Bench to Bedside

The thesis #

This guide is for the executive standing up a captive translational engine: a unit that de-risks discoveries in-house to a defined value-inflection point, then licenses or partners them out. Data stays inside the system. IP gets licensed out. The unit is paid in upfront fees, milestones, royalties, and retained rights.

Protection and translation are not a trade-off. The moat is the very thing that gets the science to the patient.The thesis

Everyone arrives at this problem framing it as a trade-off. How much do we lock down the data and the IP, versus how fast do we move it to the bedside? Hold the lever too tight and translation dies in legal friction. Hold it too loose and you lose the moat and, worse, the patient trust that makes the moat possible. That framing is the trap.

The rest of this guide builds that argument as an operating system: a flywheel (Chapter 02), the deal that monetizes it (03), the two protection systems that power it (04–05), and the governance and people choices that keep it from spinning into a doom loop (06–08).

The edge: why a health system can win this game #

Start with the base rates, because they are the reason most therapeutics fail and the reason your edge matters.

Oncology is worse — roughly a 5% likelihood of approval from Phase 1. The $2.6B figure from Tufts is the number everyone quotes, but be honest about it: it capitalizes the cost of failures and the cost of capital, it's in 2013 dollars, and critics argue out-of-pocket cost is far lower. Use it as an order of magnitude, not a budget line.

The one number that should reorganize your strategy

Inside the same BIO dataset is the figure that justifies a health-system engine existing at all:

A health system's longitudinal records, biospecimens, and bedside clinician insight are exactly the raw material for finding the right patients and the right biomarkers. So the data moat is not a defensive asset you protect out of caution. It is the mechanism by which your assets beat the base rates. That single fact reframes every protection decision downstream: you are guarding the thing that triples your probability of success.

The translational flywheel #

A flywheel, in Jim Collins' sense, is a system where each turn makes the next turn easier and momentum compounds. The translational engine is a five-stage flywheel, and what powers it sits in the hub: the data and IP moat.

1. Clinical reach generates proprietary data and biological insight. Patients seen become de-identified outcomes, biospecimens, and a front-row view of unmet need.
2. Insight and data surface translatable discoveries. Targets, biomarkers, patient-selection hypotheses — the 3x lever from Chapter 01.
3. The engine de-risks discoveries into protected assets. Invention capture plus IND-enabling work, carried to a defined handoff point.
4. Protected assets attract partners and capital on favorable terms. The license-out: upfront, milestones, tiered royalties, and retained rights.
5. Returns and retained rights fund more discovery and improve care — back to stage 1, now with more capital, reputation, and data.

The two doom loops

Collins' flywheel has an evil twin: the doom loop, where each turn makes the next harder. There are two ways to invert this one, and the whole strategy is about staying out of both ditches.

Notice that both ditches end the same way: the research never reaches the bedside. Over-protection kills translation just as surely as a trust-destroying breach does. That symmetry is why "protect" and "translate" are one problem, not two.

Value capture & the deal #

A captive engine does not try to own the asset all the way to market. That is a different, far more capital-intensive business. Instead it de-risks to a value-inflection point — typically IND-enabling completion or a Phase 1 readout — where the asset's value steps up and a partner can shoulder the decade-long, billion-dollar late-stage burden.

The deal architecture to insist on

The cash terms are standard: upfront payment, development/regulatory/sales milestones, and tiered royalties. Structure the milestones so you are paid for the de-risking you actually performed. The part academic deals routinely leave on the table is the retained rights:

• Field-of-use carve-outs — keep rights in indications the partner will not pursue.
• Right to treat your own patients — guaranteed access for your system's population.
• Manufacturing or co-development options — for capabilities you want to build in-house.
• Reach-through governance — so a platform discovery is not quietly signed away inside a single-product deal.

The real-world template for the structure exists in public view: an academic immunotherapy discovery licensed to a large pharma partner, which supplied the manufacturing scale and global commercialization to carry it to the first FDA-approved gene therapy for cancer (2017). Bench discovery, exclusive license, partner-run late stage. That is the captive-engine deal.

IP architecture #

The mandate and the mechanics come from the Bayh-Dole Act (1980). An institution may retain title to inventions made with federal funds, but in exchange must disclose them, try to commercialize them, share royalties with inventors, and grant the government a nonexclusive license plus march-in rights.

Four design principles

1 · Make capture frictionless, not bureaucratic

The most common translation-killer is a scientist who publishes before filing. The U.S. grants a one-year grace period, but most foreign jurisdictions apply absolute novelty — a conference abstract or preprint can destroy ex-US patentability the day it posts (Finnegan on the AIA grace period). The rule is simple: file before public disclosure. Embed tech-transfer liaisons in the labs so the answer is "file Tuesday, present Thursday," never "choose one."

2 · Layer the protection

Stack the rights: composition-of-matter patent (the crown jewel) → method-of-use → formulation → manufacturing know-how held as trade secret. Then add regulatory exclusivity on top — orphan (7 years), new chemical entity (5 years), biologics (12 years).

3 · Patents for the molecule, trade secrets for the data and process

Therapeutics must be disclosed to win FDA approval, so patent them. But proprietary datasets, cell-line know-how, and models are often better held as trade secrets — protection with no disclosure and no clock. This is the under-used lever that turns your data moat into licensable IP.

4 · Keep the chain of title clean

Settle inventorship — a legal fact about who conceived the claims, distinct from ownership — early; named-inventor errors can invalidate a patent. Govern material transfer agreements so shared samples don't create reach-through claims by another institution. IP fragmented across many owners is what blocks freedom-to-operate later, when it is expensive to fix.

Data architecture & the three postures #

This is the central knob of the whole strategy: how much, and how, do external partners touch patient data?

First, the lawful tiers you are working within. Under HIPAA, data can be fully de-identified (Safe Harbor's 18 identifiers removed, or Expert Determination), shared as a Limited Data Set under a Data Use Agreement, or used identifiably under broad consent plus IRB review (the 2018 Common Rule). On top of that sits privacy-preserving technology — trusted research environments, federated queries, tokenization, differential privacy, synthetic data — which turns sharing from an on/off switch into a dial.

Three postures on the dial

Game it out

Model it as a repeated game with two counterparties: patients and partners. Fortress maximizes the trust payoff but forgoes the deal payoff. Marketplace inverts that. The decisive structural fact: the game with patients is repeated, and one breach or one steering scandal is a defection that ends every future round — losing data access drains the flywheel's fuel for good.

Now stress each posture against the scenarios that actually arrive: a breach event, a competitor's splashy data deal, a sudden regulatory tightening, a blockbuster discovery you would hate to have given away cheaply. The posture that survives the most of these is the one that preserves optionality — the Governed Gateway. It lets governance, not a one-time architectural bet, do the work of matching exposure to opportunity.

Posture is per-asset, not per-institution. One blunt institutional setting either strangles value or overexposes trust.Chapter 05 · The executive insight

Operationalize this with a data-asset register that tags every asset with a posture, a lawful basis, and a re-identification risk tier — governed by a standing committee that can move an asset between tiers as risk and opportunity shift. Run every asset through the same two-question tree:

Governance & the COI counterweight #

A health system that both treats patients and profits from therapies manufactures an institutional conflict of interest by design — royalties on investigational products, equity in sponsors, officials with financial stakes.

Structure

• Separate the commercialization entity's governance from clinical and research oversight — a subsidiary or LLC with its own board insulates clinical decisions from the commercial interest.
• Firewall the clinical decision from the commercial interest: independent IRB review and independent trial monitoring wherever the system holds a financial stake, with transparent disclosure to patients.
• Three standing committees: a Data-Use Committee (governs Chapter 05), an IP Committee (disclosure, ownership, licensing), and Joint Steering Committees per partnership.
• Explicit decision rights: name who can approve a data deal, sign a license, and kill a program.

The perception that patients are steered toward the system's own products is the single fastest trigger of the over-share doom loop. Govern as if a journalist will read the minutes.

Steering the people without stifling them #

The governing principle for people is the same one used for data: bright lines and fast lanes. Bright lines are the few things that are never okay — anything that breaks trust. Fast lanes are pre-cleared paths for everything else, so the default answer is "yes, here's how," not "no." Three constituencies, three failure modes to steer around.

1 · BU leadership & strategy — own the flywheel, not every shiny asset

Their failure modes are chasing every discovery and over-indexing on near-term licensing revenue, which starves stage 5. Steer with a portfolio thesis, hard stage-gates with kill criteria, and a metric that rewards flywheel velocity (assets reaching the inflection) over raw deal count.

2 · Scientists & clinician-investigators — a fast lane, not a gauntlet

They are driven by discovery, publication, patients, and autonomy. Heavy IP and legal process kills curiosity and speed, and they will route around it. Steer by making IP capture frictionless (embedded liaisons, fast disclosure, "publish and patent — file before you present"), aligning incentives (inventor royalty share, recognition), and teaching the one-year-bar discipline without policing academic freedom.

3 · Cross-functional (legal / compliance / BD / commercial) — enable velocity within bounds

Their failure mode is becoming the "department of no" that brakes the flywheel. Reframe the job as clearing the path: standard deal templates, pre-cleared data-access SOPs, tiered playbooks so routine cases never need a bespoke negotiation. They own the bright lines; they should automate the fast lanes.

Gates, metrics & the board's decision checklist #

Stage gates — kill early, kill cheap

It is a portfolio, and most assets should die. That is not failure; it is the design.

Discovery → Target validation → IND-enabling → Phase 1 → LICENSE
            └ each gate: explicit go/kill criteria + a named decision-owner ┘

Metrics that match the thesis

• Flywheel velocity — assets advanced per year to the inflection point.
• Biomarker-attach rate — % of programs with a selection biomarker. This is the 3x lever; make it an explicit, resourced target.
• Disclosure-to-filing time — a proxy for fast-lane health.
• Retained-rights value — not just upfront cash per deal.
• Trust ledger — breaches, COI incidents, patient-consent rates. Leading indicators of doom-loop risk.

The board's seven-question checklist

1. Are we hunting platform/enabling IP, or only single-product assets?
2. Is our biomarker-attach rate an explicit, resourced target?
3. Do we have a data-asset register with per-asset postures — or one blunt institutional posture?
4. Is IP capture a fast lane (file-before-present) or a gauntlet scientists route around?
5. Is the commercialization entity's governance separated from clinical and research oversight?
6. Could a journalist read our COI minutes and conclude patients are not steered toward our own products?
7. Do our metrics reward flywheel velocity and retained rights, or vanity deal count and upfront cash?

If the answers trend "no," the unit is drifting toward a doom loop regardless of how good the science is. Keep the flywheel turning; stay out of both ditches; protect the moat because it is how the bench reaches the bedside.

The two funnels #

Everything up to here was strategy. From here down is how you actually run the thing, because this is a new business line and the org has no muscle memory for it yet. Start with the single most important operational fact: a captive engine runs two funnels at once, joined by the development pipeline in the middle.

The deal funnel & funnel-math #

Out-licensing is B2B enterprise sales. The "customer" is a pharma or biotech partner; the "product" is a de-risked asset plus its data dossier; the sales cycle runs 6–18 months. Run it with the same rigor a SaaS company runs new logos: defined stages, exit criteria per stage, a CRM, and backward funnel-math.

The seven stages

Funnel-math — work backward from the goal

You can't manage what you don't quantify. Pick the annual deal target, apply stage conversion rates, and the top-of-funnel volume you need falls out. Illustrative shape (calibrate to your own data — these are planning heuristics, not benchmarks):

Stage-gate project management #

The development pipeline is a project-management system. Each asset is a project; the stage-gate (phase-gate) method governs how it advances. Borrowed from new-product development, adapted to therapeutics.

Gates and decisions

Each stage has defined entry deliverables, an owner, a budget envelope, and a timeline. Each gate has explicit criteria, a named gatekeeper (or gate committee), and a cadence. The discipline is that a project cannot skip a gate, and a gate review must be able to say "kill" without it being a career event for anyone.

The per-asset project kit

RACI for a translational project

Ambiguous ownership is the quiet killer of a new unit. A simple RACI per decision class removes it:

Portfolio scoring & kill discipline #

You cannot fund everything, and the temptation in a new unit is to spread thin to look busy. Resist it with an explicit scoring model that ranks assets and a culture that makes killing normal.

The scoring rubric

Score each asset 1–5 on each dimension, weight, and rank. Weights are yours to set — they encode the unit's strategy — but a defensible starting set:

Portfolio balance: diversify across risk and time horizon, but concentrate around the clinical areas where your data moat is deepest. A focused portfolio in two or three therapeutic areas beats a scattered one — focus is what compounds the flywheel's data advantage.

Standing it up: the first 18 months #

A new business line needs a launch sequence, not a big-bang. Build credibility with early wins while the machinery comes online.

The founding team — hire in this order

• BU head — translational + commercial fluency; carries the thesis and the quality bar.
• BD lead — runs the demand funnel; comes with a partner rolodex.
• IP / tech-transfer lead — runs capture; the file-before-disclose discipline lives here.
• Scientific / translational lead — runs the supply funnel and triage.
• Portfolio PM — runs the gates and the cadence; the operating backbone.
• Then: alliance manager, data steward, finance partner, regulatory advisor (often fractional at first).

Operating cadence & tooling #

The meeting cadence (the heartbeat)

The tooling stack

What worked: three cancer centers #

The strategy above is not theory — three premier cancer centers have run it, and between them they form a natural experiment in the three postures. St. Jude shares by default but governs in tiers; MD Anderson monetizes governed access without letting data leave; Memorial Sloan Kettering is the field's most successful IP monetizer and the source of its sharpest cautionary tale. Each row below is posture-tagged — Fortress, Governed Gateway, Marketplace — and cited.

MD Anderson — build the moat in-house, sell governed access

Note: ORIEN / Total Cancer Care / M2Gen belong to Moffitt and Ohio State, not MD Anderson — a common misattribution. MD Anderson's analogs are APOLLO, GENIE, and the Institute for Data Science in Oncology.

Memorial Sloan Kettering — the IP machine, and the bright line it taught the field

St. Jude — share by default, govern in tiers

Bright lines & posture defaults #

Here are the two decisions the data chapter left open, now answered from the precedent in Chapter 15. Treat these as a starting policy you adjust to your institution, not a finished rulebook — but every line traces to something a peer learned, often the hard way.

Decision 1 · Your bright lines — the "never" list

Decision 2 · Default posture per asset class

Run each asset through the Chapter 05 decision tree, but start from these evidence-based defaults:

The TDIE model #

If Chapter 15 asked what worked, this chapter is the how — the single clearest working template for the IP-capture function your engine needs. St. Jude's Technology Development & Industrial Engagement (TDIE) office is worth copying almost line for line, because one office owns the entire stack — patenting, licensing, material-transfer agreements, and industry engagement — and is the sole party authorized to bind the institution. No fragmentation between legal, contracts, and business development, which is exactly the IP-fragmentation failure Chapter 04 warns about.

The signal is in the rename and the hire. The old Office of Technology Licensing became TDIE, and in early 2024 St. Jude created its first-ever SVP of Technology Commercialization and filled it with Lisa Jordan — a life-science VC and entrepreneur (Wharton MBA, prior Executive-in-Residence at a university commercialization center), not a career licensing attorney. Her brief: recruit seasoned entrepreneurs and build an accelerator to de-risk assets toward regulatory approval. The move is from "license what walks in the door" to "build and de-risk companies" — the captive-engine thesis, staffed.

The disclosure-to-commercialization workflow

How the office is built

Two structural choices make it work. First, one mandate, one signatory: TDIE alone negotiates and authorizes every technology-transfer agreement and every material transfer into or out of the institution. Second, a two-layer leadership stack: a VC/operator SVP on top to drive company-building and de-risking, sitting over career licensing counsel (JD/MBA) who run patent prosecution and deal mechanics — and the SVP reports near the very top (to a deputy director), not buried three levels down. Keep the lawyers for the deals; add the operator for the companies.

The disclosure-timing gate — the core control

This is the mechanism to copy first. At intake, TDIE asks two questions: (1) the funding source (any federal money triggers Bayh-Dole — elect title, government keeps a non-exclusive license); and (2) whether the invention has already been publicly disclosed, because prior disclosure can destroy patentability. The rule is file before the scientist publishes. And patenting is selective and instrumental — in TDIE's own framing, "patents are the glue that brings us together" with an industry partner, so where eligibility law makes a patent futile (U.S. diagnostics, software) they skip it or file abroad first. The patent is a means to a partnership, not an end. This is how an open-science institution that shares genomic data by default (St. Jude Cloud) still captures the IP that matters: openness is the default; patenting is the deliberate exception on translatable assets, governed at the disclosure gate.

Two value levers most offices miss

What the office has shipped

The copyable blueprint