It's Friday of the quarter's last week. Your ops lead has emailed 22 portfolio companies for KPIs. Six sent back the form. Four sent a board deck. Three sent a PDF with one chart and a paragraph. Two sent a Google Sheet link. Seven haven't replied. Monday's partner meeting depends on fifteen different submissions scattered across four mailboxes.
This is what KPI collection actually looks like once your portfolio crosses fifteen companies. And chasing isn't the problem.
The real problem isn't silence. It's format.
Every fund treats KPI collection as a communications problem: send the form, chase the laggards, escalate to the partner, repeat. But the non-responders are the visible cost. The invisible one is bigger, turning whatever did come back into one clean row.
A 20 company portfolio in a normal quarter produces something like fifty separate data fragments: forms, decks, PDFs, emails, spreadsheets. Someone has to manually turn all of that into the same format. That reconciliation tax is where most ops time actually goes, and it grows every time the portfolio grows.
Four ways this breaks
| Failure mode | What it looks like | What it costs you |
|---|---|---|
| Format chaos | Same KPI shows up as a slide, a form field, a PDF table, an email sentence. | The biggest hidden cost - grows with every new company you add. |
| Founder silence | Same names miss the deadline, every cycle. | Chasing hours, plus partner attention on every escalation. |
| Lost in handoff | Deck → analyst's spreadsheet → ops lead's file → CFO's recalculation. | Every handoff is a chance to introduce an error before it reaches an LP. |
| Definition drift | “ARR” means three different things at three different companies. | Breaks your fund-level numbers quietly - most teams have given up trying to catch it. |
Format chaos is the loudest complaint. Definition drift is the one that actually burns you - it's the reason fund-level metrics go quietly wrong.
What to actually ask for, by stage
Don't send the same form to a pre-revenue company and a Series C company. One will leave it mostly blank, the other will feel shortchanged.
| Stage | Core KPIs | Cadence |
|---|---|---|
| Pre-revenue | Cash, burn, runway, headcount, product milestones | Quarterly |
| Early commercial | Revenue, ARR (defined), gross margin, customers, retention, cash/burn/runway | Quarterly + monthly cash |
| Growth | Revenue, ARR, NRR/GRR, CAC payback, cash/burn/runway, hiring vs. plan | Monthly |
Two things that punch above their weight: define ARR in the form itself ("recognized subscription revenue, annualized, excluding one-time fees"), a free-text field guarantees ten founders give you ten definitions. And match cadence to stage - monthly asks exhaust a pre-revenue founder; quarterly-only leaves you finding out about a growth-stage problem three months late.
Where the fix actually lives
The workflow everyone runs is: send form → chase → receive → normalize → roll up. The waste is that steps three and four repeat the same work — the founder fills the form, then someone re-does it by hand into your schema.
That's the gap AI extraction closes. Feed it whatever the founder actually sent - deck, PDF, email, spreadsheet, and it pulls the structured KPIs straight into your taxonomy. The form doesn't disappear. It gets pre-filled. The founder just confirms it.
That shift changes the math for the ops lead too: less time spent translating other people's formats, more time spent on the two or three companies whose numbers actually need a second look.
One split worth knowing: the extraction itself is the Orca platform. The rhythm taxonomy, cadence calendar, what happens when a founder goes dark is the operational layer SerVC runs for funds that use its fund administration team. Run Orca alone and you build that rhythm yourself. Run it with SerVC and you inherit one that's already working across other funds.
Where does your fund sit?
| Level | Looks like | Cost of staying there |
|---|---|---|
| 1. Ad hoc | Individual emails, cadence lives in someone's head. | Endless follow-ups, manual work and hard to follow. |
| 2. Templated | One form, manual normalization. | Scales badly a bigger fund costs proportionally more, not less. |
| 3. Extracted | AI pulls structure from whatever arrives; form becomes a check, not a chore. | Collection cost stops scaling with portfolio size. |
| 4. Roll-up native | KPIs flow straight into fund dashboards and LP reports. | Basically maintenance-free just quarterly taxonomy review. |
Most first funds run at Level 1. Most second funds graduate to Level 2 - and then stall there, because it works, just expensively. Level 3 is where the cost of adding a company finally stops climbing.
FAQ
Does AI extraction replace the founder questionnaire?
No, it changes what the questionnaire is for. The founder confirms pre-filled numbers instead of typing them from scratch. That confirmation step matters: every founder has one metric they define their own way, and that's exactly the flag you want to catch.
My portfolio's only 8 companies - is this overkill?
Probably, for now. At that size, one person can hold the normalization in their head. Most funds feel the pain somewhere past a dozen companies. Define your taxonomy now; switch to extraction when it starts to hurt.
What if founders just don't send anything, pre-filled form or not?
Treat silence and refusal as the same problem. Most of it is a founder waiting for a clean, reasonably-timed ask - the rest needs a reporting clause in the next side letter, and a board seat to enforce it on existing portfolio companies.
Does this work outside SaaS?
Only if your taxonomy accounts for it. A marketplace's “ARR” is GMV × take rate. A fintech's depends on whether it's interchange, lending, or fees. Build category-specific KPI sets instead of forcing every company into a SaaS template - that mismatch is usually why mixed-portfolio revenue numbers are quietly wrong.
Most ops teams already know collection breaks past fifteen companies. What they usually don't know is which of the four failure modes is actually costing them time. The fix is moving extraction to the source - see how Orca's Portfolio KPI module handles it across your whole portfolio, or book a demo and run it against your own data.
