training data.

Known limitations and bias
  • Your inputs are self-reported and unverified. A weak alternative scored as strong will tilt the advice.
  • Estimates of the other side are the weakest data and can encode stereotypes. Pressure-test by asking, not assuming.
  • It runs on a general model, not a representative validated dataset, so it carries that model's biases.
  • Guided mode is rule-based templates, not reasoning.
Lifecycle

This is a pre-deployment proof of concept with no live performance monitoring. If it were productionized, the FDA's total-product-lifecycle approach would add version control for the model and prompts, output logging with consent, periodic review against real outcomes, a change-control plan for updates in the spirit of a Predetermined Change Control Plan, and bias checks on a representative sample.

The controls, mapped

◉ Transparency

Every AI value is labeled an estimate, with confidence, assumptions, and the model card above. Maps to the Transparency for ML-Enabled Devices guiding principles and the documentation step of the credibility framework.

⚖ Fairness to the absent party

The fairness lens flags one-sided positions, so you do not overreach into a deal that will not hold. Adapts the Good Machine Learning Practice focus on bias and representativeness to the side not in the room.

◆ You stay the decision-maker

The AI suggests, you decide, and nothing is auto-sent. Maps to the Good Machine Learning Practice human-AI team principle.

🔒 Privacy and security

Figures stay in the browser session and are never saved. If you connect a key directly in the browser, the tool warns plainly that it is exposed and holds it in memory only. The hosted versions move the key to a server secret.

⚑ No deception

The assistant refuses manipulative tactics like false scarcity or untrue claims. It supports honest, principled negotiation only.

↺ Contestability

Because a full answer runs with no AI, you are never dependent on a model you cannot see or question.

Guidances referenced, current as of June 2026: Considerations for the Use of AI to Support Regulatory Decision-Making for Drug and Biological Products (FDA draft, Jan 2025); AI-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (FDA draft, Jan 2025); Marketing Submission Recommendations for a Predetermined Change Control Plan for AI-Enabled Device Software Functions (FDA final, Dec 2024); Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles (FDA, June 2024); Good Machine Learning Practice for Medical Device Development: Guiding Principles (FDA, Health Canada, MHRA, 2021).

Why this exists

About

I spend my days in clinical data: standards, EDC and eCOA platforms, decentralized and wearable data quality, and the governance that keeps it all inspection-ready. Those projects are full of negotiations, with vendors, with CROs, over data ownership and scope, and the prep usually lives in someone's head or a messy spreadsheet.

So I built the prep into a tool. It treats a negotiation like a decision matrix, scores what each side values, and runs the same frameworks I trust, with a responsible-AI layer mapped to the principles in the FDA's current AI guidances, because that is the discipline I bring to data and to AI. It works with no AI at all, and it can run live analysis when you connect a model. The point is to show, not just claim, how I think about AI in the clinical-research world: useful, governed, and honest about its limits.

In industry the stakes are not abstract. A single EDC, CRO, or AI-vendor agreement carries data ownership, compliance exposure, and six or seven figures of contract value. Walking in with the trades already mapped and the governance terms protected can move the money and the risk that actually matter, and it does in an afternoon what usually takes a week of scattered prep. The return is better terms, less rework, and a cleaner audit position, from a tool that costs almost nothing to run.