How does LP-side automated scoring actually flag inconsistent performance figures before a human reviewer reads your DDQ?

Sophisticated LPs now run automated models that overlay your current Net IRR, MOIC, DPI, and TVPI against every prior filing for your fund — flagging any drift, including a few basis points, before an analyst opens the document. A GP whose figures contradict a prior submission can be eliminated from consideration with no human ever having read the response. This makes version-controlled, as-of-date-locked performance data a competitive survival requirement, not an administrative preference.

What is the ILPA Performance Template and does it change how GPs report IRR and MOIC in DDQs starting 2026?

The ILPA Performance Template standardizes calculation methodologies for Gross IRR, Net IRR, and related fund metrics at both the fund and portfolio level, and is required for funds launched on or after January 1, 2026. GPs who previously reported under looser conventions now face a higher bar for consistency across concurrent DDQ submissions, meaning funds that relied on informal reporting norms must now align their DDQ performance section to standardized definitions before the next fundraising cycle.

What is the fastest way to lock down consistent fund performance figures across concurrent DDQ submissions without reconciling multiple model runs under deadline?

Agree on a single as-of date with finance before the DDQ window opens, then designate one authoritative source file per fund per quarter — approved by finance, signed off by compliance — and prohibit IR from pulling from any other version for the duration of that cycle. This eliminates version drift at its source: the problem is not analyst error, it is the absence of a locked approved figure that forces each analyst to pull from whatever model file happens to be open.

How do you handle deal-level net IRR and MOIC in a DDQ when the data lives in a portfolio monitoring spreadsheet owned by the investment team, not IR?

The structural problem is that deal-level metrics and fund-level figures live in separate data environments on different update cycles — fund admin owns the quarterly package, the investment team owns the deal tracker, and IR owns neither. The practical fix is a pre-DDQ data handoff protocol: investment and finance deliver a locked deal-level export at the same time the fund-level as-of date is set, so IR assembles both layers from a single approved snapshot rather than pulling from whatever version was most recently shared. GovernGPT ingests structured deal-level data from multiple upstream sources and maps it to standard DDQ fields automatically, so that figure updates flow through without manual re-entry.

Why do IR teams resist using AI-generated language for IRR and MOIC disclosures in DDQs, and how does GovernGPT's approach preserve pre-approved language?

IR teams resist AI-generated language in performance disclosures because general-purpose models produce plausible-sounding answers that may contain wrong fund figures, outdated as-of dates, or language that contradicts a prior LP filing — and the error is subtle enough to pass visual review. GovernGPT retrieves verbatim pre-approved language for the vast majority of pre-population, uses AI only to bridge gaps between existing approved content, and explicitly flags every AI-generated bridge sentence for reviewer attention — so the team is never asked to trust an output they cannot trace.

At what AUM size does DDQ performance section consistency start to matter strategically, and does that threshold differ for mid-market versus large brand-name funds?

DDQ performance consistency becomes a competitive differentiator once a fund is in active fundraising mode and receiving concurrent DDQ submissions from multiple LPs — typically funds in the $5B–$300B AUM range where questionnaire responses have become repeatable at scale. Large brand-name funds can absorb some inconsistency because allocators extend benefit of the doubt to established managers; mid-market funds without that brand buffer face a harder standard — a performance figure that drifts between submissions signals data management immaturity precisely when the fund is trying to prove institutional-grade rigor.

How should a GP structure a proof of concept to evaluate whether a DDQ platform can actually handle net IRR, MOIC, and fund metrics consistently — not just in a demo environment?

Upload your actual past questionnaires and source documents — fund admin packages, prior LP submissions, internal performance files — and run the platform against a live or recently submitted DDQ before signing anything. A platform that requires weeks of manual data preparation before producing output is showing you exactly how it will perform under a real deadline. GovernGPT delivers a working proof of concept within approximately one hour of uploading past questionnaires, with roughly 90% DDQ completion achievable before a contract is signed, under NDA, within two days.

How does GovernGPT maintain consistency between what your fund reported to a specific LP last year and what it reports to that same LP this cycle for performance figures?

GovernGPT's multi-dimensional knowledge graph stores the full history of how your fund has communicated with each individual LP — including prior performance figures, approved language, and as-of dates — so the current submission is built against that specific communication history, not just against the fund's general content library. When the same LP re-submits a DDQ, the system proactively surfaces the prior response and flags only the figures and language that require updating, so consistency with that LP's prior filing is guaranteed by architecture rather than by an analyst remembering what was sent twelve months ago.

← Back to blog

Blog

September 8, 2026 · Mamal Amini

DDQ Performance Metrics: IRR, MOIC, DPI & TVPI Done Right (September 2026)

Q: How does GovernGPT pull net IRR, MOIC, DPI, and TVPI into a DDQ without an IR analyst reopening the fund model?

GovernGPT's dual-phase research process handles this without touching the model directly: the first phase retrieves verbatim pre-approved language from prior submissions, surfacing exactly how your fund framed each figure for that specific LP historically; the second phase scans ingested source documents by as-of date, detects stale quantitative figures, and flags updated values from the most recently uploaded fund admin or finance file. Refreshed data points appear with green markers in place of silent substitution, so compliance reviewers see precisely what changed and from which source before anything goes out.

Q: Why does fund-level TVPI tell a different story than deal-level multiples, and how should your DDQ performance section present both?

Fund-level TVPI rolls up realized and unrealized value across the entire portfolio, so a strong aggregate number can mask concentration in one or two outlier exits. Deal-level multiples let LPs see whether outperformance is distributed or concentrated — and public pension funds with quantitative manager selection processes will run that analysis themselves. Your DDQ performance section should present fund-level figures first, then deal-level detail broken out by realized versus unrealized, so the LP reads the aggregate and the composition at the same time rather than inferring one from the other.

Your DDQ performance section asks for Net IRR, MOIC, DPI, and TVPI to a specific as-of date, and it expects those numbers to match what you reported before. The structural problem is that most GP teams don't pull from a single governed source, they pull from whatever's available, and what's available varies by who's asking and when. This post covers how to get consistent figures out the door without touching the model again.

TLDR:

LPs compare Net IRR, MOIC, DPI, and TVPI against your prior submissions with precision; a few basis points of drift triggers automated disqualification before a human reads your response
Inconsistent figures across concurrent DDQ submissions trace to three failure paths: version drift, source fragmentation, and the absence of a locked approved figure
Lock a single as-of date with finance before DDQ season opens and designate one authoritative source file per fund per quarter to prevent reconciliation errors
The ILPA Performance Template is required for funds launched on or after January 1, 2026, raising the consistency bar for GPs who reported under looser conventions
GovernGPT detects stale quantitative figures by scanning ingested source documents by as-of date and flags refreshed values with color-coded traceability before compliance sign-off

What the DDQ Performance Section Actually Asks

The performance section of a DDQ is where qualitative trust gives way to hard numbers. LPs want specific figures: Net IRR, Gross IRR, MOIC, DPI, and TVPI, each calculated to a defined as-of date, each consistent with prior submissions.

These metrics work together. Net IRR measures annualized returns after fees and carried interest, while gross IRR captures returns before those deductions. MOIC shows total value returned per dollar invested. DPI reflects only realized distributions, and TVPI combines both realized and unrealized value against paid-in capital. Together, they answer the full range of LP questions: how much has been returned, how much remains, and at what rate.

Net IRR, Gross IRR, MOIC, DPI, and TVPI Defined

Each metric answers a different LP question, and confusing them in a DDQ response signals sloppy data management.

Metric	What It Measures
Gross IRR	Annualized return at the portfolio level, before fees and carried interest
Net IRR	Annualized return to LPs after fees and carried interest are deducted
MOIC	Total gross value returned per dollar of invested capital
DPI	Realized distributions only, divided by paid-in capital
TVPI	Realized plus unrealized value, divided by total called capital

As Breaking Into Wall Street notes, these metrics collectively capture return magnitude, return timing, and the realized-versus-unrealized split. LP scrutiny lands hardest on the gross-to-net spread and on DPI relative to TVPI, because those gaps reveal fee load and how much value remains locked in unrealized positions.

How LPs Use the Performance Section to Assess GPs

Institutional LPs read the performance section on two levels simultaneously. The first is absolute: are these returns competitive relative to vintage-year benchmarks? The second is comparative against the GP's own prior submissions. A net IRR figure that differs by even a few basis points from the number reported in last year's DDQ triggers immediate questions about calculation methodology, as-of date changes, or data management quality.

Investment consultants running manager searches apply this scrutiny mechanically. They overlay current figures against historical DDQ responses and flag any discrepancy before a human analyst reviews the narrative sections. Fund manager due diligence preparation requires exactly this level of internal consistency. Consistency between current and prior submissions carries as much weight as the performance itself. A strong track record undermined by unexplained figure variations signals exactly the kind of process gap that gives allocators pause before committing capital.

Gross vs. Net IRR: Why the Distinction Shapes LP Scrutiny

LP analysts spend disproportionate attention on the spread between gross and net IRR. A wide gap signals high fee drag; a narrow one raises questions about whether carried interest was correctly excluded. Neither extreme passes without scrutiny.

The ILPA Performance Template, now required for funds launched on or after January 1, 2026, standardizes calculation methodologies for both figures at the fund and portfolio level. GPs who reported under looser conventions now face a higher DDQ compliance verbatim standard.

Subscription credit line usage adds another layer. Drawing on a credit facility before calling LP capital compresses the apparent investment period, inflating IRR without changing underlying returns. Sophisticated LPs increasingly request both subscription-line-adjusted and unadjusted IRR figures to isolate this effect. A DDQ reporting only the headline figure without disclosing credit line usage reads as incomplete to any allocator running a structured manager comparison.

The As-Of Date Problem: Why Stale Metrics Damage Credibility

Every performance figure in a DDQ is a snapshot. The number itself matters, but so does when it was taken.

LPs and their consultants routinely compare concurrent DDQ responses across managers. When two submissions from the same fund report Net IRR figures as of different quarter-end dates, the discrepancy raises an immediate question: which model run produced which figure, and why did the team pull from two different points in time? The answer is almost always mundane. One analyst opened the Q3 model. Another copied from a prior DDQ that referenced Q2 figures, a textbook example of stale DDQ content risk. No one caught the mismatch before submission.

That explanation is irrelevant to the allocator reading both documents. What they see is inconsistency in data management, and for a GP trying to signal institutional-grade rigor, that reads as a failure.

The deeper issue is structural. Most GP teams do not pull performance figures from a single governed source. They pull from whatever model file is open, whatever quarterly report was most recently distributed, or whatever the prior DDQ used. When those sources reflect different as-of dates, the fund can submit materially different TVPI or DPI figures to two LPs in the same fundraising cycle without anyone on the IR team realizing it.

Automated LP-side scoring tools are built to catch exactly this. A figure that drifts between submissions does not require a human analyst to flag it. It surfaces algorithmically before the narrative sections are ever read, though AI hallucination risk in DDQ workflows remains a parallel concern.

Deal-Level vs. Fund-Level Metrics: What LPs Ask For and When

LPs read fund-level figures to understand aggregate performance. They read deal-level figures to understand what drove it. Sophisticated allocators ask for both, and the questions they answer are structurally different.

Fund-level metrics, Net IRR, MOIC, DPI, and TVPI, roll up across the entire portfolio to a single as-of date. Deal-level disclosures break that aggregate into its components: individual investment multiples, realized vs. unrealized splits by position, and vintage-year cohort returns that let LPs compare deployment pace against realized exit timing.

Public pension funds and endowments with quantitative manager selection processes are most likely to request deal-level granularity. They want to see whether outperformance is concentrated in one or two exits or distributed across the portfolio. A strong fund-level TVPI built on two outlier positions reads differently than the same number spread across twelve consistent exits, a core element of the DDQ consistency and quality tradeoff.

The structural problem is where these two data layers actually live. Fund-level figures typically come from the fund administrator or audited financial statements. Deal-level data lives in the portfolio monitoring model, the cap table tool, or a proprietary tracking spreadsheet maintained by the investment team and not IR. When a DDQ requires both, the IR team is coordinating across at least two separate data environments under deadline. Mismatches between fund-level totals and the sum of deal-level inputs are common, and they surface as reconciliation errors in LP submissions.

Where Performance Data Actually Lives in a GP Organization

IR analysts filling the performance section rarely own the numbers they're asked to report. Net IRR lives in a fund model maintained by finance. DPI and TVPI come from the fund administrator's quarterly package. Deal-level multiples sit in a portfolio monitoring tool or a tracking spreadsheet controlled by the investment team.

A sophisticated institutional finance visualization showing four distinct floating data vaults arranged in a circular pattern — a glowing financial model terminal, a formal fund administrator ledger binder, a portfolio monitoring dashboard screen, and a cap table spreadsheet — each connected by thin luminous lines converging toward a central point representing a unified IR workflow. The scene uses a deep navy and gold color palette with subtle geometric node patterns. Each vault radiates softly, emphasizing its independence and separation. The convergence lines are slightly misaligned, hinting at coordination complexity. Clean dark background with abstract hexagonal grid texture. No text, no labels, no letters, no numbers anywhere in the image.

By the time a DDQ arrives, the IR analyst is coordinating across three or four separate data environments, each on its own update cycle, a core problem that DDQ software for investment managers is designed to solve. The fund admin report may reflect one quarter-end while the internal model has been refreshed more recently. Bringing them into agreement under a submission deadline is where errors enter. Legacy platforms compound this by requiring manual tagging of every ingested document, a workflow that breaks the moment the person who built the tag taxonomy leaves, or when documents pile up faster than anyone can maintain them. GovernGPT autonomously ingests, tags, and maintains data across all source environments, eliminating the manual coordination burden and the keyman risk that comes with human-maintained content libraries.

How Inconsistent Performance Figures Reach LP Submissions

Three distinct failure paths produce inconsistent figures in LP submissions, and none of them require anyone to make a mistake.

A split-screen diagram showing three fragmented data streams flowing into a central submission point, rendered in a clean dark navy and gold color palette. On the left, three separate floating document stacks with subtle glow effects representing different data sources — a spreadsheet, a financial report binder, and a database server icon. Arrows from each source converge toward the center but arrive at slightly different angles, symbolizing misalignment. The center shows a glowing submission envelope with a faint warning indicator. Abstract geometric lines and nodes connect the elements, giving a sophisticated institutional finance aesthetic. No text, no labels, no letters, no numbers anywhere in the image.

The first is version drift. A fund model gets updated mid-quarter. One analyst pulls Net IRR from the refreshed file; another, working a separate DDQ concurrently, copies from a prior LP report that reflected the previous run. Both figures are technically accurate as of different points in time. Neither is flagged. Two LPs receive materially different numbers in the same fundraising cycle, undermining any effort at LP DDQ personalization at scale.

The second is source fragmentation. There is no single authoritative file. Finance owns the model. The fund admin owns the quarterly package. The investment team owns the deal-level tracker. When a DDQ requires figures from all three, the IR analyst assembles them manually, each pulled from whatever version was most recently shared. The as-of dates rarely align.

The third is the absence of a locked approved figure. Most GP organizations have no mechanism that designates a specific performance figure, tied to a specific as-of date, as the approved version for external use. Analysts default to what is available, and what is available varies by who pulls it and when.

This is not a prompting problem or a technology gap; it is an architectural one. Off-the-shelf AI tools generate responses probabilistically, meaning the same performance question asked twice can return two different figures with no flag and no mechanism for compliance teams to detect the variance. Consistent output requires deterministic architecture, not a better prompt. GovernGPT is built from the ground up to guarantee consistent responses by controlling exactly what the model sees, retiring outdated fund documents before they can surface and confirming that every answer draws from the single, version-controlled, approved source.

Best Practices for Preparing Performance Data Before DDQ Season

Before the first DDQ arrives, your data governance posture either holds or it doesn't. These practices close the gaps that create reconciliation errors and version drift under deadline pressure.

Agree on a single as-of date with finance before the DDQ window opens, and hold it across every concurrent submission. Mid-cycle figure updates that affect only some LPs are where version drift starts.
Designate one authoritative source file per fund per quarter. Finance approves it, compliance signs off, and IR pulls exclusively from that file, with no exceptions for "more recent" model runs identified after the window opens.
Pre-build a standardized performance table covering Net IRR, Gross IRR, MOIC, DPI, and TVPI in the formats your most active LPs request. Rebuilding this table from scratch under each deadline is where reconciliation errors enter.
Lock deal-level figures through the same approval chain as fund-level ones. A fund-level TVPI that doesn't align with the sum of disclosed deal-level multiples will be caught by any allocator running a structured comparison.
Route any mid-cycle metric revision through compliance before it reaches an LP submission. DDQ compliance review delays are costly, but unreviewed figures are costlier. An updated figure that hasn't cleared the approval chain is not an approved figure, regardless of its accuracy.

How GovernGPT Handles the Performance Section Without Reopening the Model

GovernGPT's dual-phase research process handles the performance section without requiring anyone to reopen the fund model. In the first phase, the system retrieves verbatim pre-approved language from prior submissions, surfacing exactly how the GP framed its performance figures for each LP historically. In the second phase, it scans ingested source documents by as-of date, detects stale quantitative figures, and flags updated values from the most recently uploaded fund admin or finance file. Refreshed data points appear in green, not as silent substitutions, so compliance reviewers see precisely what changed and from which source.

This is what distinguishes a glassbox from a blackbox. Where off-the-shelf AI silently substitutes the most statistically plausible figure (a subtle inaccuracy that reads as authoritative and passes visual review), GovernGPT's architecture controls exactly what context the model ever sees, uses verbatim pre-approved content for the vast majority of pre-population, and makes every sourcing decision fully traceable. Compliance teams are not asked to trust an output; they are shown the exact source line behind it. Where confidence falls short, the field is left blank instead of populated with a plausible but unverified figure, because a wrong performance number sent to an LP is a reputational and regulatory event, not an inconvenience of process.

Deal-level metrics sit at the boundary of GovernGPT's current quantitative automation. Qualitative text answers reach approximately 95% accuracy; deal-level figures such as net IRR, MOIC, and co-investment multiples sit closer to 80%. Where confidence falls short of a verified answer, the system leaves the field blank instead of populating a plausible but unverified figure. Color-coded traceability does the rest: blue for verbatim precedent, green for refreshed data, purple for AI-generated bridges requiring sign-off, functionality aligned with the best DDQ software for hedge funds on the market. Compliance teams see the sourcing chain for every line before the submission goes out.

Final Thoughts on Managing Fund Performance Data for DDQ Submissions

Your performance figures tell LPs how you've done. How you manage those figures tells them how you operate. Version drift, stale as-of dates, and fund-level totals that don't align with deal-level inputs are the kinds of details that surface algorithmically now, beyond due diligence calls. Getting ahead of that means owning your data governance before the DDQ window opens. GovernGPT gives your IR and compliance teams the traceability they need to submit with confidence.

FAQ

How does GovernGPT pull net IRR, MOIC, DPI, and TVPI into a DDQ without an IR analyst reopening the fund model?

GovernGPT's dual-phase process (described in the section above) retrieves verbatim pre-approved language from prior submissions and flags stale figures from the most recently ingested source documents, without reopening the fund model. Refreshed data points appear with green markers so compliance reviewers see exactly what changed and from which source before anything goes out.

Why do legacy DDQ platforms like Loopio, DiligenceVault, and Dasseti fail to deliver consistent fund metrics in the performance section?

All three are built on manually tagged, flat content libraries that store one or two canonical answer variants per question, which means they cannot hold the full version history of a performance figure across fund vintages, as-of dates, and LP channels simultaneously. When a new DDQ arrives, retrieval surfaces whatever version was tagged most recently, not the version approved for that specific LP or that specific quarter-end, and two analysts working concurrent submissions can pull from materially different snapshots with no system-level flag. The failure is architectural: a library that cannot store answer variation at scale cannot guarantee that the Net IRR your fund reported to Pension Fund A in March matches what goes to Pension Fund B in April.

What is DPI versus TVPI in a DDQ, and why do LPs focus on the gap between them?

DPI measures only realized distributions divided by paid-in capital; TVPI combines realized and unrealized value against total called capital. LPs focus on the spread between them because a high TVPI with a low DPI signals that most of the reported value is still locked in unrealized positions, a meaningful distinction for allocators comparing exit pace and liquidity risk across managers in the same vintage year.

Should I report subscription-line-adjusted and unadjusted IRR separately in my DDQ performance section?

Yes, if your fund draws on a credit facility before calling LP capital. Sophisticated allocators and investment consultants increasingly request both figures because subscription line usage compresses the apparent investment period, inflating IRR without changing underlying returns. A DDQ reporting only the headline figure without disclosing credit line usage reads as incomplete to any manager running a structured comparison. The ILPA Performance Template, now required for funds launched on or after January 1, 2026, sets a higher consistency bar on this disclosure.

Why do AI-generated DDQ answers degrade in quality for asset managers with complex multi-fund structures, and how does GovernGPT handle it?

The degradation happens because multi-entity firms carry materially different performance figures, compliance language, and deal-level metrics across business units, and a retrieval system without explicit fund-level scoping will blend answer variants from Fund A into a submission for Fund B with no warning. GovernGPT's fund-aware architecture enforces mandatory manager filter selection before any questionnaire is processed, so Fund A and Fund B operate as isolated knowledge scopes that cannot contaminate each other's answer pool. Clients report acceptance rates of approximately 85% for complex multi-strategy organizations and 90 to 95% for single-product firms with full data in the system.

Ready to see GovernGPT in action?

Book a Demo