ToltIQ

The removal test leaves the problem the product was built to solve. A virtual data room already exists and already holds the documents; what a deal team lacks is the ability to read thousands of pages of purchase agreements, quality of earnings reports, customer cohort analyses and side letters inside a deadline.

Generative models do the ingesting, categorising and extraction that turns that pile into something a person can question, and nothing else in the product survives their removal except a document viewer. The founder's own framing is that this is messy, high stakes terrain where specialised models are the right tool precisely because no two data rooms look the same.

The division of labour is stated plainly and memorably: let the machines do the sifting so humans can do the judging, with the platform described throughout as a co-pilot for investment teams rather than a decision maker. For a product whose output feeds an investment committee that is the right boundary, and putting it in one sentence is more than most vendors here manage. What is absent is everything below the headline.

No confidence indication on an extracted term, no flag when the model could not parse a document, no review step before findings reach a memo, and no account of what happens when a clause in a purchase agreement is categorised wrongly and nobody opens the original.

No accuracy figure, extraction error rate, validation evidence or model documentation was located, and the published number measures the wrong thing, since an 85 percent productivity gain describes how fast a team worked rather than whether the model read the contract correctly.

The correctness question is acute for this material because a missed change of control provision, an unnoticed side letter obligation or a misread earnout term is a live financial exposure rather than an inefficiency. Two peers show what the answer looks like, with Daloopa hyperlinking every datapoint to its source document and Nammu21 relying on the binding legal instrument being checkable against an original the vendor did not produce, and no equivalent traceability is described here.

More than 65 general partners, limited partners and family offices are stated to be using the platform, which is real adoption across three distinct buyer types rather than a single design partner. One institution is named through trade coverage, a global alternative investment manager that had the software running across three of its portfolio companies, though that was described as piloting rather than production.

Productivity gains of up to 85 percent for investment teams are claimed and carry no methodology. The founder credential is unusually specific and relevant: more than a decade as partner and chief information officer at one of the largest private equity firms, with responsibility for technology and data strategy and a seat on the operations and risk committee, which is domain authority a young company cannot manufacture. A 12 million dollar Series A was led by a fintech specialist investor that also backs another vendor in this index.

This is the sharpest unaddressed question in the profile and it is structural rather than hypothetical. In a competitive private equity process, multiple bidders review the same virtual data room for the same asset, and with more than 65 sponsors and allocators on one platform it is a matter of time before two of them are running the same deal through it simultaneously.

Nothing published describes separation between engagements, whether analysis of a target for one client informs anything surfaced to another, or whether the questions a deal team asks are visible in any form beyond that team. This is the Rogo finding, where one platform serves institutions routinely on opposite sides of a transaction, applied to an auction where the opposing sides are reading identical documents.

The personal data chain is short, since the payload is commercial documentation about companies rather than records about individuals, which is the mitigating property recorded for Nammu21 and Daloopa. What replaces it is a different and equally serious sensitivity: virtual data room contents are material non public information about a live transaction, and side letters and customer cohort data are among the most closely held documents any institution handles.

Secure ingestion is asserted and nothing describes retention after a deal closes or breaks, what happens to a target's documents when a bidder loses, or which subprocessors touch the material in transit.

No attestation, certification, trust centre or enumerated framework was located. Secure ingestion and the highest standards of accuracy and compliance are both asserted without naming a single control or independent assessment. That is the most commercially consequential gap in this profile, because a private equity firm handing a target's confidential data room to a third party will require an attestation before a single document moves, and the absence is likely to surface in every procurement conversation before price does.

No supervisor, statute, rule or guidance instrument appears anywhere, and compliance is referenced only as a standard the platform maintains. The vendor holds no licence and needs none, which is correct for a technology supplier under the index convention.

What is unaddressed is the regime the output enters: limited partners conducting operational due diligence do so under fiduciary duties, private fund advisers operate under adviser rules covering the diligence and disclosure supporting a commitment, and material non public information handling carries its own obligations for every institution on the platform. None of that is engaged in published material.

No consumer subjects and no protected class analysis applies, so the axis adapts as it did for Rogo and Rowspace. Two exposures replace it. Extraction quality will favour targets with well organised, conventionally structured data rooms, so a company with thinner or less orthodox documentation may surface more apparent risk simply because the model read it worse, and smaller or founder run businesses are the ones most likely to present that way.

Separately, a system trained on what diligence has historically flagged encodes the industry's existing assumptions about what matters, which is the institutional confirmation bias finding recorded for Rowspace and Boosted.ai, and nothing describes how the platform surfaces the risk nobody thought to look for.

No guarantee, indemnity or falsifiable accuracy commitment was located. The stated co-pilot boundary keeps an identifiable investment professional accountable for the conclusions reaching a committee, which is a real allocation, and the underlying documents remain available to check against.

Beyond that nothing exists: no correction process for an extraction found to be wrong, no notification if a categorisation error is discovered after a deal has closed, and no description of what the vendor owes a client whose diligence missed something the platform should have surfaced.

Nothing is disclosed. The product is described as generative artificial intelligence throughout and no model provider, hosting arrangement or subprocessor is named anywhere, which leaves open the question that matters most for this payload: whether a target's confidential purchase agreements and side letters are transmitted to an external model provider during analysis, and on what retention terms.

Marloo shows the standard this axis can reach, naming both its model providers and the zero retention agreements governing them, and a vendor handling live deal documentation has more reason to make that disclosure rather than less.

The product sits directly downstream of virtual data rooms and not one data room provider is named as an integration, despite that being the single connection every user needs and the market being concentrated among a handful of platforms. Expanding integrations was stated as a use of Series A proceeds, which places the integration surface in the funded roadmap rather than in the shipped product, and the Hypercore rule applies to that portion. Nothing names a deal management system, customer relationship platform or document repository on the other side either, and no developer documentation was located.

No hosting provider, region selection, residency commitment or private deployment option was located. That absence carries more weight than usual because the payload is live transaction documentation, and a sponsor's own information security team will treat where a data room copy rests as a deal risk rather than a procurement detail. Nothing indicates whether a customer can constrain processing location or whether documents are held in a dedicated environment.

No pricing, packaging or basis of charge is published and every route in is a contact request. The unit question is live for this product because diligence is episodic rather than continuous, so whether a firm pays per deal, per seat, per data room or as an annual platform fee changes the economics entirely for a mid market sponsor doing a handful of processes a year against a large one running dozens. Nothing addresses it.

Three distinct buyer types are served and they sit on different sides of the same market: general partners evaluating acquisitions, limited partners assessing managers and commitments, and family offices doing both, with diligence advisory firms as a fourth. That spread matters because the same document analysis serves a buyout team reading a target's contracts and an allocator reading a fund's side letters.

The limits are function and geography: this is deal diligence rather than portfolio monitoring, valuation or fund operations, and the evidenced footprint is North American with no international presence demonstrated.