Research deck automation · index construction

An in-house index, and decks that write themselves.

Scripbox wanted two things off the analyst's plate: the recurring research deck, and the manual spreadsheet work behind their own index. We rebuilt both as agent pipelines — same house format, same rules, produced in minutes and fully sourced.

Deck turnaround: 3 days → 18 min
Index rebuilds: Quarterly, automated
Screens run: 1,860 / cycle
Claims sourced: 97%

The starting point

Two processes that ate the whole week.

The research team was assembling decks by hand from filings, factsheets and internal notes, then re-deriving index constituents in spreadsheets every quarter. Both were correct and both were slow — and neither left an audit trail anyone could read six months later.

Deck assembly by hand

Every chart, table and paragraph copied across from sources, reformatted, then checked twice.

Index work in spreadsheets

Screens, weights and exclusions lived in linked workbooks that only one person fully understood.

No standing challenge

Strategies were reviewed when someone had time, not systematically before publication.

Workstream one

Research deck automation.

A pipeline that reads the source material, drafts the narrative in Scripbox's own slide grammar, and hands back a deck with every figure linked to the document it came from.

IngestExtractDraftChartCiteReviewExport

01
Ingest
Filings, factsheets, call transcripts and internal notes land in one workspace.
02
Extract
Agents pull the figures and quotes, each tagged with page and paragraph.
03
Compose
The narrative is drafted against the house template, section by section.
04
Check
A reviewer agent flags anything unsourced or inconsistent before export.

Time on a standard monthly deck

Indicative data

Manual, before1440 min

First automated run62 min

Steady state today18 min

Where the content comes from

Indicative data

Regulatory filings38%

Fund factsheets24%

Internal research notes21%

Market data17%

StepBeforeWith agents

Source gathering6 hoursContinuous, automatic

Chart production4 hoursGenerated from the data

Narrative drafting1.5 days12 minutes

Fact checkSpot checkEvery claim, every run

Workstream two

Index construction, in-house.

The screening rules, factor weights and exclusion logic now live in a workspace rather than a workbook. Agents propose candidate rules, back-test the screen, and write the methodology note from the run itself.

Index vs benchmark, indexed to 100

Illustrative

Y1Y2Y3Y4Y5

Scripbox index (indicative)Broad benchmark

Factor weights in the live screen

Illustrative

Quality32%

Low volatility24%

Value22%

Momentum14%

Size tilt8%

Universe screened

1,860 names

Constituents held

Rebalance

Quarterly, rules-driven

The critique panel

Agents that argue with the strategy.

Before anything reaches investors, a standing adversarial panel reads the strategy and returns the strongest case against it. Dissent is recorded, not averaged away.

ConcentrationFlagTop ten names carry 34% of the book — inside policy, but worth stating in the methodology note.

LiquidityPassEvery constituent clears the twenty-day traded-value floor with room to spare.

Factor overlapFlagQuality and low-volatility sleeves share 41% of names; the tilt is thinner than it looks.

SurvivorshipBlockBack-test excluded delisted names. Rerun on the point-in-time universe before publishing.

Cost dragPassModelled turnover keeps implementation cost inside the stated tracking budget.

Outcome

What changed on the desk.

Analysts moved up the stack

The team spends its time on judgement calls, not assembly.

Index owned internally

No external vendor between Scripbox and its own methodology.

Every claim sourced

Compliance reads the citation trail instead of asking for it.

Challenge is standing

Nothing publishes without the counter-case on record.

Want the same on your desk?

Send us one research task you repeat every week and we'll run it through the same pipeline on your own material.