How to turn portfolio updates into searchable firm memory
Eric Friedman Co-founder, Railblocks
Short answer: Keep every original update, pull out its facts, and file them by company and reporting period. Have a person check the numbers that matter, and link every answer back to its source. Do that and your firm can ask questions across years of updates and trust what comes back.
Most firms already have the raw material. Updates land in a partner’s inbox, attachments end up in a shared folder, and the context stays with whoever read them. When someone asks how a company’s burn has changed over the past year, an associate goes digging. The goal here is to make that question take seconds.
What your firm memory should hold
Searching your inbox can find the email that mentions burn. It can’t tell you that burn has risen three months in a row, or that a company changed how it reports revenue in March. For that you need four things kept together:
| What to keep | Why it helps |
|---|---|
| The original email and attachments | Anyone can check what the company said |
| The facts: metrics, people, asks, and risks, filed by company and period | You can ask questions across time and across the portfolio |
| Your definitions of revenue, burn or ARR | Numbers stay comparable, and you can see when a company’s differ |
| Who reviewed a number and what they changed | Your team knows what’s been checked |
How to build it
1. Give updates one way in
Set up a shared address like [email protected] that partners can copy or forward to. Founders can send to it directly too. It’s tempting to connect everyone’s inbox on day one, but that brings in personal mail and raises hard permission questions. A forwarded update is a clear signal that someone wants it remembered.
2. Keep the original
Store each email and attachment unchanged, with its sender and date, and have every extracted fact point back to it. This also protects you later: when extraction tools improve, you can run them again over everything you’ve kept.
3. File facts by company and period
Pull out the metrics, people, asks, and anything qualitative worth keeping, then file each one against the company and the period it describes. An email sent in October about September belongs to September. Match company names against your CRM or portfolio list. A misspelled name then gets flagged instead of becoming a new company.
4. Decide what a person checks
A rough summary is fine when a partner is skimming for context. A number going into an LP letter needs a person. A simple rule works well: someone reviews a figure when it’s new, when it moves past a threshold you set, when it conflicts with an earlier update, or when it’s about to leave the firm. Keep a record of corrections, because they show you where extraction needs work.
5. Set permissions before anyone searches
Good search makes old access mistakes easy to find. Decide who can see deal team notes, board materials, and anything under confidentiality terms, then test it by logging in as someone who shouldn’t have access. Check what happens when someone leaves the firm too.
6. Show the source with every answer
Every answer should link to the update it came from and show which numbers it calculated. Here’s what that looks like, using a made up company:
What changed at Val AI, and where can we help? Revenue grew 20.8%, calculated from $745K in August and $900K in September. There’s one open ask from the September update: help hiring a Principal Voice Engineer.
The reader can open the email, see which number the founder stated and which one was calculated, and decide what to do next.
7. Test it with questions you already know
Write 20 questions your team asks, using updates you’ve already read. Include a restated number, a company that stopped reporting, and a question the record can’t answer. When it can’t answer, it should say so. Our guide to evaluating an investment AI agent goes deeper on this.
How this differs for VC and PE firms
VC firms mostly deal with founder emails in every format imaginable. The hard part is inconsistency: metrics change from month to month, and some companies go quiet. A good first output is a monthly view of what changed across the portfolio and which asks are still open.
PE firms usually get monthly KPI packs, financials, and board decks. The formats are more regular and there’s far more detail. The hard part is keeping definitions consistent across portfolio companies. A good first output is a monitoring view against plan, along with a draft board summary.
Portfolio monitoring and board reporting
If you’re a PE firm, start with one recurring pack from one company. Map its KPIs to your firm’s definitions, compare actuals to plan, and draft the variance commentary, then let the deal team correct it. After a few cycles you’ll know how much time it saves and how many corrections it still needs. From there, adding the next company gets much easier.
People should still decide what a board or an LP is told. The system’s job is to gather the numbers, spot what changed, and hand your team a good first draft.
LP letters and follow on decisions
Once updates are filed by company and period, the same record can support an LP letter or a follow on brief. These carry the most risk. We’d add them after the record has proven itself on internal questions. Where your fund administrator holds the official figures, cite theirs rather than calculating your own.
Should you buy this or build it?
Look at portfolio monitoring products first. If one covers your sources, your definitions, and the way your team reviews numbers, buy it.
Building makes more sense when you want updates connected to everything else your firm knows, like CRM history, IC memos, and partner notes, or when you want several agents working from the same record. If a partner has already written something that reads updates, you’ve learned what your firm wants from this. The steps above are what turn that into something the whole team can rely on. Whichever way you go, it’s worth asking a few questions up front: Can we export everything, including our corrections? Where is the data stored, and which AI providers see it? Who fixes it when a company changes its format?
Mistakes we’d avoid
- Connecting every inbox first. Start with the updates people choose to share.
- Keeping summaries without sources. If nobody can check a number, nobody will trust it.
- Assuming extraction is right. Spot check the output regularly.
- Ignoring silence. A company that stops sending updates is telling you something. Make it visible.
- Automating the last step first. Have a person review anything that leaves the firm.
Frequently asked questions
Do we need to replace our CRM or portfolio software?
No. They stay where they are and become sources. The memory connects what they hold to what arrives by email and in documents.
How much history should we bring in?
Start with the past few quarters for a handful of companies and get that working. Then bring in older material when someone has a question for it. Older files tend to be messier. Expect more review.
Does our data leave our environment?
Ask for the data flow, whoever you work with. Your storage can sit in an account your firm controls while text still goes to an AI provider for processing. You want a clear answer on both.
Can we trust AI with portfolio numbers?
Yes, with the right setup: keep the source, mark what was calculated, have a person review what matters, and test it against answers you already know.
How we can help
This is what we build at Railblocks. Our Email Ingestion block turns company updates into context linked to its source, and Firm Memory keeps that context with the originals. It all runs in your firm’s own Cloudflare account, and you own it. We explain why that matters in Every Firm Should Own Its Memory.
Bring us one recurring update or board pack. Reach out.