The DD Architecture
How The Startup Mentor™ structures investment due diligence — what each stage of the funnel represents, what each DD activity does to a company at that stage, where the DD pipeline sits within the broader deal pipeline, and how TSM acts as connective tissue across the existing AI stack.
1.The picture
Every investor — angel, fund, family office, accelerator — runs the same selection pipeline whether they articulate it or not. A company starts in a wide universe of candidates, gets narrowed by Policy DD (the activity most investors simply call screening) to a long list of policy-matched companies. Conventional practice runs that long list through a thin filter — pitch decks at best — and reserves real assessment for the short list. The Startup Mentor™ adds a step in between. Public DD assesses every company on the long list against the same 22 pillars, 5 evidence levels and 6 readiness gates that conventional firms only apply much later. The output is the Assessed List — and it is the architectural wedge that the rest of the pipeline depends on.
After the Assessed List, the funnel proceeds through Private DD (pitch deck review), Confidential DD (data room review under NDA), Vetting DD (independent field verification), and Portfolio-Level DD (decision support and ongoing monitoring). Six DD activities span the funnel; the survivors of all six enter the portfolio for active management.
The funnel geometry is not decorative. As evidence grows (the disks on the right enlarge), selection narrows (the trapezoids on the left taper). These are two faces of the same underlying mechanic: depth and volume trade off. A long-list assessment running against a hundred companies cannot reach the depth of a vetting assessment running against two — there is not enough investor time, not enough company access, not enough signal-to-noise. Every stage of the funnel buys deeper evidence at the cost of fewer companies, and every disk visualises the depth available at that stage. The two trends are inverse by construction; the diagram is honest about that constraint rather than pretending the same depth is reachable everywhere.
Selection narrows significantly at each stage in a well-run process. Evidence depth grows from public-only (Long List → Assessed List) to public + private (Assessed → Short) to public + private + confidential (Short → Vet) to public + private + confidential + field (Vet → IC). The funnel and the disks tell that story together.
Universe rule. For investment cycles, the universe is the market minus the firm's existing portfolio. For re-investment and divestment cycles, the universe is the portfolio — the firm's held positions become the candidate set for follow-on, exit, or write-down decisions, flowing back through the same six-DD funnel.
2.The broader deal pipeline
The DD pipeline is one part — the central, selection-bearing part — of the broader deal pipeline that every investor runs. The broader pipeline includes activities that come before DD (sourcing and origination), activities that interleave with the late stages of DD (negotiation and closing), and activities that follow once the company is in the portfolio (active management and exit). The DD pipeline does not replace any of those; it occupies the middle of the workflow and feeds them.
| Phase | What happens | Relation to the DD pipeline |
|---|---|---|
| Sourcing / Origination | Generating deal flow — networks, conferences, inbound platform submissions, outbound research, accelerator demo days, founder referrals. | Fills the Universe. Out of TSM scope; sourcing is the firm's relational and brand work. |
| Initial contact | First meeting, intro call, expression of mutual interest. Often surfaces companies that already passed informal Policy DD. | Sits between Sourcing and the Universe entering the funnel. |
| Policy DD (Screening) | Filtering against sector, stage, geography, ticket size — the firm's basic eligibility rules. | First DD activity in the funnel. Out of TSM scope — the policy belongs to the investor; the screening tools that apply it already exist at industrial scale. |
| Public DD | Public-information assessment of every company on the long list. New step TSM introduces. | Produces the Assessed List. The TSM value-add wedge. |
| Private DD | Review of pitch deck and other materials shared in confidence — typically without formal NDA. | Produces the Short List. |
| Confidential DD | Data room review under NDA. | Produces the Vet List. |
| Vetting DD | Independent field verification — customer references, legal, technical, expert calls. | Produces the IC List. |
| Negotiation | Term sheet, valuation negotiation, structure, board composition, governance terms. | Interleaves with the IC List → Portfolio transition. Out of TSM scope but informed by the company data file the funnel produces. |
| Closing | Legal documentation, regulatory clearances, capital call, wire transfer, cap-table update. | Sits at the IC List → Portfolio gate. Out of TSM scope; transactional rather than evaluative. |
| Portfolio-Level DD | Decision support pre-close, then ongoing monitoring, board contributions, follow-on / divestment recommendations. | Operates inside the Portfolio collection. In TSM scope. Continues using the same company data file built across the earlier DDs. |
| Active management | Hands-on partnership work — strategic guidance, executive hiring, customer introductions, follow-on capital. | Runs continuously inside the Portfolio. Portfolio-Level DD informs it. |
| Exit | IPO, strategic M&A, secondary sale, write-off, structured wind-down. | Runs after Portfolio. Re-investment and divestment cycles flow back through the funnel via the universe rule. |
TSM is not asking an investor to adopt a new pipeline architecture. The seven-collection funnel is the same set of stages every investor runs, however they are labelled internally — top of funnel, active pipeline, IC review, closed, active management. What TSM provides is the assessment layer that sits on top of those stages: a unified evidence vocabulary applied consistently from Public DD through Portfolio-Level DD, against axes the firm doesn't have to invent. Existing point tools feed into TSM; TSM rolls their outputs up into the same structured record. The workflow doesn't change. The vocabulary unifies across stages the firm already runs.
3.Seven collections
Collections differ in two dimensions: volume (how many companies are in the set) and depth of evidence (how much is known about each company). The funnel narrows because both dimensions move in opposite directions: as the volume drops, the depth of evidence rises.
| Collection | What it is | Typical population |
|---|---|---|
| Universe | Every company that could conceivably be relevant — sourced from data platforms, warm intros, scouts, deal-flow networks, conferences, public registries. | Thousands to tens of thousands. |
| Long List | Companies that have passed the policy filter — they fit the thesis, the stage, the geography, the sector. Eligible for assessment. | Tens to a few hundred per quarter for an active fund. |
| Assessed List | Companies on the long list that have received a TSM Public DD — full 22-pillar diagnostic on public information alone. New step TSM introduces. | Same population as the long list, but now structurally assessed. |
| Short List | Companies that have passed pitch-deck review (Private DD) — the thesis holds against the founder's own narrative. | Single digits to low tens per quarter. |
| Vet List | Companies whose data-room review has held up and that are undergoing field verification. | One or two at a time per partner. |
| IC List | Companies whose vetting is complete and whose investor memo is ready for IC review. | Whatever is on this week's IC agenda. |
| Portfolio | Companies the firm has invested in. Active management responsibility. For re-investment / divestment cycles, the portfolio is the universe. | 15–40 for an early-stage fund. |
The funnel narrows because evaluation deepens. A long-list assessment running against a hundred companies cannot reach the depth of a vetting assessment running against two — there is not enough investor time, not enough company access, not enough signal-to-noise. The funnel's geometry isn't a marketing flourish. It is a consequence of how attention and access scale.
The generic TSM framework is policy-agnostic by construction — it doesn't know any specific investor's thesis. Configured for a specific investor (the policy encoded as criteria on top of the framework — sector, stage, geography, ticket size, hold-period preferences, value-add capability constraints), it becomes policy-aware. Portfolio-Level DD then monitors each company against both its operational dimensions and its policy fit, and flags drift in either for IC review. Sector pivot, stage mismatch, hold-period overrun, value-add capability becoming irrelevant — all become structural signals rather than partner intuition. The flag is automatic; the decision remains the IC's.
4.Six DD activities
DD is often framed as judging whether a company is good or not — as if there's an absolute scale and the investor's job is to read it correctly. That framing is wrong. The question DD answers is not is this company good but is this company good for this investor — does it match the investor's thesis, value-add capability, ticket size, hold period and portfolio shape? A turnaround investor selects for distress. A deficient balance sheet is the criterion that gets a company onto their long list, not the criterion that knocks it off. Many growth-stage operator-investors specialise in helping companies that are deficient in one or more areas and select for companies missing the layer they specialise in providing. The same company can be a clear yes for one investor and a clear no for another, and both can be right.
What TSM provides is a 37-dimension surface (22 elemental + 15 composite) that makes the matching legible — what's strong, what's weak, what evidence supports each claim. What the dimensions mean for advance / no-advance is the investor's call, not the framework's. The framework doesn't have a thesis. The investor does.
| DD activity | Acts on | Decision it informs |
|---|---|---|
| Policy DD | Out of TSM scope. The universe of available companies, filtered through the investor's policy on sector, stage, geography and ticket size. Tools used: PitchBook, Crunchbase, Dealroom, Harmonic, Affinity, warm-intro flow. | Does this company match the investor's policy? Companies that pass enter the long list with the policy match recorded as the foundation. |
| Public DD | A company on the long list. Inputs limited to public sources: website, regulatory filings, news, founder LinkedIn, sector reports, market research, patents. The new step TSM introduces. | Should this company advance to a pitch request and the time investment of an investor meeting? |
| Private DD | A company on the assessed list. Adds the pitch deck and other materials the founder has shared in confidence — typically without formal NDA. | Should this company advance to a data-room request? The Short List is the cohort the investor is committing to deeper diligence on. |
| Confidential DD | A company on the short list. Adds the data room under NDA — financial model, cap table, customer pipeline, signed contracts, IP filings, technical specifications, board reports. | Should this company advance to independent field verification? |
| Vetting DD | A company on the vet list. Adds field-level evidence — customer references, legal review, technical review, expert calls. The only layer the founder cannot curate. Produces the investor memo. | Should this company go to the IC? |
| Portfolio-Level DD | Companies on the IC List (pre-close decision support) and companies in the Portfolio (post-close ongoing monitoring). Re-runs the framework against fresh evidence on a schedule. | Approve / kill / follow-on at IC. Then on a continuing basis: hold / increase / reduce / exit. The IC vote remains the investor's prerogative. |
A capability worth calling out separately, because it is structurally different from how stand-alone pitch review tools work. Most pitch review tools assess the deck on its own terms — clarity of narrative, slide-by-slide structure, presence of expected sections. They have no view of the company beyond what the deck itself says. The reviewer's only reference point is the pitch, so the only feedback available is about the pitch.
TSM's pitch review is different by construction. By the time Private DD reads the deck, TSM already holds a Public DD-level assessment of the company — 22 elementals, 15 composites, evidence-graded, with red flags, green flags, gates and valuation already computed. The deck is then read against that assessment, not in isolation. What TSM detects is the gap between what the company actually is (per the assessment) and what the deck claims it is. Where the deck overclaims relative to the evidence, the system flags it. Where the deck understates a real strength the public record supports, the system flags that too. Where the deck omits a material fact — competitive entrant, regulatory exposure, founder background gap, tarpit-pattern risk — the system surfaces it.
Pitch review without an underlying assessment is a stylistic exercise. The reviewer can comment on whether the deck reads well, whether the slides are in conventional order, whether the financial slide has the conventional numbers. They cannot tell whether the deck is accurate — whether the claims hold up against what the company is actually doing — because they have no independent view of the company. TSM's pitch review is the opposite: the company is assessed first, the deck is read against that assessment second, and the output identifies where the founder's narrative and the structural reality diverge. That divergence is the most decision-relevant signal a buyer can get from a deck.
5.Four evidence tiers
Each DD activity adds exactly one tier of evidence to the company data file. The four tiers are gated by access — what the investor can see at each stage of the relationship with the founder. The tier structure is what makes the funnel architecture work: each DD is named after the tier it adds, and each disk's colour signals that tier.
| Tier | What it includes | Where it becomes available |
|---|---|---|
| Public | Anything in the public record: website, blog, press, regulatory filings, patents, founder LinkedIn, sector reports, market data, competitor analysis, news. | Available from the start. Free to access, no founder coordination required, scalable to dozens of companies. The default starting point for any assessment. |
| Private | Materials the founder shares in confidence without formal NDA — typically the pitch deck, plus follow-up email exchanges, founder narrative, light financial projections. | Available from the Assessed List → Short List transition. The founder shares because the investor has signalled enough interest to warrant the time, but neither side wants to slow things down with paperwork yet. |
| Confidential | Data room contents under NDA — financial model, cap table, customer pipeline, signed contracts, IP filings, technical specifications, board reports, internal memos. | Available from the Short List → Vet List transition. The NDA marks a step-change in mutual commitment. |
| Field | Evidence the investor gathers independently — customer reference calls, legal review, technical review, expert interviews, on-site visits, hands-on product evaluation. The only tier the founder cannot curate. | Available from the Vet List → IC List transition. The most expensive tier; reserved for serious candidates because of cost and because customer outreach signals near-investment status. |
Public DD is sometimes dismissed as shallow because it lacks confidential inputs. That dismissal misreads the architecture. Public evidence is foundational — every later DD builds on it rather than replacing it. The question Private DD asks is not whether the pitch deck contradicts the public record (rare), but whether it adds enough beyond what was already known to justify advancing. The question Confidential DD asks is whether the data room confirms what the pitch deck claimed. The question Vetting DD asks is whether independent field evidence corroborates what the confidential data implied. Each tier tests the previous one rather than discarding it.
This is also why running real depth at the Public DD stage matters. A short Public DD is not a reduced version of a fuller assessment — it is the assessment the rest of the pipeline conditions on. If the Public DD is shallow, the Private DD asks worse questions, the Confidential DD looks at the wrong evidence, and the Vetting DD verifies things that didn't need verifying. Conventional firms that defer real assessment to the short-list stage carry forward less structure than firms that do the work earlier.
The point isn't that more tiers is better. Each DD uses exactly the evidence appropriate to its stage and its decision. Forcing Field evidence on Public DD would make it unrunnable at scale. Forcing Field evidence on Portfolio-Level DD on a quarterly schedule would make it impossibly expensive and still wouldn't refresh — customer reference calls don't get re-done quarterly. Periodic monitoring re-runs the framework against the evidence that does refresh: public news, board reports, the financial model as updated, the cap table, the KPI dashboard.
6.One company data file
The architectural claim that gives the system its leverage is that all TSM DDs — Public DD, Private DD, Confidential DD, Vetting DD, Portfolio-Level DD — read from and write to the same company data file. Policy DD precedes the file's existence; the policy match is recorded as the foundation when the file is first created. A company data file is a structured record of every score, evidence reference, gate result, valuation, red flag, green flag, and homework item the engine has ever observed about a single company.
When Public DD assesses a company, it writes a company data file populated with public-tier evidence on top of the policy match foundation. When the company advances to the short list, Private DD reads that same file and enriches it with the founder's pitch and supporting materials. Scores rise where new evidence warrants. Evidence levels jump from E1 (founder assertion) to E2 (internal documentation) where the pitch deck confirms what was previously assumed. The file is not rebuilt — it is deepened.
When the company advances to Confidential DD, the data room provides a step-change: financial model, cap table, customer pipeline, contracts. Evidence levels jump again, this time from E2 to E3 or E4 where independent third-party signal exists in the confidential materials. Vetting DD then adds field-level evidence, and the investor memo is rendered from this combined state. Portfolio-Level DD reads the memo and the file together for IC support, then continues reading the same file once the company is in the portfolio, re-running against fresh evidence on a schedule. Every monitoring cycle is structurally comparable to the last because the framework is identical and the company's prior state is preserved. Drift is measurable. Trajectory is measurable. Cohort patterns become measurable across the firm's portfolio.
The structure of the file is what makes it stable across stages. Each company is scored on the same axes, evidence-graded against the same scale, gated against the same readiness thresholds, and valued through the same method stack.
Each new DD activity adds evidence on top of this same structure. The data file deepens; the schema does not change. That stability is what makes Public DD's customer-definition score comparable to Vetting DD's — the dimension is the same; what's different is the evidence tier behind it.
7.The existing AI stack
A reasonable question: haven't VC and PE houses already automated most of this? Affinity reports that 85% of private-capital dealmakers now use AI in their daily workflow. Sourcing runs through Harmonic, Specter and PitchBook. Relationship intelligence through Affinity. Data-room extraction through Hebbia, V7 Go and AlphaSense. IC-memo drafting through Rogo, V7 Go, Lyzr and WorkWise. Meeting capture through Granola and Jamie. Portfolio monitoring through Standard Metrics, Visible and Chronograph. The labour at every stage has tooling around it. What none of it does is connect the stages.
The associate produces one memo for the long list. A few weeks later a different associate with the same tools writes a different short-list memo with a different structure. A third stitches together legal, technical and reference reports into the IC memo. Six months after the deal closes a fourth writes a portfolio refresh that doesn't reference any of the previous three. Each document may have been drafted with strong AI assistance. None of them shares a spine with the others.
The gap is structural, not technical. Each AI tool in the stack does its step well. None of them produces output that the next tool can read as structured input. The associate is the integration layer — manually rewriting findings into the next memo's format, manually deciding what to carry forward and what to drop. Integration is an architectural property, not a deliverable. It can't be assembled retrospectively from well-drafted memos that didn't share a spine when they were written. The fix isn't more tools or better prompts; it's a layer that the tools feed into.
And yes — Affinity and DealCloud already maintain structured records across the deal lifecycle. TSM doesn't compete with the CRM. The continuity TSM offers isn't relational schema; it's content continuity — the same 22 dimensions, evidence grades and gates carried across stages. The CRM tracks who, when and how much. TSM tracks what changed in the assessment, against what evidence, and whether the company is structurally readier than the last time anyone looked. The two layers stack.
Two things happen at once. The automation is the visible saving. Assessment work that consumes analyst hours — or doesn't get done at all because it's too expensive per company — gets compressed into hours-not-weeks per company. A Public DD that would cost a fund €15–25K of analyst engagement runs at €2,500 of TSM-priced work; a Portfolio-Level monitoring cycle that would cost a partner days of focused review runs in minutes against the same data file. That's the part of the pitch a CFO can quantify.
The clarification is the bigger structural claim. Most investors do not run the same assessment vocabulary across their deal pipeline stages. The Long-List screen uses one set of criteria, the Short-List memo uses another, the IC pre-read uses a third, the post-investment update uses a fourth. Each stage produces a document; the documents don't speak to each other. TSM's claim is that the assessment vocabulary should be constant across the deal pipeline. Same 22 elementals, same 15 composites, same E1–E5 evidence scale, same 6 readiness gates, regardless of stage. What changes between stages is evidence depth, not the framework. The fragmentation that makes the pipeline impossible to read longitudinally is removed because the vocabulary is shared.
Specific API-level connectors to Hebbia, Rogo, V7 Go and the rest don't yet exist — but every tool in the stack already produces human-readable output (memos, transcripts, extracted-fact summaries, KPI exports) that TSM can ingest directly. The firm doesn't wait for integration engineering to start getting value; they paste, drop, or upload the output and TSM decomposes it against the framework. API connectors are polish that improves throughput and reduces friction over time. They are not the gating step.
8.Scope
The vertical bracket on the left of the diagram makes the scope question explicit. Policy DD — the universe-to-long-list filter — is out of TSM scope. Public DD, Private DD, Confidential DD, Vetting DD and Portfolio-Level DD are in TSM scope. Within Portfolio-Level DD, the IC's actual decision (the partner vote) is the investor's prerogative, but the structured input the IC reads and the structured monitoring that follows are entirely TSM.
Policy DD — what most investors simply call screening — operates at the universe level, sifting thousands of companies almost all of which will be discarded. It rewards breadth, speed, and signal aggregation. The right tools for it are data platforms like PitchBook, Crunchbase, Dealroom, Harmonic. None of these tools are methodology-bound. They scale by ingesting more data, not by deepening evaluation. And the policy itself — sector, stage, geography, ticket size — belongs to the investor. TSM has no business writing the investor's thesis.
The IC's decision is a partner-level human judgement. Even with a perfect investor memo, the call depends on the partnership's risk appetite, the fund's portfolio shape, the firm's pacing, the tax of an additional board seat on a busy partner. TSM produces the structured inputs. It does not pretend to override what the IC concludes from them.
Once a company has crossed the policy filter and entered the long list — however the investor got it there — TSM applies the same methodology consistently across the five in-scope DDs. The investor stops worrying about whether the long-list memo was written by a strong analyst or a weak one. They stop worrying about whether the vet-list framework matched the long-list framework. They stop worrying about whether the quarterly portfolio refresh refers back to the investor memo. The framework is constant; the evidence depth changes with stage; the company data file accumulates.
9.Claims and limits
An architecture document that overclaims gets discounted. The strongest version of the case states what TSM doesn't yet have, explicitly, and lets the architectural argument carry the rest.
Public DD and Private DD produce structured assessments against the same data file. Confidential DD and Vetting DD outputs (investor memo, technical dossier, closure plan, white paper, science article, glossary) are generated from that same data file. One company has been through the full Public → Private → Confidential → Vetting cycle.
Portfolio-Level DD is architecturally complete and runs on individual companies; it has not yet been deployed against a multi-company portfolio at any single fund. API-level connectors to specific tools (Hebbia, Rogo, Standard Metrics) are deployment polish that reduce friction over time, not prerequisites for value.
No TSM-assessed company has gone through a complete outcome cycle that has fed back into the framework. The methodology reflects accumulated mentoring judgement; it has not yet been retrospectively tuned against return data. A fund needing a proven correlation between TSM scores and downstream performance will need to wait for the calibration cycles to run, or run their own.
The total population of TSM assessments is in the tens, not the thousands. The architecture compounds; the dataset hasn't yet. PitchBook holds millions of records. TSM cannot reach that and won't try. The bet is that structured-judgement scale is worth more per data point than raw record count. It's still a bet.
A fund with strong engineering capacity could build a TSM-equivalent internally — and the make-or-buy question is real. Two architectural points are worth making about that path. First, encoding the framework took many years of structured mentoring practice; the schema is the easy part, the resolved decision logic inside it is not. Second, the calibration that builds over time only builds across a portfolio of funds, not inside one. A fund building internally compounds against its own deal flow alone. A fund using TSM compounds against the broader population the platform assesses across all customers. Continuity and population scale beat point-tool depth — but the choice is genuinely the firm's to make.
The pipeline is the same. The vocabulary changes everything.