Research deck automation · asset allocation engine

The research deck, and the allocation behind it, both automated.

Sapient Finserv's advisers needed a house research deck on a fixed cadence and allocation guidance they could defend to a client in the room. We automated the deck end to end and built an allocation engine that treats mandate limits as hard constraints.

Deck turnaround: 2 days → 14 min
Mandates modelled: 9 profiles
Stress scenarios: 24 per run
Claims sourced: 96%

The starting point

Two bottlenecks, one week apart.

The deck was rebuilt manually every cycle from the same sources, and allocation guidance was assembled from a model portfolio plus judgement — defensible in conversation, but hard to reproduce or evidence afterwards.

Cadence pressure

A fixed publication date meant the deck was always the constraint, never the analysis.

Allocation by hand

Limits, liquidity windows and concentration caps were checked manually, late in the process.

Thin audit trail

Reasoning lived in email threads rather than alongside the numbers.

Workstream one

Research deck automation.

One pipeline produces the whole deck in the Sapient house format — macro read, asset class views, fund shortlist and the adviser-facing summary — with the citation trail attached.

IngestExtractDraftChartCiteSign-off

01
Collect
Market data, manager updates and internal notes gathered continuously between cycles.
02
Read
Agents summarise each source and tag the figures they will reuse.
03
Assemble
Sections drafted into the house template with charts generated from the data itself.
04
Sign-off
The reviewer sees only what changed and what needs a human decision.

Cycle time per deck

Indicative data

Manual, before960 min

First automated run47 min

Steady state today14 min

Deck composition

Indicative data

Macro and rates28%

Asset class views31%

Fund shortlist24%

Adviser summary17%

StepBeforeWith agents

Data refreshManual pullContinuous

Chart rebuild3 hoursAutomatic

Draft narrative1 day9 minutes

Citation trailOn requestBuilt in

Workstream two

The automated allocation engine.

Give it a mandate — target return, horizon, eligible classes, liquidity and concentration limits — and it returns policy weights, a stress read and the reasoning for every move it makes.

Balanced mandate, proposed weights

Illustrative

Domestic equity34%
International equity18%
Fixed income26%
Alternatives14%
Cash8%

Modelled drawdown by scenario

Illustrative

Rate shock, +200bp9%

Credit widening12%

Equity drawdown17%

Liquidity squeeze7%

Combined stress21%

Glide path vs policy target

Illustrative

Y1Y3Y5Y7Y10

Engine allocation (indicative)Policy target

Constraint checks

Limits are enforced, not advised.

Every run reports which constraints bound the answer, so the adviser can explain not just the allocation but what it was not allowed to do.

Liquidity windowPass92% of the book redeemable inside thirty days against a 75% floor.

Single-name capPassLargest holding at 4.1% against a 5% mandate cap.

Alternatives sleeveFlagBinding at the 15% ceiling — the engine would hold more if permitted.

Duration bandFlagSitting at the top of the permitted band; sensitive to a further rate move.

Unrated creditBlockTwo candidate funds excluded outright by mandate. Recorded, not overridden.

Outcome

What changed on the desk.

Publication is no longer the constraint

The deck lands on time without the week-long scramble.

Allocation is reproducible

Same mandate, same inputs, same answer — with the reasoning attached.

Advisers can defend it

Each weight arrives with the trade-off it accepts stated in plain language.

Compliance reads the trail

Sources and constraint checks are part of the output, not an afterthought.

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.