Reference architecture / Venture fund operations / August 2026
A stack is not a shopping list. It is two systems of record and eight feeds into them. Almost every bad stack fails the same way: it has three.
The one idea
Every tool a fund buys wants to become the place where truth lives. Sourcing platforms want to own the pipeline. Fund admin wants to own the portfolio. Portfolio monitoring wants to own the LP report. Left alone, a firm ends up with four half-populated databases and a partner who trusts none of them.
A working stack forces the question early: where does a fact go to be true? There are exactly two answers, and the discipline is refusing to add a third.
Spine 01 — Relationships and pipeline
Every company, person, meeting, intro and pass. Nothing enters the pipeline that does not enter here. Sourcing tools, inbound forms and notetakers write into it; they never sit beside it.
Spine 02 — Numbers and positions
Ownership, cost, NAV, KPIs, cash. Fund admin and portfolio monitoring resolve to one set of figures that survives an audit. If your LP deck and your fund accountant disagree, you do not have this spine.
The ten layers
Layer 01 touches a company before anyone has heard of it. Layer 10 touches an LP three years after the cheque cleared. Build them in this order — a fund that buys layer 05 before layer 01 is monitoring a portfolio it did not source well.
01
Find it first
Company data plus the leading indicators that move before a round is announced: hiring velocity, web traffic, commit activity, domain registrations, founder movement. Coverage is table stakes; the edge is in the screen you point at it.
Default
Harmonic
Pre-announcement signal at seed. Widely adopted by emerging managers.
Europe
Dealroom, Beauhurst
Best ecosystem depth for UK/EU; Beauhurst for UK Companies House granularity.
Reference
PitchBook, Crunchbase
History, cap tables, LP ownership. Growth-stage diligence more than seed sourcing.
Indicative Harmonic $12k–36k/yr · PitchBook $12k–30k/yr · Dealroom from €10k/yr · Crunchbase $588–2,388/yr
02
Kill it fast
The highest-leverage layer and the one most firms skip. A structured intake form plus an LLM that scores the deck against a written thesis turns partner triage from a 45-minute read into a 10-minute decision — reported figures, but the direction is consistent across firms. The prerequisite is a thesis specific enough to screen against.
Default
Tally or Typeform → CRM
Structured intake beats an inbox. Everything lands as a record, not an email.
Purpose-built
Deckmatch
Parses decks and matches against a stated thesis automatically.
Roll your own
LLM + prompt + webhook
A scoring rubric in a prompt, run on every submission. Cheap and yours.
Indicative £0–500/mo · this layer is mostly configuration, not licence spend
03
Hold the truth
The single most consequential purchase. The real question is not features, it is whether relationship intelligence is load-bearing for your sourcing. If warm intros are how you win, buy the relationship graph. If you source from data and inbound, buy something flexible and cheap and spend the difference on layer 01.
Relationship-led
Affinity
Category leader. Automatic email and calendar ingestion builds the graph for you.
Data-led
Attio
Highly customisable, credible free tier, popular with European funds. Weaker graph.
Lean / solo
Folk, 4Degrees
Folk for under ~200 live companies. 4Degrees if warm-intro pathing is the whole game.
Indicative Affinity ≈$2,000–2,700/user/yr (~$20k floor) · 4Degrees ≈$1,200–1,800/user/yr · Attio $29–69/seat/mo · Folk $24–48/seat/mo · DealCloud enterprise-only
04
Know it cold
Market maps, competitive teardowns, financial model decomposition, reference call synthesis, memo drafting. This is now overwhelmingly general-purpose LLM work rather than a specialist product — which is why the adoption numbers here are the highest of any layer in the stack.
Default
Claude
Highest reported adoption among data-driven funds (90.5%). Long-document and memo work.
Second seat
ChatGPT, Gemini
73.3% and 56.2% adoption. Most firms run two, not one.
Cited search
Perplexity
Fast market scans where the citation trail matters more than the prose.
Indicative $20–200/user/mo · the cheapest layer with the largest measured time saving
05
Never forget
The most underbought layer in venture. A fund’s real asset is why it passed on something in 2023, and that is almost never written down anywhere retrievable. Notetakers plus a searchable store turn three years of partner meetings into a queryable corpus. This should write into the CRM, not sit next to it.
Default
Granola
Built for people who take their own notes; strong cross-meeting querying.
Alternatives
Fathom, Fireflies, Otter
Bot-joins-the-call model. Better for recorded external calls and transcripts.
Store
Notion or Slack + search
Whatever it is, one place, with retention rules and a search layer over it.
Indicative $10–30/user/mo · disproportionate return on the second fund
06
Make it happen
This is the layer where VCs work with founders to create strategy and plan, make introductions, track impact, and leverage their network to maximise what moves the company valuation and shorten the path to the next round. Every stack map published — this one included, until now — skips it, because no vendor sells it. That absence is the finding, not an oversight: the job is real, it decides more of the return than any layer above it, and there is no category to buy.
Vendor category
None established
Marketplaces sell hours. Consultancies sell projects. Neither carries scope, acceptance or settlement.
At scale
Platform team on payroll
The a16z answer. Around $165k per hire before on-costs, funded from a fee base most funds do not have.
Improvised
Advisor equity, or nothing
Dilutes the position you just paid for, and produces no record an LP can read.
Learn about Execution Capital →
Indicative $0 to $1m+/yr, entirely depending on which of the three above you pick · the widest spread in the stack
07
Watch it grow
KPI collection, benchmarking, flags. The failure mode is collecting quarterly data nobody reads — reporting theatre. Buy this when you have enough companies that you can no longer hold their numbers in your head, which is usually around fifteen, not five.
Institutional
Standard Metrics
Automated collection and benchmarking; used by 150+ institutional managers.
Founder-friendly
Visible.vc
Lowest friction for portfolio companies, which is what actually drives response rates.
Analytics-heavy
Vestberry, ScaleX, Chronograph
Deeper modelling and valuation roll-up. ScaleX is the European option, with IPEV-compliant marks built into the monitoring; Chronograph skews to larger, LP-facing funds.
Indicative Visible from ≈$149/mo · platform tier commonly $5k–20k/yr
08
Keep the books
Capital calls, NAV, waterfall, K-1s and their local equivalents, cap table. Regulated, audited, unglamorous, and the one layer where building your own is close to indefensible. Note that most administrators do not publish pricing — expect a quote process.
Emerging manager
Carta, Infra One, Further
Carta’s emerging-manager tier around $1,500/mo. Infra One covers formation and admin together for a first fund; Further runs VC and PE administration at a larger book. Neither publishes a rate.
Institutional
Juniper Square, Allvue
Full fund accounting, LP portal, compliance. Quote-based, materially more expensive.
SPVs and vehicles
Odin, Bunch, Sydecar, Allocations
Odin and Bunch are the European defaults; Sydecar and Allocations the US ones.
Indicative $5k–20k+/yr at emerging-manager scale · quote-based almost everywhere, Carta excepted
09
Stay legal
LP subscription documents, AML and sanctions screening, adverse media, regulatory filings. Painful, mandatory, and the fastest place to lose an LP: a bad subscription experience is the first operational signal an institutional LP receives about your firm.
Subscription
Passthrough
E-subscription with KYC/AML built in. The category default for new funds.
Alternative
Anduin, DocuSign
Anduin for adaptive sub docs; DocuSign where the workflow is already legal-led.
Screening
ComplyAdvantage
Sanctions, PEP and adverse-media monitoring against a large source set.
Indicative Quote-based · budget as a fund expense, not a firm expense
10
Show your work
Quarterly narrative, NAV marks with an audit trail, and a room to raise the next fund from. Whatever tool produces the LP report must read from the ledger, not from a partner’s spreadsheet. If those two can diverge, they will, and you will find out during a re-up.
Reporting
Visible, Standard Metrics
Both generate LP-facing reports directly from collected portfolio data.
Valuation
73 Strings, Chronograph
Defensible marks with an audit trail. Matters the moment you have an auditor.
Data room
DocSend, Ansarada, Intralinks
DocSend for fundraising decks with page analytics; the others for full diligence rooms.
Indicative DocSend from ≈$45/mo · valuation platforms quote-based
On the glue. Underneath all ten sits the layer nobody budgets for: Zapier or n8n for automation, 1Password for credentials, Deel or Rippling for payroll, and — above roughly $150M — a small warehouse (Postgres or BigQuery) that the CRM, the monitoring platform and the ledger all write into. Firms discover this layer when the analyst who built the automations leaves.
Configurations
The same ten layers, in three configurations. The stack does not scale linearly: it scales in three jumps, and each jump is triggered by an event rather than an AUM number: the first LP who demands an audit, the fifteenth portfolio company, the first fund where sourcing coverage stops being achievable by hand.
Execution Capital — scoped delivery at any of the three, without the payroll →
Figures aggregate published list prices and secondary estimates; most fund-administration and enterprise vendors quote privately. Firms building proprietary AI and data infrastructure report year-one totals well above the software line — plausibly $250k–700k at mid-size and into the millions at mega-fund scale, though those estimates come from a single secondary source and should be treated as directional.
What the data says
The clearest finding in the 2026 Data-Driven VC Landscape is not about tools at all. Funds have stopped treating tooling as a procurement question and started treating it as a headcount question — and the headcount they are adding is engineering, at the direct expense of junior investing roles.
345
data-driven VC firmsidentified in 2026, up from 151 three years earlier.
49%
plan to hire an engineerin the next year. Just 2% plan to add a junior investor.
57%
are ramping internal tools, up from 37% a year earlier.
1:1
engineering spend vs data and tools spend, having moved from roughly 2:1.
Two archetypes have separated out. Workflow Builders (42%) run no engineers and assemble bought tools well — median around $151M AUM and seven people. Fullstack Builders (58%) keep engineering in house — median around $800M AUM, 23 people, two engineers. Both work. The failure is the middle: a firm that wants proprietary infrastructure and staffs it with an analyst and a weekend.
The stated bottleneck is not technology. 49% cite time and bandwidth as their biggest constraint, and 41% cite data quality — which is to say the two spines, and whether anyone is maintaining them.
Read that against layer 06 and it stops being a tooling finding. Bandwidth is the binding constraint in every layer that requires someone to do something rather than run something — and layer 06 is nothing but that. It is also the one layer a fund cannot hire its way out of below a certain fee base: a platform hire is a fixed cost, and a £10–25m fund earning £200–500k in gross management fee is not choosing not to make it. So the layer with the largest effect on the return is the layer with no vendor, no budget line, and the constraint firms name first.
Build versus buy
The line is sharper than most firms make it. Anything audited, regulated or commoditised is a purchase — you will never out-build a fund administrator, and no LP has ever been impressed by one. Anything that encodes your specific thesis is the only place a technology edge can exist, because a screen every fund can run has stopped being a screen.
Failure modes
Every one of these is more common than a bad tool choice, and none of them is fixed by buying something.
The sourcing platform holds companies you have not put in the CRM. The monitoring tool holds numbers the fund administrator has not seen. Nobody decides which one wins, so partners quietly build private spreadsheets and the stack becomes decorative. Fix: name the winning system per fact type, in writing, and delete the losers’ write access.
Signal platforms return everything. Without a written thesis specific enough to screen against, layer 01 produces a firehose that a junior person is then paid to hand-filter — the exact cost the tool was bought to remove. Fix: write the screen first, in prose, then buy the data to run it.
Quarterly KPIs are collected, chased, chased again, formatted, and never used to change a decision. Response rates fall, data quality follows, and by year three the portfolio dataset is worthless. Fix: cut the metric set to what you would actually act on, and show portfolio companies their own benchmarks so collection buys them something.
One analyst built forty automations across six tools, documented none of them, and left. The stack keeps working for about a month. Fix: automations live in one platform, in version control, with an owner named in the fund’s operating manual.
Layer 07 is easy to buy and layer 06 is not, so firms buy layer 07 and call it portfolio management. The result is a fund that measures companies with increasing precision and helps them no more than it did before — a dashboard of deterioration you have no capacity to act on. Monitoring is not an intervention. Fix: before buying a monitoring platform, write down what you would do with a red flag. If the answer is “send an introduction”, buy capacity first.
Nine tools at $50 per seat is invisible at four people and material at fourteen, and by then every tool has one loyal user defending it. Fix: an annual line-by-line review against the layer map — if a tool is not the system of record for its layer or feeding one that is, cut it.
Sourcing is fun, so it gets four tools. Compliance is not, so it gets a folder of PDFs. The stack ends up strongest where the partners’ attention already was and weakest where the fund’s actual risk sits. Fix: budget by layer, not by enthusiasm.
Sources
Ranked by how much weight the document puts on them. One survey does the heavy lifting; everything else is corroboration, and a share of it is written by vendors in the categories being ranked.
Primary survey
datadrivenvc.io · n = 345 firms
Firm counts, engineer hiring, the two archetypes, tool adoption. The strongest source here, and self-selecting: it surveys firms that already identify as data-driven, so the levels describe the leading edge, not the median.
Pricing and cost bands
valueaddvc.com
CRM per-seat pricing and the total stack cost bands behind the tier table.
Category map
evertrace.ai
Category taxonomy and firm-size cost tiers. One of the published stack maps that has no execution layer.
Back office
portfolioiq.ai
Compliance, valuation and data-room vendors in layers 09 and 10.
Fund admin
vcbeast.com
Emerging-manager fund admin pricing, and the finding that of 86 fund-operations providers, twelve published any rate at all.
Sourcing
vcbacked.co
Layer 01 cost ranges and vendor positioning.
Vendor-authored
crustdata.com
Motherbrain, Beacon and DIG detail. Written by a vendor in the category it describes; treated as directional, not as measurement.
Secondary estimate
tommasomariaricci.com
AI programme cost bands and workflow timings. Sole source for the $250k–700k build figure in the tier table footnote; directional only.
Self-reported
eqtgroup.com · standardmetrics.io
Firm and vendor material about their own products. Existence proofs for what in-house infrastructure can do; the attributed wins are selected.
Pricing is the weakest data in this document. Most fund-administration, valuation and compliance vendors publish nothing and quote privately; the figures above are list prices and secondary estimates, they vary by seat count, region and negotiation, and they should be read as order-of-magnitude only. Where a claim originates with a vendor about its own product — time savings, accuracy percentages, lead times — it is marked as reported rather than stated as fact.
Disclosure — the publisher
Execution Capital builds in layer 06, and in no other layer. We do not sell a CRM, a data provider, a fund administrator or a monitoring platform, and nothing of ours is recommended anywhere above.
What we operate is the category that does not exist yet: portfolio support as scoped work rather than goodwill. A company’s gap is diagnosed against a shared maturity scale. A senior operator who has solved that exact problem is matched to it. The work is written down as a deliverable with acceptance criteria and a date before anyone starts. Settlement releases only when the founder accepts. Roughly thirty percent settles in cash; the rest settles in Venture Capital Interests — ring-fenced units at the ecosystem level, never on the company’s cap table and never against the fund’s reserves.
The by-product is the thing layers 07 and 10 have always struggled to produce honestly: a delivery record generated by the transaction itself, rather than a report someone writes about their own support afterwards.
Structure
Raise
Source
Select
Deploy
Support
Follow-on
Realise
Execution Capital Ltd, London. Venture Capital Interests™ is a trade mark of Execution Capital Ltd. Early-stage ventures are high-risk and there are no guaranteed outcomes. Nothing here is an offer of securities or investment advice. Talk to us.