How AI is Transforming Due Diligence for Funders

Due diligence is the part of grantmaking that nobody enjoys but everybody needs. Before awarding a grant, funders must confirm that the applicant organisation is registered, governed properly, financially viable, and capable of delivering the work. Traditionally, this has meant a grants officer spending two to four hours per applicant: searching registers, downloading accounts, reading governance documents, cross-referencing trustee lists, and compiling notes in spreadsheets or emails.

The problem is not the thoroughness. The problem is that this manual approach does not scale. UK charitable foundations increased their grant-making to a record 8.24 billion pounds in 2023-24, with the largest 200 ACF members alone distributing 6.1 billion pounds, a 12 percent increase on the previous year (ACF, Foundations in Focus 2025). More money flowing through more applications means more due diligence work, and most grantmaking teams have not grown to match.

AI is changing this by automating the routine lookups and document analysis that consume the bulk of due diligence time, while keeping humans in control of interpretation and final decisions. This guide explains what AI-powered due diligence actually involves, where it helps most, and where human judgement remains essential.

What does AI-powered due diligence actually check?

AI-powered due diligence is not a black box that produces a pass or fail score. It is a set of automated checks against trusted data sources, combined with document analysis, that produces a structured report for human review.

In practice, the checks fall into several categories:

Register lookups. The system queries whichever national register governs the applicant — the Charity Commission and Companies House for an organisation in England and Wales, OSCR in Scotland, the IRS Form 990 series for a US nonprofit — to verify registration status, trustee and director information, filing history, and any regulatory flags. These are the same registers a grants officer would check manually, but the system does it in seconds rather than minutes per register, and it does not need the applicant to be in the same country as the funder.

Sanctions screening. The system screens the applicant organisation and each named director against the US Consolidated Screening List, which merges the OFAC Specially Designated Nationals list, the Entity List and the other federal restricted-party lists into one source. Names are screened individually rather than in a batch, because a fuzzy match on one director is exactly the signal that gets lost when names are run together. Matches are surfaced for human review rather than acted on automatically: false positives are common on fuzzy name matching, and a sanctions hit is never a decision a system should make on its own.

Financial history, not a single year. A registration check tells you an organisation exists. The filings tell you how it has been doing. The 990 series carries several years of revenue, expenses, assets, liabilities and net assets, from which months of cash, programme-expense ratio and fundraising efficiency follow directly — so the question shifts from "is this a real nonprofit" to "is this organisation's position improving or deteriorating, and how fast." A single year cannot answer that; three can.

Relationships, not just the organisation in front of you. The most useful facts about an applicant are often about its position in a field rather than its own record:

• Co-funders. Which other funders already support this organisation, and how often those funders appear alongside you elsewhere. Derived by grouping every grant in the corpus by recipient, so it is computed rather than looked up.
• Related and affiliated entities. Whether the applicant is a standalone organisation or one filer within a larger system — affiliated entities filing separately under shared control, detected as a family and required to be corroborated before two filers are treated as one. That changes what the financials mean (a family's combined giving is not the sum of its filers counted separately) and whether two applications in front of you are really one applicant.
• Related-party transactions. What the filing itself discloses about business dealings with insiders, which is the disclosure a conflicts check is actually looking for.
• Governance overlap. Where an applicant's board overlaps with organisations already in your portfolio — a conflict-of-interest signal you are obliged to notice. Person-level detail sits behind authentication rather than on any public page.

Document analysis. This is where AI adds the most value beyond simple lookups. The system reads uploaded governance documents, safeguarding policies, equality and diversity policies, insurance certificates, accounts, and bank statements. Rather than simply confirming a document exists, AI analyses the content for specific indicators: whether a safeguarding policy names a designated lead, whether a governing document includes a dissolution clause, whether accounts show declining income or negative reserves.

Financial health assessment. AI analyses submitted accounts to identify income trends, reserve levels, income concentration risks, and any discrepancies between filed and claimed figures. It flags specific concerns with numbers rather than vague warnings.

Why manual due diligence struggles at scale

Manual due diligence works well for small programmes processing a handful of applications. It breaks down when volume increases. According to Xapien, traditional donor or applicant research can take around eight hours per subject, with full reports often taking up to two days to compile to the expected standard.

For a funder receiving 200 applications per round, that represents 400 to 800 hours of staff time on lookups alone. The consequences are predictable:

Inconsistency. Two grants officers checking the same organisation will not necessarily examine the same things in the same depth. One might scrutinise Companies House filings closely; another might skip them for registered charities. Without rigid protocols, the quality of due diligence varies by reviewer.

Fatigue errors. When a grants officer is processing their twentieth application in a week, the quality of their checks declines. This is not a reflection of competence but a reality of repetitive cognitive work. The Charity Commission concluded 4,375 regulatory concern cases in 2024-25, an 18% increase on the previous year (Charity Commission Annual Report 2024-25), and many of the underlying issues could have been caught earlier through consistent checking.

Decision delays. Applications cannot progress to panel until checks are complete. When checks take hours each, funders either delay decisions or pressure staff to cut corners. Neither outcome serves applicants or funders well.

Fragmented records. Manual checks often produce notes in emails, Word documents, and personal spreadsheets. When a trustee or regulator asks what due diligence was performed on a specific grant, reconstructing the answer from these fragments is difficult.

How AI reads governance and policy documents

One of the most time-consuming elements of due diligence is reading and assessing uploaded documents. A grants officer might spend 30 minutes reading a governing document to determine whether it includes the right clauses, or 20 minutes checking whether a safeguarding policy names a designated lead and references current legislation.

AI document analysis transforms this process. For governance documents, the system assesses:

• Whether the legal type of the organisation is appropriate
• Whether the document is up to date, including provisions for remote board meetings (increasingly expected since 2020)
• Whether the document includes a dissolution clause and asset lock
• Whether there is a conflict of interest policy
• Whether the board size and procedures are appropriate
• Whether the organisation name and charity number in the document match registry records

For safeguarding policies, the system checks for a named designated safeguarding lead with contact details, references to DBS checks (flagging outdated "CRB" language), coverage of online safety, and regular review dates.

For equality and diversity policies, the system verifies reference to the Equality Act 2010, coverage of the nine protected characteristics, and provisions for preventing harassment and discrimination.

For accounts, the analysis goes deeper: identifying income sources and concentration risks, checking for ongoing deficits or excessive surpluses, reviewing reserve levels relative to expenditure, and noting whether accounts were prepared by an independent examiner or auditor as required.

Each analysis produces specific findings with supporting evidence rather than a simple pass or fail, giving the reviewing grants officer the information they need to exercise their own judgement.

Manual vs. AI-assisted due diligence: a comparison

The important distinction is that AI-assisted due diligence does not eliminate human involvement. It shifts human effort from mechanical lookups to interpretation and decision-making, where professional judgement matters most.

What proportionate due diligence looks like with AI

Good due diligence is proportionate. A two-thousand-pound grant to a well-established local charity does not need the same depth of checking as a two-hundred-thousand-pound grant to a newly formed organisation. IVAR's Open and Trusting Grant-making initiative has encouraged funders to take a collaborative, proportionate approach to due diligence rather than applying one-size-fits-all requirements.

AI makes proportionality easier to implement rather than harder. With automated baseline checks, funders can:

Apply consistent minimum checks to every applicant. Register verification, sanctions screening, and basic document presence checks can run automatically regardless of grant size. This ensures no application bypasses fundamental compliance requirements.

Tier deeper analysis by risk. For larger or higher-risk grants, AI can perform more detailed document analysis, including reading governance documents for specific clauses, analysing accounts in depth, and cross-referencing information across multiple documents. For micro-grants, a lighter set of checks may be sufficient.

Use findings to guide manual investigation. When automated checks flag a concern, such as late filings at Companies House or a safeguarding policy that lacks a named lead, the grants officer can focus their manual time on those specific issues rather than spending equal time on every application.

Record the rationale. The system logs not just what was checked but also which tier of checks was applied and why, creating a clear audit trail that demonstrates proportionality to trustees and regulators.

This approach aligns with the Charity Commission's own guidance, which emphasises risk-based regulation. Over 42 percent of charities had expenditure exceeding their income in 2023 (Charity Commission Annual Report 2023-24), highlighting the importance of financial health checks while recognising that not every applicant carries the same level of risk.

The role of public registers and data sources

AI-powered due diligence is only as reliable as its data sources, and those sources are national. A funder checking an applicant in Ohio and a funder checking one in Ayrshire are asking the same questions of completely different systems, which is why register coverage — how many countries a platform can actually check, and how much each of those registers gives up — matters more than any single feature.

IRS Form 990 (United States). The 990 series is the deepest financial record in public charity data anywhere: full revenue and expense detail, total assets and liabilities, net assets, filing history, and voting and independent board member counts, from which months of cash, programme-expense ratio and fundraising efficiency can all be computed directly. Board members are counted rather than named, so identifying individual directors still needs another source.

Charity Commission for England and Wales. The register contains details on registered charities including their objects, trustees, filing history, financial data, and any regulatory actions. Over 5,000 new charities were registered in the 2024-25 financial year (Charity Commission), and the register provides an API for automated access.

Companies House. For organisations registered as companies (including CICs and CLGs), Companies House provides director information, filing history, accounts data, and company status. Its public API enables automated lookups.

OSCR (Scotland). The Office of the Scottish Charity Regulator maintains a separate register for Scottish charities, including financial history and regulatory status. Data can be accessed for automated screening.

Registers vary enormously in what they will settle. Some publish current standing, the people responsible, and full accounts; some confirm only that a body exists; a few carry detail but will not say whether the organisation is still live — which is the one question a funder most needs answered. A platform that treats every register as equivalent is overstating what it knows. Plinth's registry coverage map sets out all 51 countries it can check, ranked by how much of a funding decision each register can settle on its own.

The quality of AI-powered due diligence depends on connecting to these sources accurately and keeping results current. Systems that only check one register leave gaps. For example, a charity registered with the Charity Commission may also be a company registered at Companies House, and cross-referencing both can reveal discrepancies that checking either alone would miss.

Where human judgement remains essential

AI handles routine lookups and document analysis well. It does not replace the need for human judgement in several critical areas.

Complex governance structures. Organisations with unusual structures such as fiscal sponsorship arrangements, unincorporated associations, partnerships between multiple entities, or overseas parent bodies require human interpretation. AI can flag these as unusual, but assessing whether the structure is appropriate requires contextual understanding.

New organisations. A charity registered three months ago will have no filed accounts. An unregistered community group will not appear on any register at all. Due diligence for these applicants relies on other evidence, including references, individual track records, and project plans, which require human assessment.

Sensitive assessments. Some due diligence involves judgement calls that should not be delegated to automation. Evaluating the quality of a safeguarding framework beyond its mere existence, assessing an organisation's approach to sensitive issues, or considering reputational risks not captured in registry data are all areas where human expertise is irreplaceable.

Relationship context. Experienced grants officers bring knowledge about the local landscape, relationships between organisations, and historical context that no automated system can replicate. A funder who has worked with an applicant over several years brings valuable intelligence to the due diligence process.

The 2024 Charity Digital Skills Report found that 61 percent of UK charities are now using AI in their operations (Charity Digital Skills Report 2024), but 62 percent still want to develop a better understanding of AI and its uses. This gap between adoption and confidence reinforces why human oversight remains essential: the technology should support professional judgement, not replace it.

How Plinth approaches AI due diligence

Plinth implements AI-powered due diligence as an integrated part of the grant management workflow rather than a standalone screening tool.

When an application is submitted, Plinth performs automated checks against the register that governs the applicant, drawing on 55 register adapters across 51 countries — the IRS 990 series for a US nonprofit, the Charity Commission and Companies House for an English or Welsh charity, OSCR for a Scottish one. Where an organisation appears in more than one register, they are cross-referenced automatically: a charity that is also a registered company is checked in both, and discrepancies between the two are flagged.

Checks are organised into seven assessment pillars — legal standing, governance, financial management, track record and reputation, programme and technical capacity, compliance and sanctions, and safeguarding — and a funder can switch off any pillar that is not proportionate for a given fund. A micro-grants programme need not run the full set; a capital grant should.

Where no register can answer the question — a country with no public charity register, or an unregistered community group anywhere — the system falls back to AI web research and sanctions screening rather than returning nothing, and says plainly which of the two produced each finding.

For document analysis, Plinth reads uploaded governance documents, safeguarding policies, equality and diversity policies, insurance certificates, accounts, bank statements, and budgets. Each document type is assessed against a specific set of criteria. For example, governance documents are checked for dissolution clauses, conflict of interest provisions, board composition, and consistency with registry data. Accounts are analysed for income trends, reserve levels, income concentration, and discrepancies with filed data.

The system produces a structured report for each category, with a plain-text summary and any issues flagged by severity (medium or high). This gives the reviewing grants officer a clear starting point: they can see at a glance which areas need attention and focus their manual review accordingly.

Critically, Plinth does not make approval or rejection decisions. The AI analyses and flags; the human reviews and decides. Every check is logged with a timestamp and evidence source, creating a complete audit trail. Results can be exported to a Word document for sharing with panels, trustees, or external assessors.

Plinth also supports proportionate checking by allowing funders to configure which checks are run for different programmes or grant sizes. A micro-grants programme might use only register lookups and basic document presence checks, while a capital grants programme might run the full suite of document analysis.

Plinth's entry tier is priced to put AI-assisted due diligence within reach of smaller funders without a large upfront investment — see pricing for current tiers in your region.

Getting started: a practical approach

Implementing AI-powered due diligence does not require a wholesale transformation of your grantmaking processes. Most funders find it practical to start small and expand:

1. Audit your current process. Document what checks you currently perform, how long they take, and where inconsistencies or bottlenecks exist. This gives you a baseline to measure improvement against.

2. Start with register checks. Automated lookups against the applicant's own national register are the lowest-risk, highest-return starting point. They replace manual typing and searching with instant cross-referencing.

3. Add document analysis for your highest-volume programme. Choose one programme where volume creates the most pressure and pilot AI document analysis for governance and safeguarding policies. Compare the AI findings to manual review to build confidence.

4. Review and refine. Examine the automated results critically. Are there false positives? Are genuine issues being flagged? Use the findings to calibrate the system and your internal protocols.

5. Expand across programmes. Once you are confident in the automated checks, extend them across your full portfolio. Configure different check levels for different programmes based on risk and grant size.

6. Train your team. Ensure grants officers understand what the automated checks do and do not cover, how to interpret the results, and when to conduct additional manual investigation.

The goal is not to automate everything but to automate the parts that consume disproportionate time relative to the insight they generate, freeing your team to focus on the analysis and judgement that only humans can provide.

Frequently asked questions

Can AI due diligence completely replace manual checks?

No. AI handles routine lookups and document analysis effectively, but human judgement is essential for interpreting findings, assessing complex governance structures, evaluating new organisations with limited public records, and making proportionality decisions. The best approach is human-in-the-loop: AI performs the checks, humans make the decisions.

What data sources does AI due diligence use?

AI due diligence systems check the register that governs the applicant — the IRS Form 990 series in the US, the Charity Commission and Companies House in England and Wales, OSCR in Scotland — plus sanctions screening for the organisation and its directors. Some systems also analyse uploaded documents such as governance documents, safeguarding policies, accounts, and insurance certificates. Coverage is the thing to compare: a system that checks one country will return nothing for an applicant based anywhere else.

Is automated due diligence compliant with GDPR?

Yes, provided it is implemented correctly. Register lookups use publicly available data. Where applicant-submitted documents are processed, the system must handle them in accordance with GDPR requirements, with appropriate data processing agreements, purpose limitation, and data retention policies in place.

How long does AI due diligence take compared to manual checks?

Manual due diligence typically takes two to four hours per applicant. AI-powered checks complete register lookups in seconds and document analysis in minutes. The human review of the AI-generated report then takes a fraction of the time that a full manual check would require, because the grants officer is reviewing compiled findings rather than performing raw research.

What happens when AI flags a concern?

The AI produces a structured report with findings categorised by severity. Medium and high severity issues are highlighted for the grants officer, who then investigates further as needed. The system does not reject applicants; it surfaces concerns for human review and decision-making.

Can we customise which checks are performed?

Yes. Systems like Plinth allow funders to configure checks by programme and risk profile. Micro-grant programmes might use only register lookups, while larger programmes might include full document analysis across governance, safeguarding, accounts, and insurance.

Does this work for unregistered community groups?

Automated register checks will not return results for organisations that appear in no public register, or that are based in a country with no register to check. In these cases, due diligence relies more heavily on uploaded documents, references, and manual assessment. AI can still analyse submitted documents, but the register cross-referencing element will be limited.

How do we explain AI-assisted due diligence to our board?

Frame it as a tool that improves consistency and frees staff for higher-value work. Every check produces a timestamped audit trail, so trustees can see exactly what was examined and what was found. The system supports human decision-making rather than replacing it, which addresses common governance concerns about algorithmic decision-making.

Recommended next pages

• How to Automate Due Diligence in Grantmaking -- A step-by-step guide to implementing automated checks in your grantmaking workflow
• Manual vs. Automated Due Diligence -- A detailed side-by-side comparison of both approaches with cost analysis
• What is a Due Diligence Check? -- Fundamentals for teams new to the due diligence process
• Human-in-the-Loop Grantmaking -- Why keeping humans central to AI-assisted decisions matters
• AI for Funders: The Future of Grantmaking -- The broader landscape of AI applications in philanthropy

Last updated: February 2026