Flagship Case Study · CSL Capital · Private Credit / MCA Fund

One system. Six connected stages.
Every deal, start to decision.

The same mess a solar or home-services operator lives every day: leads, financing, install status, commissions, and reporting sitting in systems that don't agree, and no number you'd bet a decision on. This is what it looks like when one system finally ties them together.

Here, that system ran CSL Capital's entire lead-to-decision pipeline: automated ingestion, scoring, ML risk assessment, a proprietary loan-status system, automated reporting, and a live dashboard. Nektar built it as one connected system, end to end, on a governed warehouse that fails its own build if a number stops tying out.

Role
Fractional data & analytics lead
Domain
Private credit · merchant cash advance
Scope
6 connected components, 1 pipeline
Status
~14 months live, ongoing relationship
01 The Problem

Fragmented lead flow. Manual analysis. No single system.

CSL Capital, a private-credit fund making merchant-cash-advance and specialty-finance investments, was running its entire operation on data that couldn't keep up with the decisions riding on it.

None of this was a judgment problem. The team knew the business. What they didn't have was one connected system that captured a deal, graded it, tracked it, and reported on it the same way every time.

02 The System Built

The entire lead-to-decision pipeline: six components, built as one system.

Not six tools stitched together after the fact. Six stages designed to hand cleanly to the next. A deal that enters at Stage 1 flows through scoring, risk, status tracking, and reporting without a single manual re-entry point.

1 Ingest 2 Score 3 Risk Assess 4 Status Track 5 Report 6 Dashboard
STAGE 01

Automated lead ingestion

Deal emails from funding partners are read, extracted, and written into a structured record automatically. No one re-keys a deal by hand. Every incoming deal is captured the same way, every time.

STAGE 02

Lead scoring

A hard-fail gate plus a multi-point scoring rubric grades every deal that lands. The same rules apply whether the deal arrived at 9am or midnight. No exceptions made on memory or gut feel.

STAGE 03

Risk assessment

Deals that clear scoring feed an ML-based risk/underwriting model, retrained on a regular cadence. This is what turns "this looks like a good deal" into a modeled, auditable signal.

STAGE 04 Bespoke IP

A proprietary loan-status system

Not a repurposed CRM stage field. A purpose-built status taxonomy and determination logic designed for how this specific loan book behaves, with every status change audited and reversible only through a governed override path.

STAGE 05

Automated reporting suite

Executive, portfolio, and fund-level reports are generated and delivered on a schedule. No manual assembly step, no version where a number drifts from what's in the warehouse.

STAGE 06

A live dashboard

The governed decision surface leadership actually runs the fund on, pulled live from the same warehouse every other stage in this pipeline reads from and writes to.

Stage 4 is the piece that couldn't have been bought off a shelf. Nektar designed the status taxonomy and its determination logic specifically for how this loan book behaves (which statuses are terminal, which are watchlist-worthy, which require escalation) and made every change to it auditable. It's the single source of truth for "where does this deal actually stand," and it's differentiated IP, not a checkbox in someone else's CRM.

— why Stage 4 is called out, not just listed
03 The Governance Layer

Every number, tested before a decision-maker sees it.

All six stages sit on top of the same governed warehouse. The standard here wasn't "does the dashboard look good." It was "would this number survive an audit."

76
Tested data models: staging → intermediate → marts
109
Automated tie-out tests, fail the build on a mismatch
6
Connected components, one pipeline, no gaps
4
Outside funding-partner portals reconciled automatically
Production test suite · 109 / 109 passing

Every number that reaches any of the six stages (a deal score, a risk signal, a status change, a report, a dashboard figure) first passes through this layer. That's not a marketing number. It's the literal count of transformation steps and automated checks standing between raw source data and anything a decision-maker reads.

04 What Changed, Concretely

Manual, ad hoc, and unaudited → automated, governed, and tested.

LayerBeforeAfter
Lead ingestionAnalyst hand-keys deal terms from email into the CRMAutomated extraction pipeline, no manual re-entry
Deal scoringJudgment-based, no consistent rubricHard-fail gate + scoring rubric, applied identically every time
Risk assessmentSized on institutional memory of past dealsML risk model, retrained on a cadence
Loan statusWhatever field was closest, no audited definitionProprietary taxonomy, single source of truth, audited overrides
Collections accountingReversed/voided payments never ingested, overstated collectionsDedicated reversal tracking nets every reversal before a number is reported
Partner reconciliationEach portal reconciled by hand, ad hoc cadenceAutomated reconciler, shared identifier, scheduled tie-out
Reporting layerHand-written queries, no test coverage76 tested dbt models, 109 tie-out tests
Recurring reportsManually rebuilt weekly/monthlyAutomated, scheduled, no manual rebuild
Decision surfacePeriodically rebuilt spreadsheetLive dashboard, same governed warehouse
05 What It Enabled

Faster decisions, on numbers that tie out.

06 Why It Matters

This fund moves real capital on these numbers.

Every model is tested. Every partner tie-out is automated. Every deal that enters the pipeline is captured, scored, risk-assessed, status-tracked, reported, and surfaced on a dashboard, through one connected system, not six separate tools someone has to manually stitch together.

That's the same rigor available to any operator running a sales, install, or lending business across systems that don't agree with each other. The stakes are different, but the discipline is identical: capture once, score consistently, track status honestly, report automatically, and never make a decision-maker guess which number is right.

BigQuerydbtPythonLLM extractionBQMLCloud RunCloud SchedulerAutomated CI/CD

Want this rigor applied to your numbers?

This is the same discipline behind the Single Source of Truth Build. I build the reconciliation layer underneath your CRM, sales tools, and financing systems, so leadership runs on one governed set of numbers instead of four disagreeing ones. Start with a 2-week paid diagnostic that shows you exactly where your numbers disagree.

Find the right data fix Explore Nektar Insights