menu_bookGuide

Automating capital calls with a complete audit trail for UK PE firms

How a UK private equity firm runs the full drawdown cycle - calculation against commitment schedules, notice drafting, issuance, receipt tracking and chasing - with AI agents doing the work, maker-checker approval before anything leaves the building, and every action timestamped and exportable.

Why capital calls carry more risk than their volume suggests

A mid-sized UK private equity firm might issue a few dozen drawdowns a year. That low volume is exactly why the process stays manual: it never looks big enough to industrialise. But each notice moves real money from an LP's account into the fund on a fixed date, on a number your team calculated, against a commitment schedule held in a side letter that may itself carry an excuse provision, an equalisation adjustment or a different management fee basis.

Get it wrong and three things happen at once. There is a financial consequence - an over-call to be returned, an under-call to be topped up, or an interest calculation to unwind. There is an LP relationship consequence, because institutional investors read a miscalculated drawdown as a signal about the firm's operational maturity, and they say so at re-up. And there is an evidence consequence: when the depositary, the auditor or the FCA-regulated adviser asks how that number was reached and who approved it, the honest answer in most firms is a spreadsheet, an email thread and someone's memory.

This guide covers the drawdown cycle specifically, for UK PE, at implementation level. For the general version across fund types, see automating capital call processing and notices. For the governance mechanics that sit underneath, see building an audit-ready fund operations process.

1. The process, and where it breaks down

Strip out the firm-specific detail and every capital call runs through the same four stages. Each one has a characteristic manual failure.

Stage What has to happen How the manual version fails
Calculate the call Total requirement split across LPs by unfunded commitment, with side-letter terms, excused investors, equalisation for later closes and fee treatment applied per investor. A commitment schedule held in one spreadsheet, side letters in a document folder, and manual reconciliation between them. Excused or defaulting LPs get missed; the maintained copy diverges from the ledger.
Draft the notices A per-LP notice: amount, due date, bank details, purpose split between investments, fees and expenses, and remaining unfunded commitment. Mail-merge from the same spreadsheet. One wrong reference or stale bank detail propagates to every notice, and the version issued is not necessarily the version filed.
Issue and track receipt Distribution through the investor portal or data room, then matching incoming cash against expected amounts per LP. Receipts reconciled by eye against a bank statement. Partial payments, FX shortfalls and payments referencing the wrong fund sit unresolved in a thread.
Chase and resolve Late payers chased on a defined clock; LP queries answered; disputes escalated; default provisions applied where required. Chasing depends on someone remembering. The reason a payment was late, and the decision not to apply default interest, is never recorded anywhere durable.

The common thread is that the evidence is a by-product of nothing. Nothing in the manual process produces a record; it produces a result, and the record has to be reconstructed afterwards from whatever survived in mailboxes and file versions.

2. What automating this actually looks like, step by step

In Next Matter the drawdown runs as one orchestrated process across the systems you already use - the fund accounting or administration platform, the investor portal or data room, the bank feed and your email or CRM. An AI agent executes the mechanical work at each step and raises anything it cannot resolve cleanly as an exception, with the context a human needs to decide, rather than asking a person to re-check every case.

Pull the commitment position

The agent reads each LP's commitment, drawn-to-date and unfunded balance from the administration system or ledger, together with the applicable side-letter terms held against the investor record. Every value read is stored with its source and the timestamp it was read at.

AI agent

Calculate the per-LP allocation

The agent applies the call amount pro rata to unfunded commitments, then applies the exceptions: excused investors, capped participations, equalisation for investors admitted at a later close, and the fee and expense split. The calculation basis is retained, not just the result.

AI agent

Flag what does not reconcile

A commitment in the ledger that disagrees with the register, an LP whose allocation exceeds remaining unfunded commitment, a missing or expired bank mandate, an investor still mid-KYC refresh. Each becomes an exception task routed to a named owner, carrying both sides of the mismatch and the investor's history - not a warning in a log.

AI agent

Draft the notices

Per-LP notices are generated from the approved template with the calculated figures, the purpose breakdown, the due date and the remaining unfunded commitment. The generated set is held as a single versioned artefact so what was reviewed is what is issued.

AI agent

Review and approve before issuance

The calculation basis, the exception log and the drafted notices go to a checker who did not prepare them. Nothing is sent until that approval is recorded. See section 3.

Human approval

Issue through the investor channel

Notices are distributed through the portal or data room and, where required, by email, with delivery status captured per LP. The issued document, the recipient and the timestamp are recorded against the case.

AI agent

Match receipts against expected amounts

Incoming cash from the bank feed is matched per LP against the expected amount. Full matches close automatically. Partial payments, FX shortfalls, unreferenced credits and overpayments raise exceptions with the comparison attached.

AI agent

Chase on a defined clock

Unpaid positions escalate at the intervals your policy sets rather than when someone notices - reminder before the due date, chase after it, escalation to the named partner or IR lead beyond the grace period. Every chase and its response is part of the record.

AI agent

Resolve disputes and queries with a stated reason

An LP query or a disputed amount routes to the accountable person with the full calculation behind that investor's number. The decision - amend, confirm, waive default interest - is recorded with its reasoning, and any amendment re-enters approval rather than being edited in place.

Human decision

Close the call as a complete record

The cycle closes when every position is settled, waived or escalated. The case record - inputs, calculation, exceptions, approvals, notices, receipts, chases and overrides - is complete at the point of closure, with nothing left to assemble.

Human sign-off

The point of the agent is not that it is clever about the hard cases. It is that it removes the volume of easy cases so your team's attention lands on the mismatched commitment, the late payer and the LP query - each arriving with the context already gathered.

3. Where maker-checker approval sits

Capital calls are among the clearest cases for four-eyes approval, because the consequence of an error is immediate and external: money is requested from investors on a number you published. A mistaken NAV can often be restated; a mistaken drawdown has already been sent to every LP in the fund.

What runs without a human

  • Reading commitment, drawn and unfunded balances from source systems.
  • Performing the pro-rata allocation and applying documented side-letter and equalisation terms.
  • Generating draft notices from the approved template.
  • Matching receipts that reconcile exactly to the expected amount.
  • Sending scheduled reminders and chases within policy.

What always requires human sign-off

  • Issuing the call. No notice is distributed until a checker who did not prepare the calculation has approved the basis, the per-LP schedule and the drafted notices.
  • Any amended figure. If an allocation changes after review - a corrected commitment, a re-applied excuse - the change re-enters approval; it is never edited into an approved set.
  • Recording a payment that does not match. Partial settlements, overpayments and unreferenced credits are accepted by a person, with the treatment stated.
  • Waiving or applying default provisions. Default interest, grace extensions and forbearance are decisions with a named decision-maker and a reason.
  • Changing the gate itself. Altering who can approve, or the thresholds at which escalation applies, is versioned and attributable in the same way as the work.

Separation is enforced by the platform rather than by policy: the checker cannot be the maker, routing to an eligible approver is automatic, and there is no urgency bypass when the call is going out on Friday. Deeper: enforcing four-eyes and maker-checker approvals.

4. What the audit trail captures

The record is produced by execution, not by anyone remembering to save it. For a single call, it holds:

Captured Detail held
Inputs Every source system read, the values returned per LP, and the version and timestamp of each read.
Calculation The allocation basis, the terms applied per investor, and the resulting figure - so the number can be re-derived, not just re-stated.
Notices The exact document issued to each LP, versioned, alongside the template version it was generated from.
Approvals The named approver at each gate, what was presented to them, their decision and the timestamp. Rejections and their reasons persist.
Issuance Recipient, channel and time of dispatch per LP, with delivery status where the channel reports it.
Receipts Cash matched per LP, date received, and any shortfall, overpayment or FX difference with how it was treated.
Exceptions Every break raised, its context, owner, escalation history, resolution and the stated reason.
Chases Each reminder and escalation sent, when, to whom, and the response.
Overrides Any deviation from the standard path - attributed, reasoned, and held in the same record rather than a side channel.
Agent actions Actions taken by an AI agent are labelled as such, with the inputs they acted on, and distinguished from human actions.

Every element is timestamped and exportable on demand as a single pack for the call, or filtered to one LP. The test that matters is that someone else picks the case and you produce the pack without preparing it.

5. The UK regulatory context

Next Matter is a platform, not a regulatory opinion, and nothing here is a claim of certification or sign-off against a specific FCA rule. What the platform does is produce the kind of evidence UK firms are expected to be able to produce, in the form supervisors and depositaries ask for it.

  • Evidence produced, not reconstructed. UK supervisory practice consistently tests whether a firm can demonstrate that a control operated in a specific case, not whether it is documented in a manual. A per-action record created at execution answers that directly; a pack assembled for the visit invites the question of what an ordinary day looks like.
  • Record-keeping around investor funds movements. Drawdowns move investor money on a schedule, through named bank mandates, against contractual commitments. Holding the mandate used, the amount requested, the approver and the receipt in one continuous record supports the recordkeeping expectations that apply to firms handling investor money and their delegates.
  • AML and KYC status as a gate, not a lookup. Where an investor's verification is expired or mid-refresh, that condition can block issuance or route for a compliance decision, rather than being a check someone was supposed to perform. UK AML obligations include ongoing monitoring rather than one-off onboarding checks, and treating status as a live gate makes that continuous by construction.
  • Operational resilience (PS21/3). Where capital call processing forms part of an important business service, firms are expected to map it, set impact tolerances and evidence that they monitor and control it in practice. Live visibility of where each call sits, which positions are breaching tolerance and what the control record shows is exactly that evidence. More: UK operational resilience under PS21/3.
  • Accountability for AI-supported steps. Where an agent performs part of the process, its actions are labelled, its inputs retained, and a named person approves the consequential step. That is the same principle underlying EU AI Act oversight and record-keeping, and it is what a UK firm's own governance committee will ask for regardless of framework.
  • Delegation and oversight. Where the drawdown is run by an administrator rather than in-house, the same record gives the manager oversight evidence over a delegate rather than a monthly summary and trust.

The practical framing for a UK PE firm: you are not trying to prove the process is perfect. You are trying to be able to show, for any call, what was calculated, on what data, who checked it, when it went out and what happened afterwards - including the cases that went wrong.

6. One call cycle, end to end

A fund with 38 LPs issues a 24M GBP call, split 21M GBP for a follow-on investment and 3M GBP for management fees and fund expenses, with a 10 business day due date.

Day 0 - inputs and calculation

The agent reads unfunded commitments for all 38 LPs from the administration system and the applicable side-letter terms from each investor record. It allocates the 24M GBP pro rata, applies the excuse provision for two LPs excluded from this sector, applies equalisation for three investors admitted at the second close, and splits fees per the applicable basis.

Day 0 - one exception raised

LP 17's unfunded commitment in the ledger reads 1.8M GBP; the investor register shows 1.55M GBP following a partial transfer completed the previous quarter that was never reflected in the ledger. The agent raises an exception carrying both figures, the transfer documentation and the resulting allocation difference of roughly 96k GBP, and routes it to the fund controller.

Day 1 - exception resolved

The controller confirms the register is correct, records the reason with the transfer document attached, and the allocation is recalculated. The original figure, the correction, the evidence and the resolver stay in the record; nothing is silently overwritten.

Day 1 - maker-checker review

The CFO, who did not prepare the schedule, reviews the allocation basis, the per-LP table, the single exception and its resolution, and the drafted notices. She approves. The approval is recorded with a timestamp and what she was shown.

Day 1 - issuance

38 notices are issued through the investor portal with email notification. Each dispatch is logged per LP with the exact document version sent.

Days 2-10 - receipts and chases

Payments match automatically as they arrive. On day 7 the agent sends the scheduled pre-deadline reminder to the six LPs still outstanding. On day 9, one LP pays 40k GBP short after FX; the mismatch raises an exception rather than closing as paid.

Day 11 - shortfall and late payer

The FX shortfall is accepted by the IR lead as a top-up to be collected with the next call, with the reason recorded. One LP is still unpaid; the escalation routes to the named partner on the policy clock, and payment arrives on day 12 with default interest waived - a decision recorded against a named person with its reasoning.

Day 12 - closed and audit-ready

The call closes. The pack - inputs and versions, calculation basis, the commitment exception and its evidence, the CFO's approval, 38 issued notices, 38 receipts with two treated as exceptions, every chase, and the interest waiver - exports in minutes when the auditor picks this call six months later.

The whole cycle involved two genuine human decisions beyond the approval gate: a commitment mismatch and an FX shortfall. That is the intended shape - people spending their time on the two cases that needed judgement rather than checking the 36 that did not.

Evaluating a platform for this specific use case

Most tools that claim capital call automation are strong on one part of the cycle. Fund accounting systems calculate well and evidence weakly across boundaries; investor portals distribute well but do not govern; generic workflow tools orchestrate but leave you to build the controls. Questions worth asking of anything you evaluate:

  • Does it span the whole cycle? Calculation, notice, issuance, receipt matching and chasing in one governed case - or does the trail stop at the system boundary?
  • Is maker-checker enforced by the system? With separation of maker and checker, no urgency bypass, and configuration changes to the gate versioned.
  • Is the record produced at execution? Or assembled by a reporting layer afterwards from partial logs.
  • Can you re-derive a figure, not just re-read it? The calculation basis and source versions, not only the final number.
  • Are exceptions first-class objects? With owner, context, escalation clock and stated resolution - rather than comments on a task.
  • Are AI actions labelled and bounded? Distinguished from human actions in the record, with consequential steps gated.
  • Does it orchestrate your stack or replace it? Your administrator, ledger, portal and bank feed should stay where they are.
  • Who changes the process? If a side-letter term or an approval threshold changes, can your operations team change it in days, with the change itself versioned?
  • Can you export a full case pack on demand? For a call someone else picks, without preparation.

These controls run in production in regulated environments today: at Ocorian, where 300+ fund specialists work in the platform daily across fund operations, at Trade Republic in high-volume client operations, at b2venture across a portfolio of around 800M EUR AUM, and at Swan in embedded finance. The platform is SOC 2 Type II and ISO 27001 certified, with SSO/SCIM and data residency you control.

Walk through your next drawdown

Bring a real capital call - the commitment schedule, the side-letter quirks, the LPs who always pay late - and we will show the record it would produce, gate by gate.

Book a demo