AI implementation for fundamental investment teams

Your research process. Engineered by a Forward‑Deployed Investor.

A Forward-Deployed Investor is a buy-side investor who embeds with your team, finds where the hours go, and decides what is worth building. NSigma engineers build and maintain the AI workflows around how you invest.

Book 45 minutes See what we build

Maintenance, mapped.

More judgment per analyst-hour.

↓

The implementation gap

Licenses are not capability.

Almost every fund now pays for AI. Very few have changed how research gets done.

Use generative AI at work

95%

Have an AI usage policy

Just over half

Have trained staff to use it

About 1 in 4

Have built it into the research process

The gap

Source: AIMA, 2025 survey of fund managers on generative AI use. The last step is not measured in the survey. It is the work we do.

The question every allocator now asks

“What are you doing about AI?”

The answer that worries them

“We pay for three AI subscriptions.”

The answer that reassures them

“We built AI into the process you invested in, and every output is on the record.”

Allocators backed your process. Generic tools pull it toward everyone else's. That is how style drift starts.

Where the hours go

Upkeep fills the week. Judgment gets what is left.

The same filing can be either kind of work. Refreshing the model from it is maintenance. Reading it to test a new idea is judgment. The work is to separate the two steps inside each task, then automate the first.

34hmaintenance

20hjudgment

0hAn analyst's week in earnings season54h

Work that should shrink Maintenance

Extracting disclosed KPIs
Reconciling them to the model
Re-keying into templates
Building earnings prep packs

Work that should grow Judgment

Interpreting what changed
The questions nobody had time for
One more name covered
Earlier warning on the thesis

Illustrative baseline from Thesis & Method's mock diagnostic. Not a client measurement and not a performance claim. The diagnostic measures your own split.

Two parts, one service

An investor defines the question. Engineering makes it work.

The Forward-Deployed Investor

An investor who learned to build, not an engineer who learned finance.

Every engagement is led by a CFA charterholder with 17 years of buy-side equity research, who has led AI implementation for a global equity team. Your FDI diagnoses your process, recommends what to build, and proves whether it works.

NSigma engineering

The engineering arm of the service.

About 30 developers, PhD-level quant and machine learning support, and ISO 27001-certified information security management. NSigma engineers build, test and maintain the agreed workflows.

Your team

Keeps every investment decision.

Analysts approve every proposed change. The PM owns the thesis, the sizing and the decision. No tool sizes positions, enters orders or publishes.

Meet the team

Show the work

A research record. A clear build order. A better question.

Three outputs, recreated with sample data so you can see their shape. In an engagement they are built on your names, your models and your note standard.

Thesis trackerIllustrative, sample data

Broadcom (AVGO)

Core long · Reviewed Oct 2026 · On track

Custom accelerators for three hyperscalers become a second growth engine next to networking.

What has to be true

Assumption	Status
Accelerator ramp at three hyperscalers	On track
Networking share holds against rival Ethernet	Watch
Software margins hold after price changes	On track
Gross margin holds as custom mix rises	Watch

FY27 AI revenue, $bn: house vs Street

House 74 vs Street 71. The gap narrowed from 7% to 4%. That is the conversation for the next review.

Sell triggers

A top-three customer takes accelerator design in-house
AI revenue guide cut two quarters in a row

Sample record. Not a view on the company.

Diagnostic readoutIllustrative, sample data

Where the week goes

63%of analyst hours are maintenance

Holdings with no thesis update this year	9
Prints last season with no note at all	6

One print, step by step

Retyping figures40m

Transcript70m

Drafting60m

170 minutes from release to note. Sent 21:54, with no tie-out to the model.

Build order

First: Thesis tracker and post-print digest
Then: KPI history per name
Later: Public-data collection
Not yet: Card-data dataset (vendor rights not confirmed)

Before any build: data rights cleared, AI policy agreed, a named reviewer.

AI Bubble MonitorIllustrative, sample data

Separate signal from story.

Thematic observatory · 48 signals

41.2CAPE

+73%Data center build

23.8%US AI adoption

$2.40H100 spot, hr

0.87%IG credit spread

27Hot, dangerous 6Hot, healthy 10Weak, deteriorating 5Calm, fine

Question for the analyst

Is installed capacity turning into demand at the rate the thesis requires?

Sample grid. Not current market observations.

Built for how you invest

Your style determines what we build first.

Prepare faster. Analyze deeper. Research further. The diagnostic decides where the next hour of engineering is worth spending.

Style	Where the hours go	Typical starting points
Earnings-driven long/short	Post-print notes, KPI tie-outs, sell-side triage, crowding checks	Post-print digest with proposed model updates, thesis and estimate record, crowding check
Long-only quality and compounders	Watchlist upkeep, moat durability, capital allocation, management meeting prep	Thesis record with history, signpost monitors per holding, moat and pricing-power investigations
Value and contrarian	Filing re-reads, earnings normalization, asset valuation	Cited filing search, normalized earnings and asset-value models, value-trap investigations
Event-driven and activist	Deal and proxy review, ownership, regulatory diligence, catalyst timing	Evidence room, ownership and voting maps, catalyst and regulatory trackers, scenario workbooks
Thematic	Signposts across many companies, value chains, company exposure	Thematic monitors, exposure maps, theme-to-earnings scenarios
Small and mid cap	Fragmented data, management and ownership, funding needs	Company evidence room and starter model, public-data collection

Multi-manager pods, global and emerging markets, quantamental and income strategies are covered in the diagnostic. The first build follows your style, not a template.

See examples of what we build

The engagement

Start with the process. Commit at each decision gate.

The diagnostic stands alone. A build is a separate decision, and so is everything after it.

Phase	What happens	Your team's time	Price and decision gate
01Diagnostic	Three weeks, 2 to 6 days on site by team size. A baseline, a working sample, a build order and a priced recommendation.	About one day for the PM, half a day per analyst, plus a one-hour readout.	From $25,000. Approve a build, change the scope, or stop with the report.
02Build	NSigma engineers build and test on your past quarters, with your FDI specifying and checking.	A named reviewer, about 2 hours a week.	Agreed cap. Your reviewer accepts the output before use.
03Training	4 workshops and 2 clinics within 30 days. 15 live hours, at most 6 analysts per cohort.	Attendance at the workshops and clinics.	Priced in the proposal. Your team shows it can use and check the workflow.
04Managed	Maintenance, improvements and FDI time. 12-month initial term from agreed go-live.	Priorities and exceptions on an agreed cadence.	From $5,000 a month. Optional NSage: $5,000 to $10,000 a month by coverage.

A typical first year for a two-to-three analyst team costs less than one analyst.

Read the scope, schedule and terms

For the COO and data lead

Your environment. Agreed controls.

Fund-hosted, managed or hybrid, chosen with your IT and your licensed-data permissions.
Named access only, separate client environments, ISO 27001-certified information security management.
Your positions, theses, models and notes never enter a shared component or another client's work.
A personal trading restriction for everyone with access to your data.

Read security and data

What we build · Configured to your process

Better questions. Working systems.

Choose the point where your research loses momentum. See how an engagement turns it into a repeatable workflow, and which tools it draws on.

Get to the debate while it still matters.

A filing should enter your research process with its definitions, discrepancies and implications intact.

InputRelease, filing and house model

WorkflowExtract, reconcile, flag

ReviewAnalyst-approved update

Workflow schematic, configured during the engagement.

What we would build

A repeatable extraction and reconciliation pipeline, a queue of exceptions, and a draft research update linked to the source passages.

How we prove it

Tested on your own past quarters, using only data available at the time, before it goes live.

The deliverable

A reviewable update in the format your team already uses.

A sample from the library

Nine examples. Nearly 40 in the library.

Every workflow answers one of three limits on a research team: time, skill or reach. These are examples. The diagnostic decides which workflows fit your process, and most engagements use some that are not shown here.

The limit is time

Prepare faster

Recurring work an analyst could do, in a fraction of the hours.

Post-print digestRelease, filing and call in; KPI tie-out, model lines and a house note out.
Morning event briefOvernight filings and news on each holding, ranked by thesis condition.
Coverage primerA cited first-look memo and starter model in your template for a new name.

Without AI, who could do it?Your analyst, given enough time.

The limit is skill

Analyze deeper

Quant-grade tools, in the language the PM already uses.

Factor exposure and attributionReturns split into factor and stock-specific, in plain language, ready for the LP letter.
Scenario P&L and driver bridgesEach decision with a range, built from the drivers that move it.
Thematic observatoriesSignal grids for the themes that carry the book, with exposed holdings ranked.

Without AI, who could do it?A trained quant team.

The limit is reach

Research further

Questions no team could answer by reading alone.

Read-through alertsEach customer, supplier and competitor of a holding, flagged when it reports.
Contradiction monitorEvery new document tested against your written thesis conditions.
Red team and pre-mortemAn independent, cited counter-case before a position changes.

Without AI, who could do it?Nobody.

Connected by design

Every workflow should leave the firm smarter.

Every workflow reads from and writes to a small set of shared records, so each one makes the next cheaper to build. All of it runs in your environment.

Thesis and estimate ledger

Theses, triggers and dated estimates, joined to outcomes. What you believed, what you expected, and which evidence changed the view.

Evidence bank

Cited search across years of your own notes, models and transcripts. Source passages, definitions, dates and unresolved contradictions.

Relationship graph

Customers, suppliers and competitors of every name, refreshed each filing cycle. The economic links through which a development could matter.

Case · All three limits in one question

Could cooling content protect Vertiv's growth if data-center builds slip?

The chain: a grid interconnect delay pushes a data-center go-live, fewer megawatts are installed this year, and the timing of Vertiv's cooling revenue moves.

Megawatts installedWatch: hyperscaler capex, utility interconnection queues

Liquid-cooled shareWatch: rack power density, GPU platform mix

Content per MWWatch: Vertiv and peer disclosures, order mix

Vertiv shareWatch: competitor wins, pricing

Cooling revenuePlus retrofit and service

Base case100 MW, 50% liquid, 1.00 content, 1.00 share

100

Builds slip 20%80 MW, everything else unchanged

Slip, content rises60% liquid, 1.10 content per MW

106

Slip, content rises, share fallsShare drops to 0.90

Illustrative inputs, indexed to base case = 100. The dashed line marks the base case. The diagnostic replaces them with sourced figures. A proposed investigation, not a completed finding or a forecast.

Builds can slip 20% and cooling revenue still grow if liquid share and content per MW rise. The debate is which moves faster.

Scope follows the question

Engagement patterns, not a shelf of software.

We select and configure the work against your data rights, investment style and review requirements. Specialized forensic, event-driven and quant tools are kept for funds whose mandate calls for them, and the diagnostic decides.

The engagement · From diagnosis to daily use

Begin with the desk. Earn the right to build.

The engagement starts with what your team is trying to learn. Each phase ends with a concrete output and a decision about what happens next. Nothing proceeds without your approval.

01Diagnostic · three weeks

Find the constraint. Price the remedy.

Your Forward-Deployed Investor sits with the team and follows the work from the source document to the investment conversation. Interviews are checked against actual work samples and observed handoffs.

Week one, on site

Day one with the PM or CIO, including the team meeting
Half a day with each analyst
At funds with 6 or more analysts, one day with operations, compliance and the data or vendor lead
Playback with the PM on the last day

Weeks two and three

Analysis of where the hours go
A working sample, built on one of your holdings
Build order and acceptance tests

End of week three

A one-hour readout with the PM
A priced recommendation: build, prepare first, train, or stop

On-site days by team size

PM + 2 analysts2 days

PM + 4 analysts3 days

PM + 6 analysts5 days

PM + 8 analysts6 days

PM or CIO dayTwo analysts a dayOperations and compliance day

Your team's time: about one day for the PM, half a day per analyst, one day for operations and compliance at larger funds, plus a one-hour readout.

What you receive

Where your time goes

Hours by task, maintenance vs research, and print-to-note time.

Your research backlog

The questions that could change a position, ranked by the PM.

A build order

First, then, later and not yet, each with the hours it gives back.

A working sample

The first build, run on one of your own holdings.

A recommendation and price

Build, prepare first, train, or stop. With costs and acceptance tests.

Standalone engagement. Use the report with us, with another provider, or not at all. From $25,000, fixed by team size.

Decision gate. Approve a build, change the scope, or keep the diagnostic and stop.

Before any build

Four things are confirmed with you first. The diagnostic checks each one, so none is a surprise.

Data

Which sources you are licensed to process with AI.

Policy

An AI usage policy. If you don't have one, we help write it.

Owner

One person on your team who reviews outputs, about 2 hours a week.

Where it runs

Your environment or ours, agreed with your IT.

02Build · capped hours

Prove it on the work you already know.

NSigma engineers build to capped hours agreed before work starts, with your FDI specifying and checking the work. Each workflow is tested on your past quarters, using only data that was available at the time.

Acceptance targets agreed per build before implementation
Example targets: 70 to 85% of proposed model lines accepted on a held-out quarter, and a draft within 60 minutes of the print
Unused hours are not billed; overruns within unchanged scope are absorbed by us
Weekly reporting of hours and forecast cost
Working code, documented source and calculation logic
Acceptance results, known limitations and a release plan

Decision gate. Your reviewer accepts the output before the workflow goes into use.

03Training · with the people using it

Your analysts leave able to build, not just use.

workshops, three hours each

clinics of 90 minutes, within 30 days

analysts at most, in one cohort

live hours in total

Build the data room and model for a new name
Turn a company question into an evidence plan
Build KPI dashboards and memos from their own data
Backtest an idea using only data available at the time
Spot when an output is wrong, and say why
Use Claude Code and Codex day to day on their own names

Decision gate. The team shows it can use and check the workflow.

04Managed · ongoing

Research changes. The system has to follow.

New disclosures, revised definitions and changing priorities all create maintenance work. Managed support keeps it working, keeps improving it, and keeps FDI time on your calendar.

Failures and unresolved exceptions
Analyst edits and recurring corrections
Data and model changes
The next useful improvement

Decision gate. Ongoing support is scoped separately around what is live. From $5,000 a month, 12-month initial term from agreed go-live.

Commercials

Priced to sit next to a hire, not a consultancy.

Phase	Price
Diagnostic	From $25,000, fixed by team size
Build	Capped hours, agreed in the proposal before work starts
Training	Program fee in the proposal; clinics included
Managed services	From $5,000 a month; 12-month initial term
NSage add-on	$5,000 to $10,000 a month, by coverage

A typical first year for a two-to-three analyst team costs less than one analyst.

Diagnostic50% at booking, 50% on delivery.

BuildBilled monthly up to the cap, with weekly reporting of hours and forecast cost.

Training50% at scheduling, 50% after workshop four.

Recurring feesMonthly in advance.

Scope changesWritten approval of cost and timing first.

Third-party costsHosting, model usage and data costs are identified in the proposal.

Questions a skeptical PM asks

Asked and answered.

Why not just give the analysts Claude?

They already have it. The gap is set-up: your KPI definitions, your models, your note standard and what counts as material. That is investment work first.

Who will we actually work with?

A Forward-Deployed Investor leads the diagnostic in person and stays on the engagement. Named NSigma engineers build and maintain the workflows. The proposal names everyone who will have access.

How much of our time does the diagnostic take?

One day with the PM or CIO, half a day with each analyst, one day with operations and compliance at larger funds, and a one-hour readout.

Does our data leave the building?

Only if you choose. Fund-hosted, managed or hybrid, set by your data permissions and agreed before release.

Do you work with funds that own the same names?

Yes, behind a strict information barrier: separate environments, no client data in any shared component, and a personal trading restriction. See security and data.

What if an output is wrong?

It carries its sources, so the analyst can check it. Overrides are logged, and every workflow is tested on your past quarters first.

What if the diagnostic says don't build?

Then we say so. You keep the plan and the baseline, and can use them with any provider or later.

How do we measure success?

Each build ships with acceptance tests agreed before work starts, and every analyst edit is kept and reviewed monthly.

Is this only for earnings-driven hedge funds?

No. The diagnostic starts with your style. A compounder fund gets a thesis record and monitoring; an event fund gets deal-document tools.

Can our analysts maintain it without you?

Training is designed for that. Managed support is optional.

Team

An investor who learned to build, not an engineer who learned finance.

Thesis & Method connects the people who understand the investment question with the engineers who build the system to investigate it.

Founder · Forward-Deployed Investor NL

Founder

Nikola Legetic, CFA

Designed the Forward-Deployed Investor method and the product suite behind it. His role in every engagement is to understand the question beneath the task, make the analytical requirements explicit, and judge the result alongside the analysts who know the companies.

17 years of buy-side equity research; multi-sector analyst at Artisan, GQG Partners, Eagle Asset and RLCM
Led the implementation of data analytics, machine learning and AI for Artisan Global Equity
Co-founder of NSigma; Head of Product for NSage.ai and NSight Private Credit
Columbia Business School, AI for Finance & Business (2026); Claude Certified Associate, Foundations (2026)

NSigma engineering

The depth to build. The discipline to maintain.

An investment workflow is only as dependable as the engineering underneath it. NSigma implements the integrations, calculations, tests and operating controls behind every workflow. Technology choices follow the use case and your environment.

~30

developers, averaging 9 years of specialized experience

PhD

level support in quant research and machine learning

27001

ISO-certified information security management

Vendor-agnostic: built around your needs, not software licenses. Product work includes NSage.ai and NSight Private Credit.

One engagement, clear responsibilities

Who does what.

Forward-Deployed Investor

Defines the question.

Translates research intent into analytical requirements and judges whether the output is useful.

NSigma engineering

Makes it work.

Implements, tests and supports the agreed workflow.

Your firm

Owns the judgment.

Your analysts review the evidence. Your PMs keep every investment decision.

We start up to two new diagnostics a month, so your Forward-Deployed Investor stays on your calendar.

Research standard · Designed to be challenged

An answer is only useful if it survives scrutiny.

Every workflow must make it possible to retrace the evidence, reproduce the calculation, and see where human judgment entered the result.

Source documentEvidence and definitionsCalculation and interpretationHuman reviewAccepted research record

Keep the evidence attached.

Outputs lead the reviewer back to the source passage, the reporting period and the definition. A useful summary cannot hide a missing input or a contradictory disclosure.

Respect what was knowable.

Historical tests use an explicit information cutoff. A restated filing or revised dataset cannot silently stand in for what the analyst originally had. Where the original vintage is missing, that limitation stays visible.

Make calculations reproducible.

Extraction and language tasks can use AI. Arithmetic, reconciliations and model checks run in inspectable code with explicit inputs, units and assumptions. Same input, same answer, every time.

Separate fact from judgment.

Confirmed evidence, interpretation and unresolved questions stay distinguishable. A missing figure stays missing until a defensible source establishes it.

Preserve the disagreement.

Analyst edits and their reasons are kept and reviewed monthly. If nobody ever disagrees with the machine, nobody is checking it, so an acceptance rate near 100% triggers a spot check.

Keep approval with the team.

Drafts and alerts support review. Analysts approve research outputs and model changes. Position sizing, order entry and publishing are never built.

Security and data · For the COO, compliance and the data lead

Your environment. Agreed controls.

Your data rights, research ownership and operating environment are part of the specification. They are set before the build, not after.

Before we start

A mutual NDA is signed before the diagnostic begins. Engagement terms are set out in the proposal.

Deployment

Fund-hosted, managed or hybrid, chosen with your IT to fit your data permissions and existing systems. Access, operating responsibilities and the support route are agreed before release.

Certification

ISO 27001-certified information security management.

Who touches your data

Each engagement has a named team: your Forward-Deployed Investor and the NSigma engineers assigned to it. Only named people get access, inside the environment you choose and under your access controls. Access is logged and removed when the engagement or managed service ends.

AI providers

We work within the AI providers your firm has already approved. For our own build work, we use Claude Code and Codex under business terms that exclude training on your data.

Licensed data

We process only sources you are licensed to use with AI. The diagnostic confirms each one before any build.

Information barrier

Your positions, theses, models and notes are never shared with another client and never used to build or improve NSage, Research OS or any shared component.

Each client works in a separate environment. Reusable components contain no client data.

Personal trading

Anyone on our side with access to your data does not trade individual securities you hold or are researching.

During the engagement and for 30 days after. Diversified funds and ETFs are excluded. Each person attests to this annually.

Ownership

You keep

Bespoke deliverables and documentation
Client-specific code
Your own research and data

Licensed

NSage AI Earnings Predictor
Research OS
Reusable proprietary components

No tool sizes positions, enters orders, publishes, or changes a model without analyst approval.

Your vendor due-diligence questionnaire is welcome. We answer it before the diagnostic starts.

Start with the work

What is your team working around?

A model update that takes too long. A thesis that is hard to monitor. A research question you never get to. Bring one concrete example.

Your investment style, coverage and research process
The recurring work that consumes analyst capacity
The questions you would like to investigate more deeply

You leave with a proposed diagnostic scope, fee and on-site week, or a clear view that it is not the right time.

Book 45 minutes

Or prepare a short introduction and email it to us.

Your introduction

Send to FDI@nsigma.io

Your research process. Engineered by a Forward‑Deployed Investor.

Licenses are not capability.

Upkeep fills the week. Judgment gets what is left.

An investor defines the question. Engineering makes it work.

An investor who learned to build, not an engineer who learned finance.

The engineering arm of the service.

Keeps every investment decision.

A research record. A clear build order. A better question.

Broadcom (AVGO)

Where the week goes

Separate signal from story.

Your style determines what we build first.

Start with the process. Commit at each decision gate.

Your environment. Agreed controls.

Better questions. Working systems.

Get to the debate while it still matters.

What we would build

How we prove it

Hold every position to the conditions you wrote down.

What we would build

How we prove it

Follow the event to every name it touches.

What we would build

How we prove it

Answer the question the team never had time for.

What we would build

How we prove it

Keep what the team already knows.

What we would build

How we prove it

Nine examples. Nearly 40 in the library.

Prepare faster

Analyze deeper

Research further

Every workflow should leave the firm smarter.

Thesis and estimate ledger

Evidence bank

Relationship graph

Could cooling content protect Vertiv's growth if data-center builds slip?

Engagement patterns, not a shelf of software.

Begin with the desk. Earn the right to build.

Find the constraint. Price the remedy.

Week one, on site

Weeks two and three

End of week three

Where your time goes

Your research backlog

A build order

A working sample

A recommendation and price

Four things are confirmed with you first. The diagnostic checks each one, so none is a surprise.

Prove it on the work you already know.

Your analysts leave able to build, not just use.

Research changes. The system has to follow.

Priced to sit next to a hire, not a consultancy.

Asked and answered.

An investor who learned to build, not an engineer who learned finance.

Nikola Legetic, CFA

The depth to build. The discipline to maintain.

Who does what.

Defines the question.

Makes it work.

Owns the judgment.

An answer is only useful if it survives scrutiny.

Keep the evidence attached.

Respect what was knowable.

Make calculations reproducible.

Separate fact from judgment.

Preserve the disagreement.

Keep approval with the team.

Your environment. Agreed controls.

Before we start

Deployment

Certification

Who touches your data

AI providers

Licensed data

Information barrier

Personal trading

You keep