Reducing deal-screening cycle time at a GCC sovereign wealth fund.

At a glance.

All client identifying details are anonymised under NDA. A senior lead reference call is available to qualified prospective clients on request.

The client is one of the three largest sovereign wealth investors in the GCC, with mandates spanning private equity, infrastructure, listed equities and direct strategic stakes. The deal team — roughly one hundred and forty professionals across three offices — was receiving an average of three hundred and forty inbound opportunities per month from banks, advisors and government referrals. Every opportunity required a first-pass triage memo: a structured read of teaser, CIM, sector context, sponsor history, and preliminary fit against the active investment thesis. The memo went into an internal screening committee that met twice weekly. The committee chair was a senior managing director whose calendar was the binding constraint on the entire pipeline.

Two things were breaking simultaneously. First, analyst capacity: associates and senior associates were spending between sixty and seventy percent of their week on triage, leaving the actual diligence on live deals chronically under-resourced. Second, consistency: the screening memos varied widely in depth and tone depending on which associate had drafted them, and the committee chair was making decisions on materially uneven inputs. The client had previously evaluated three off-the-shelf investment-research products and rejected all three: the data residency story was unacceptable for a sovereign mandate, the Arabic-language handling on deal collateral from regional sponsors was poor, and none could be tuned to the house investment thesis without exposing it to a third-party tenant.

We deployed a fourteen-person pod inside the investment office for eleven months. The pod was co-located with the deal team on the same floor and operated under the client's information-security regime from day one. The senior engineer on the proposal — a founding partner of the firm — was the senior engineer on the engagement, and remained so for the full duration. We built an agentic investment-memo system on the client's sovereign-cloud tenancy, with the model layer running entirely inside the client's perimeter. The system ingests teasers, CIMs and sponsor materials, performs structured extraction in Arabic and English, retrieves relevant precedent transactions from the client's own deal history, and drafts a first-pass committee memo in the client's house format. The memo is never automatically committed: the originating analyst reviews, edits and signs the final version. We also rebuilt the screening committee's weekly pre-read into a single dashboard with explicit thesis-fit scoring and an auditable provenance trail for every claim.

By month four the system was drafting the first pass on every inbound opportunity. By month seven the time from teaser receipt to a committee-ready memo had fallen by eighty-two percent. Analyst capacity reclaimed from triage was reallocated to live diligence, which the client measures internally as a four-point-three-times increase in deals screened per analyst per quarter at constant headcount. The screening committee chair reports that the inputs are now uniform in depth and tone, and that committee meetings have shortened by approximately thirty-five percent. The engagement remains in flight: the pod is currently extending the system to portfolio monitoring and to the post-mortem function that closes the learning loop on rejected opportunities.

Three. First, the binding constraint was never model quality — it was the trust contract with the originating analyst, which we earned only by making the system's outputs editable, auditable and visibly conservative. Second, Arabic-first ingestion is not a localisation feature, it is a precondition: roughly a third of inbound collateral from regional sponsors arrives in Arabic and the off-the-shelf products were silently degrading it. Third, the sovereign-cloud constraint was an advantage, not a tax — running entirely inside the client perimeter is what made the house investment thesis safe to encode in the system in the first place.