Start with repetitive, high-volume work: portfolio data collection, KPI reporting, newsletters, meeting notes, intro tracking and resource requests. Automating these first frees a lean platform team to focus on the high-value work that needs a person: founder relationships, strategic programs and crisis support.
Most platform teams I talk to don't lack ideas for automation. They lack a way to choose. Every vendor demo and every article a partner forwards points somewhere different, and a two-person team can only change so much at once. "Where do we even start?" is the question I hear most.
So here's a sorting method plus a task-by-task list. The aim is to pick the first two or three things to automate this quarter, get them working, and leave everything else for later.
One ordering rule first: start with the data. The newsletter, the intro, the partner prep and the LP update all depend on knowing what's happening at each company. Get the data flowing on its own and everything after it gets easier. Do the rest first and you're drafting from stale context.
Score each task before you automate it
Give each recurring task four scores from 1 to 3:
- Frequency. 3 if it happens weekly or more, 1 if it happens once a year.
- Time per instance. 3 if it takes hours, 1 if it takes minutes.
- Predictability. 3 if the steps are the same every time, 1 if each instance is different.
- Relationship risk. 3 if a founder would feel the difference if a machine did it, 1 if nobody would notice.
Add the first three, then subtract the fourth. Tasks scoring 7 or higher are your first candidates. Tasks scoring 3 or lower should stay with people for now.
Here's how that plays out for a typical platform team:
| Task | Freq | Time | Predictable | Relationship risk | Score |
|---|---|---|---|---|---|
| Quarterly portfolio metric collection | 2 | 3 | 3 | 1 | 7 |
| Weekly portfolio news digest | 3 | 2 | 3 | 1 | 7 |
| Meeting notes into the CRM | 3 | 2 | 3 | 1 | 7 |
| Monthly portfolio newsletter | 2 | 3 | 2 | 1 | 6 |
| Intro request tracking | 3 | 1 | 3 | 1 | 6 |
| Resource requests (templates, vendor lists) | 3 | 1 | 2 | 1 | 5 |
| Founder dinner seating and invites | 1 | 2 | 1 | 2 | 2 |
| Supporting a founder through a layoff | 1 | 3 | 1 | 3 | 2 |
Your scores will differ, since the table is illustrative. What matters is that the exercise turns "we should use more AI" into a ranked list you can defend to partners.
The tasks most worth automating
Portfolio data collection
This is the time sink I hear about most. Every quarter someone emails 30 founders, waits, chases, chases again, and then retypes numbers from decks and spreadsheets.
And most of that chasing is for numbers the fund already has. Founders are actually pretty decent at telling you what you need. It's just scattered across decks, updates and email threads, sometimes on page 37 of a board deck. I was on the other side of this as a startup chief of staff. I ghostwrote our investor updates, then got nine different forms and Excel templates asking for the same numbers. I hate forms. I think forms are dead. Ask from a real person's inbox, pre-fill what you already have, ask only for what's missing, and let founders reply in any format. The fund looks like it did its homework, and more founders reply.
What to automate: request → reminders → pulling the numbers out of replies and attachments. What stays human: deciding which metrics to ask for, and reviewing anything that looks off before it reaches a partner or LP.
Cura handles this loop. It sends metric requests from your own inbox, pre-fills what it already has, and asks founders only to confirm or fill gaps. It follows up automatically and extracts metrics from the replies and documents founders send back, with every number linked to its source. There's a longer walkthrough in automatic metrics.
KPI reporting
Once the data comes in on its own, the internal report is mostly assembly. A monthly portfolio snapshot with the same structure every time is a good fit for a template that fills itself, with a person writing the two-line commentary at the top.
Newsletters
AI can draft the portfolio newsletter from a list of inputs: new hires, launches, fundraises, open roles, upcoming events. Set it up to draft, never to send. The platform lead edits for voice and picks what leads. Expect to rewrite the intro every time, because that's the part founders read (if they read anything at all, which is never a given).
Meeting notes
Call recording and note tools are mature now. The useful step is connecting them, so a summary of every founder call lands in the CRM against the right company, with action items pulled out. Then the partner who takes the next call starts with context.
Intro tracking
The intro itself should stay personal. The tracking around it is a good fit, as long as nobody has to log anything by hand. People who log manually forget. Capture it from the email thread itself: who asked, who was introduced, the date, and a follow-up two weeks later asking "did this go anywhere?" That follow-up is the step most teams skip, and it's how you learn which intros work.
Over a year, that record becomes a value-add log: how many customer intros, investor intros and candidate intros the fund made. That count helps in LP diligence and in competitive deals against bigger brands.
Resource requests
"Do we have a template for an option pool refresh?" "Who did another company use for SOC 2?" These questions repeat. The tempting answer is a resource hub, and I'm not a fan. In my experience founders don't go there, because nobody remembers the login or the link. Answer where founders already ask: Slack, email or WhatsApp. An assistant connected to your portfolio context can draft a reply to the common ones, and a person sends it or routes it to someone who knows.
Keep a count of which questions come up most. If the same request arrives five times in a quarter, write a standard answer for it. Each standard answer should also name a person to ask when it doesn't fit.
Signals that trigger a reach-out
The most useful thing to automate might be noticing. A founder posts three new roles. An ask sits in paragraph four of a monthly update, where the people who could help never see it. The same hiring struggle shows up three months in a row. Silence from a founder doesn't mean nothing's happening, and the signal is often already public.
Have the system watch for these signals and put each one in front of a person: signal → a person decides → reach out. Think of it as signal-based sales applied to platform work. The best platform teams act before they're asked, and they help where the founder already is.
Roll it out in three stages
A crawl/walk/run plan keeps a small team from breaking everything at once.
Crawl (weeks 1 to 4). Pick one task from the top of your list. Run it with AI help, but keep every step visible. Note where the output is wrong.
Walk (months 2 to 3). Automate the steps that held up. Name one owner. Add an alert for when the workflow fails, because silent failure is how automations die.
Run (month 4 onward). Once the first workflow is boring and reliable, start the next one. Report the time saved to partners, along with what you did with it.
That last part is the one people skip. Partners will judge automation on what the team does with the freed hours, so plan that before you start. The post on how AI is changing the platform team covers how that shift changes the role, and scaling a platform without losing the human touch covers which moments to protect.
How much time can AI save a VC platform team?
It depends on how much of your week is repeatable work, and the only honest way to know is to measure it. Log your time for two weeks, tag each task as repeatable or judgment-based, and estimate savings on the repeatable share only. That number is your realistic ceiling.
A simple version of the log:
- Task: "Chased three founders for Q2 numbers"
- Time: 45 minutes
- Type: Repeatable
- Could AI do the first draft or the whole thing? Whole thing, with review
Say a two-person team finds that a third of its combined week is repeatable. Even if automation handles only half of that, you've won back real hours every week. Write the estimate down before you start, so you can compare it with what actually happens.
What should never be automated in portfolio support?
Keep people on anything where trust is the product: crisis support, layoffs and down rounds, key executive hires, founder conflict, and the first conversation with a new portfolio company. Automation can prepare the context for these moments, but a person should show up for them.
My test is to imagine the founder finding out a machine handled it. If they'd shrug, automate. If they'd feel let down, keep a person on it. Automation still helps here, just indirectly. When the routine work runs on its own, the platform lead has the time and the context to be fully present when a founder calls on a bad day.