Native Alpha is the advantage you already have and have not proven

Native Alpha is an assessment. We find the edge already sitting in your customer access, your data, or your track record, test whether it holds up, and end with a decision instead of a deck.

Native Alpha is a hypothesis, not a clever idea, a patent, expertise, important contacts, or access to AI. Those things may supply clues, but evidence must show that the advantage matters to somebody else. The test is short. An advantage counts as Native Alpha when it changes a real decision, and a competitor with money can't just buy it. A tool anyone can license doesn't count. A six-month head start doesn't count. Neither does a good story about a market. What counts is the customer who picks up when you call, the data you have a legal right to use, or a track record nobody else holds.

The idea grew out of founder Shahid Shah's writing about Native Alpha and years spent inside regulated businesses. The alpha is native because it comes from the work itself. Nobody here invents hospital discharge software from a whiteboard. The useful ideas come from having been in the building: a health system's IT shop, a device maker's regulatory review, a trial operations team, or an insurer's records department, and noticing what nobody had fixed. Shahid keeps working through the idea in his Native Alpha series.

AI sharpens the question. The easy ask is whether a tool makes today's job faster or lets a department run with fewer people. The better ask is which work becomes possible that nobody could staff before: reading every denial a plan issues, or keeping a device submission's evidence current while the product is still in the field. An old job done faster is a nice feature. A job nobody could do before is worth proving, filing, and building around.

Try the assessment. It turns what you already know into a prompt you can paste into whatever AI tool you use. There is a worksheet to download, and eight scenarios from regulated healthcare: medical devices, SaMD, healthcare IT, trials, and payer operations. They show how the test holds up. Nothing you type is stored or sent anywhere.

One thought, tested from several directions

The units are not stages in a pipeline. IP tests what may be true. Press tests whether the argument can be understood and used. Capital tests whether money should move and whether a better-positioned founder already exists. Studios runs a small experiment when the market has not produced the right builder. Network finds the people needed throughout. What each encounter teaches returns to the hypothesis.

The units are not a pipeline. Each encounter with reality changes the thesis.

Every pass ends with a choice: Continue · Change · Back someone else · Build it ourselves · Stop.

Founder's writing

Native Alpha in Shahid's own words

Shahid first wrote down the argument, then kept testing it against how companies choose what to build, buy, fund, sell, and stop.

Visit the full Native Alpha series

2026-09-10
Some Companies Are Already on Life Support. AI Just Hasn’t Pulled the Plug Yet.
Revenue is not the same thing as being alive. A test for founders of mature software companies: if you started this company today, knowing what AI makes possible, would you build anything resembling what you now operate?
2026-08-30
Native Alpha: Start With the Advantage You Already Have, Not the Opportunity You Just Noticed
Alpha, in investing, is the return you cannot explain by the market. Native Alpha is the business version: the unusual advantage already sitting inside your company, your customers, your data, and your relationships, and the discipline of letting it decide what you build, buy, fund, sell, and stop.
2026-08-29
Alpha, Benchmarks, and Why I Borrowed the Word From Investors
Alpha in finance is the return you cannot explain by the market. Here is what that actually means, why investors bother separating skill from luck, and which parts of the idea carry over to running a company and which do not.
2026-08-28
Capability Is Cheap Now. Advantage Is Not.
AI made building things dramatically cheaper, which quietly destroyed a lot of differentiation that was never real. How to tell commodity execution from differentiated execution, and where to spend the judgment you have left.
2026-08-27
Someone Outside the Room Has to Care
Enthusiasm inside a company proves nothing. The strength of evidence for an advantage is measured by what somebody outside gave up to act on it, from a compliment at the bottom to a renewal at the top.
2026-08-26
Product-Market Fit Does Not Tell You Who to Sell To
The same product can be ordinary for one buyer and transformative for another. Product-asymmetry fit and the Ideal Alpha Profile explain why the customer with the biggest problem is often the wrong customer.
2026-08-25
Where Native Alpha Actually Lives, and How It Decays
Advantage is contextual, it hides at surprisingly small scales, it needs protection in the place it actually lives, and it expires. How to find it, where to put the fence, and when to admit it is gone.
2026-08-24
First Principles Got Turned Into a Slogan
Reasoning from first principles means stripping a problem down to what you can verify and rebuilding from there. In executive hands it usually means ignoring people with experience. This is the real version, and how it fits with Native Alpha.

Try it yourself

Find My Native Alpha™

Tell us about a problem you've lived with. The page builds a prompt you can paste into whatever AI tool you already use. Nothing you type leaves your browser, and nothing is kept after you close it.

Organization or teamWho does the work, and in what market or setting?Worked problemDescribe the workaround, delay, or failure that keeps coming back.Customer accessWhich customers, buyers, or decision-makers can you reach that others can't?Unusual knowledgeWhat has doing the work over and over taught you that the manuals miss?Data and rightsWhat data, contracts, approvals, licenses, or IP rights do you legally hold?Workflow positionWhere do you sit in the flow of work, information, or decisions?Relationships and trustWhich relationships get you access, evidence, or speed others don't have?Operating capabilityWhat can this team do better, faster, or with less risk, over and over?Economics and compoundingWhat gets cheaper, faster, or harder to copy each time you do it?

Your prompt

Native Alpha analysis prompt

0/9

You can skip questions. The prompt says which ones you left blank instead of guessing.

Act as a skeptical Native Alpha analyst. Native Alpha is an unusual and disproportionate advantage already present in a person, team, organization, workflow, position, asset, or relationship. It must change a real decision and must not be something a capable competitor can simply buy, copy, or recreate.

CONTEXT
Organization or team: Not provided
Worked problem: Not provided
Customer access: Not provided
Unusual knowledge: Not provided
Data and rights: Not provided
Workflow position: Not provided
Relationships and trust: Not provided
Operating capability: Not provided
Economics and compounding: Not provided

ANALYSIS RULES
1. Separate every important statement into Fact, Inference, Assumption, Estimate, or Hypothesis.
2. Never invent evidence, customers, rights, performance, market size, regulatory status, or economics. Write “not established” when support is absent.
3. Treat access to common AI models and tools as a commodity, not Native Alpha.
4. Look for advantage in customer access, unusual knowledge, relationships, lawful proprietary data, rights, workflow position, trust, reputation, distribution, operating capability, economics, IP, and unusual combinations of assets.
5. Argue against each hypothesis. Name the evidence that would disconfirm it.
6. In regulated work, reserve clinical, regulatory, quality, security, legal, and release judgments for qualified people.

REQUIRED OUTPUT
A. Restate the worked problem in one plain sentence.
B. Produce up to five ranked Native Alpha hypotheses. For each, state: classification; why it may be unusual; who should care; supporting evidence; missing evidence; disconfirming evidence; and confidence (low, medium, or high).
C. Distinguish durable advantage from a temporary lead, common capability, or attractive story.
D. For the top three hypotheses, design the smallest credible real-world experiment, including setting, owner, measure, decision date, and stop condition.
E. State how repeated use could compound the advantage through evidence, rights, workflow, relationships, distribution, or reusable artifacts.
F. End with a decision table: continue, change, back someone else, build it ourselves, or stop. Within that choice, say whether the best form is research, publication, rights, licensing, partnership, investment, an experiment, software, or a company. Give the reason and the next evidence required.

Be concise, plain spoken, and willing to conclude that no Native Alpha has yet been shown.

Review the prompt before sharing sensitive or regulated information.

Download the Discovery WorksheetFive-page PDF · evidence, experiments, and decisions

What you get back

What a good answer looks like

You don't get a brainstorm. You get a written record that shows where the evidence is thin, what's still a guess, and when to stop.

01
Ranked hypotheses
Up to five claims, best first, each tied to the evidence behind it.
02
Evidence labels
Facts stay separate from inferences, assumptions, estimates, and guesses.
03
The case against
What's missing, what would disprove the claim, and why it might fail anyway.
04
Smallest experiment
A real setting, something you can measure, an owner, a decision date, and a reason to stop.
05
Compounding path
What each round of use adds: evidence, rights, workflow, distribution, or work you can reuse.
06
Decision
Continue, Change, Back someone else, Build it ourselves, Stop, with the reason written down.

Apply the method

Native Alpha scenarios in medical devices, SaMD, and healthcare IT

These are research scenarios. Nobody is claiming the advantage is proven yet. Each one starts from evidence that regulated work already produces, and each one says what would weaken or kill the idea.

Scenario 01

Connected medical devices · post-market surveillance

Continuous evidence for connected medical devices

Operating problem: Complaint, service, telemetry, cybersecurity, and adverse-event records often live in separate systems, so a manufacturer rebuilds the safety picture from scratch for each review.
Possible Native Alpha: The manufacturer already has legal access to its installed-device telemetry and its quality team’s complaint decisions. Together they could keep a living evidence record competitors cannot recreate from public data.
Evidence already present: Device logs, complaint files, service history, risk controls, vulnerability records, CAPA decisions, and prior submissions.
Smallest credible experiment: Trace one known field issue from telemetry through complaint assessment, risk management, corrective action, and the current regulatory record.
Disconfirming evidence: The records cannot be joined reliably, the telemetry lacks the context a decision needs, or qualified reviewers do not trust the resulting trace.
Stop condition: Stop if the pilot cannot reproduce a previously approved safety decision without someone rebuilding it by hand.
Next decision: Build the evidence pipeline, narrow its intended use, or stop.

Medical-grade device research Proof in the register

Scenario 02

SaMD · medical device software development

Software as a Medical Device traceability

Operating problem: Software requirements, hazards, code changes, verification results, clinical evidence, cybersecurity controls, and labeling drift apart as a SaMD product changes.
Possible Native Alpha: A team that already controls its engineering history and regulatory rationale may be able to maintain requirement-to-evidence traceability continuously rather than reconstruct it before a submission.
Evidence already present: Design inputs, software requirements, risk files, source-control history, test results, clinical-performance records, SBOM, threat models, and labeling.
Smallest credible experiment: Select one released software function and reconstruct the full trace from intended use through hazards, implementation, tests, clinical support, and labeling.
Disconfirming evidence: Generated links omit important controls, produce false relationships, or add more qualified-review time than they remove.
Stop condition: Stop if the trace cannot pass a blinded review by the people responsible for software quality and regulatory approval.
Next decision: Extend to change control, constrain the automation, or stop.

Impossible Work research Healthcare use cases

Scenario 03

Clinical decision support · digital health

Clinical decision support boundary review

Operating problem: A health software team can spend months building a clinical decision support feature before deciding whether its intended use and operation make it a regulated device function.
Possible Native Alpha: Direct access to clinicians, product behavior, and the intended workflow may let a team test regulatory boundaries and user comprehension before committing to a product architecture.
Evidence already present: Intended-use language, user roles, input data, output explanations, workflow observations, clinician interviews, risk analysis, and current FDA guidance.
Smallest credible experiment: Run three realistic cases with intended users and test whether they can independently review the basis of the recommendation within the actual workflow.
Disconfirming evidence: Users cannot understand the basis, rely on the output without independent review, or the intended use still requires a regulated claim.
Stop condition: Stop the non-device path when the workflow evidence contradicts the proposed regulatory boundary.
Next decision: Revise the claim, pursue a device pathway, or abandon the feature.

Research areas Submit a worked problem

Scenario 04

Healthcare IT systems · EHR and FHIR interoperability

Healthcare interoperability exception handling

Operating problem: Interfaces can pass conformance tests while clinical messages still fail, arrive late, lose meaning, or require people to repair queues by hand.
Possible Native Alpha: A health system’s accumulated interface exceptions, repair decisions, and workflow consequences may reveal repeatable failure classes unavailable in specifications alone.
Evidence already present: Interface-engine logs, FHIR and HL7 messages, error queues, help-desk records, data-quality reports, downtime logs, and downstream workflow outcomes.
Smallest credible experiment: Classify one month of production exceptions and test whether the resulting rules identify preventable failures before they reach a clinical queue.
Disconfirming evidence: Exceptions are too local to generalize, source data is insufficient, or prospective detection does not reduce repair work or clinical delay.
Stop condition: Stop if the rules fail prospectively or merely rename work without removing it.
Next decision: Turn the exception model into a product, keep it local, publish it, or stop.

Use-case register Market potential

Scenario 05

Medical device cybersecurity · quality systems

Medical-device cybersecurity evidence

Operating problem: Threat models, SBOM changes, vulnerability findings, compensating controls, verification, and submission claims are often reviewed as separate documents.
Possible Native Alpha: A manufacturer’s installed-base knowledge and product-specific control history may support a living cybersecurity evidence chain that improves both response and regulatory review.
Evidence already present: SBOM versions, architecture, threat models, vulnerability feeds, exploitability assessments, patches, verification results, field deployment, and risk decisions.
Smallest credible experiment: Take one known component vulnerability and trace affected products, exploitability reasoning, controls, tests, deployment, and residual-risk approval.
Disconfirming evidence: The inventory is incomplete, product context cannot be resolved, or the evidence chain produces unsafe false confidence.
Stop condition: Stop automation where authoritative product identity or qualified security judgment is absent.
Next decision: Build continuous evidence, repair the inventory first, or stop.

Medical-device research Related filings

Scenario 06

Clinical trial operations · regulated research

Protocol drift in clinical trials

Operating problem: Protocol amendments, site instructions, eligibility logic, consent language, data-capture rules, and monitoring plans can diverge across systems and sites.
Possible Native Alpha: A research organization’s protocol history and site-level deviation record may reveal where amendments predict operational failure before the next site repeats it.
Evidence already present: Protocol versions, amendments, consent forms, EDC configuration, monitoring findings, deviations, queries, training records, and site correspondence.
Smallest credible experiment: Trace one amendment across every affected artifact and compare predicted failure points with subsequent deviations and data queries.
Disconfirming evidence: The historical record cannot establish sequence, predicted issues do not occur, or intervention fails to reduce deviations.
Stop condition: Stop if prospective use cannot improve a measurable site or data-quality outcome.
Next decision: Deploy as trial assurance, narrow to one artifact chain, publish, or stop.

Clinical-trial use cases Family records

Scenario 07

Health plans · prior authorization

Prior-authorization denial intelligence

Operating problem: Health plans and providers process each authorization or denial while rarely learning systematically from the full history of reasons, evidence requests, appeals, and reversals.
Possible Native Alpha: A lawful longitudinal record of decisions and reversals may reveal preventable evidence gaps or policy inconsistencies that no public benchmark contains.
Evidence already present: Authorization requests, clinical attachments, policy versions, reason codes, reviewer notes, decisions, appeals, reversals, and elapsed time.
Smallest credible experiment: Choose one high-volume service and test whether historical patterns predict avoidable denials or missing evidence before submission.
Disconfirming evidence: The apparent patterns disappear after policy and case-mix controls, or prospective guidance does not change outcomes.
Stop condition: Stop if the model cannot distinguish documentation defects from legitimate medical-necessity decisions.
Next decision: Build a prospective evidence check, revise the policy workflow, or stop.

Market-potential register Bring a worked problem

Scenario 08

Medical devices · adverse events and complaints

Post-market signal review at full scale

Operating problem: Qualified reviewers cannot manually compare every new complaint with the entire complaint history, service record, recall history, and current risk file.
Possible Native Alpha: A manufacturer’s own history of adjudicated complaints may support a review system grounded in decisions and product context a general model cannot see.
Evidence already present: Complaint narratives, reportability decisions, investigations, service and manufacturing records, adverse-event reports, recalls, risk files, and reviewer corrections.
Smallest credible experiment: Replay a blinded historical set and test whether the system surfaces known clusters without hiding rare serious events or changing the qualified reviewer’s authority.
Disconfirming evidence: Serious cases are missed, false clusters consume more review time, or performance fails across product families and reporting periods.
Stop condition: Stop if recall falls below the approved safety threshold or reviewers cannot inspect why a case was surfaced.
Next decision: Use as reviewer triage, narrow the product scope, or stop.

AI-enabled work research Regulatory execution research

Primary references

FDA: Software as a Medical Device (SaMD) (opens externally)FDA: Clinical Decision Support Software guidance (opens externally)FDA: Cybersecurity in Medical Devices guidance (opens externally)

Synthetic worked example

A result that can still say no

This one is made up. It's the shape a real analysis takes, not a finding about any actual company or product.

Hypothesis: A device maker’s adjudicated complaint history may support earlier detection of field-safety clusters.
Classification: Hypothesis. The record exists, but nobody has shown it predicts better than the alternatives.
Supporting fact: The synthetic company holds product-linked complaints and approved reportability decisions.
Missing evidence: No prospective test shows earlier detection without masking rare serious events.
Smallest experiment: Replay a blinded historical set, then run prospectively beside qualified reviewers for one product family.
Stop condition: Stop if serious-case recall misses the approved threshold or review time increases.
Decision: Prove before build. Do not claim the operating record is Native Alpha until it changes a safety decision.
Compounding path: Reviewer corrections may improve the evidence record; common AI tooling alone does not compound.

The stages run in order because each one is expensive to skip. Without Prove we would be buying opinions. Without Decide we would be filling a library nobody opens. Without Compound we would start over every January.

Stage 1

Find

What is somebody working around every day?

Good inventions rarely announce themselves. They show up as a workaround: the nurse who keeps a paper list because the discharge screen is too slow, the records clerk with a spreadsheet that catches the access requests the system misses, the trial coordinator who keeps a paper binder because the study system will not show the current protocol version. We log the workaround, who does it, and how often, before anybody argues about what it means.

Write down the workaround before writing down the theory.
Attach the actual artifact: the spreadsheet, the log file, the denial letter.
Say who does the workaround and how many hours a week it costs them.

Stage 2

Prove

What would show this does not work?

Before money moves, we write down what would kill the idea. A prior-authorization idea gets tested against a payer's real denial history. A device idea gets shown to the regulatory people who would have to file it, early, when the answer is still cheap. A state modernization idea gets checked against what procurement can actually buy. Ideas that fail get written up too, because knowing in six weeks beats knowing in two years.

Name the kill condition in writing before spending.
Test against production data or a live operating environment.
Ask the person who has to sign the FDA submission or the state contract, up front.

Stage 3

Decide

Who acts on it, and by when?

An idea that survives goes to one person who can do something about it: file the application, publish the disclosure, size a position, charter a company, or stop. That person writes the date and the reason. Without a named owner and a date, a good finding goes into a deck, circulates for a quarter, and quietly dies of committee. We have watched it happen elsewhere and we do not allow it here.

One named owner carries the decision.
A date on every decision, including the decision to wait.
The reason written down while the evidence is still fresh.

Stage 4

Compound

What does the next one inherit?

Claim language drafted for a hospital interoperability filing gets reused for the trials version. A test harness built for device telemetry gets pointed at metric queues. The people who have lived through a large system replacement know exactly which month an integration will slip and why. All of that accumulates, which is why the fourth filing in a family costs a fraction of the first.

Every artifact goes back into the shared record.
Reuse claim language, test harnesses, and diligence checklists across markets.
Track how long a cycle takes, and what it produced.

What we turn down

We pass on work we cannot defend to a skeptical customer: savings estimates that assume a hospital changes its staffing model, device software that nobody has run past a regulatory reviewer, insurance analytics that only work on a cleaned extract, and proposals that require a procurement path the buyer does not have. We also pass on secrecy for its own sake. Trade secrets stay closed where secrecy is genuinely the asset; everything else gets published.

A method only counts when it changes what actually gets filed, funded, published, or built. If a stage never once stopped us from spending money, it was decoration.

Work with us

Take the next step

Challenge an analysis Design an experiment Submit a worked problem Request an evaluation

See what the method has produced in the register