IThe short version
Affinity's relationship score is a legitimate frequency-and-recency signal, and it correctly ranks the warm paths your team actually has — but it treats every contact as equal influence, so it's confidently wrong on exactly the relationships that matter most: the target who just went quiet, and the decisive call that got logged late or never. Treat the number as triage for who to loop in on an introduction, never as a ranking of relationship quality.
That's the whole verdict. The rest is how I got there, and where the edges are.
I build AI systems for venture funds for a living, which means I live inside this feature in real instances — not the sales demo. Nobody has published an independent test of the scoring itself; search the term and you get Affinity's own marketing pages, back to back. So I ran the score against the one thing the marketing never does: a fund's actual network, with partners sitting next to me telling me who they'd really call first.
IIWhat Affinity's relationship score actually measures
The score is a number from 0 to 100, and it is built from one thing: interaction metadata. Not what you said — how you interacted. How often two people email, how recently, whether replies actually come back, and whether meetings show up on both calendars. A relationship with frequent, recent, two-way contact scores higher than a single introduction three years ago, as Affinity's own product page puts it, "regardless of how either party would describe it."
This isn't a black box someone sprinkled the word "AI" onto. An independent technical breakdown from Salesforce Ben — not Affinity's marketing — confirms the same mechanic: a proprietary algorithm computes a 0-to-100 score from the frequency and recency of communication between your team and their outside contacts, then surfaces it on the connection records and inside the inbox and browser extensions.
One thing worth saying plainly, because buyers get nervous about it: the base score does not read your email or meeting text. Message content is only touched if a firm opts into a feature that needs it — an AI agent that drafts a follow-up or proposes a record update, for example. The raw relationship score is metadata only, and Affinity carries SOC 2 Type II and ISO 27001 certification on top of that. It's a real engineered system. The problem isn't that it's fake. The problem is what it quietly leaves out.
IIIThe mechanics nobody puts on the pricing page
Here's the operational detail you won't find in the marketing copy, pulled from Affinity's own help center.
Scores recalculate every six hours. So the email you just sent won't move the number right now — if you're checking a score to decide whether to send an intro in the next ten minutes, you're looking at a stale picture. A first sync can take up to 24 hours to populate connections, which matters the week you onboard a new partner and the network suddenly looks thin because it hasn't finished importing.
The formula is not configurable. You can't weight it, you can't tell it that a board seat matters more than a newsletter reply, you can't tune it to your fund's reality. Everyone gets the same math. And a manually logged interaction — you tap in that you had a call, or a LinkedIn conversation — counts exactly the same as a synced email or calendar invite. That's a genuinely useful escape hatch, and it's also where a lot of the score's accuracy quietly leaks out, because people don't log consistently.
Two more terms the score leans on. A 1st-degree connection is someone on your team directly knows and has communicated with. A 2nd-degree connection is someone who knows one of your 1st-degree connections but whom nobody at the firm knows directly — click a target founder and Affinity surfaces everyone at your firm who could make the warm introduction. That warm-intro-path surface is the feature's real payoff. It's also, as we'll see, only as honest as the data feeding it.
IVRunning the score against a composite fund's real network
I'm not going to name a fund — I never name a client, and I'm not going to publish any firm's private network graph. What I can do is describe the pattern I see repeated across the instances I work inside, as a composite of real fund networks rather than a demo dataset.
The test was simple. For a slate of live and target relationships, I pulled Affinity's ranking of the strongest warm paths, then sat with the partners and asked a blunt question for each one: if you had to get this intro done tomorrow, who on this team actually makes the call? Then I compared the two lists — the algorithm's ranking against the humans' ranking. Where they agreed, the score was earning its keep. Where they split, that gap is the whole story.
VWhat it got right
More often than I expected, the two lists agreed. High-frequency, two-way, recent contact reliably surfaced the real warm path. When a partner had genuinely been in a founder's inbox all year, the score knew it, and it pointed at the right person to send the intro. That tracks with what a VC who has run deal flow across 65-plus investments on Affinity, 4Degrees, and Salesforce calls "the feature I'd actually pay for," his point being that roughly 80% of the deals that matter in venture come through a warm path, so a tool that ranks warm paths is pointed at the right problem.
The quietly valuable part is resurfacing. The same reviewer gives the sharp example: knowing a partner emailed a target founder 40 times last year versus once in 2021 changes who should do the outreach — and that's a signal a spreadsheet of "contacts" will never give you, because it has no sense of time. Affinity's own marketing claims the right warm path can close a deal 25% faster than a cold approach. I'd take that number as a vendor's number, not gospel. But the direction is real and well documented across the category: warm introductions convert at up to roughly five times the rate of cold outreach, and referral-driven deals have closed up to 69% faster than non-referred ones. That's the whole reason every vendor in this space, Affinity included, builds scoring on frequency and recency. The signal is genuinely worth something.
VIWhere the score gets confidently wrong
Then you hit the gaps, and they're not random — they're structural.
Frequency is a proxy for influence, not the same thing, and a third-party CRM teardown names the failure mode exactly: "A weekly thread with a procurement analyst scores high while a single decisive conversation with the economic buyer scores low." Translate that to venture and it's the associate you trade emails with every week outscoring the GP who can actually write the check but only talks when it counts. The score rewards volume. Decisions don't move on volume.
The second failure is the partial-adoption trap, and it's the one that bites funds hardest. Relationship intelligence only works when every relevant inbox is connected. Leave one partner's email out of the sync — the one who happens to own the relationship that matters — and the score doesn't say "unknown." It says "weak," confidently, because from its point of view no contact happened. The same teardown is blunt about it: partial adoption "produces confidently wrong relationship scores." One missing inbox silently corrupts the ranking for the whole team, and nothing on screen warns you.
The third gap is simply everything the score can't see. Phone calls, texts, in-person conversations at a conference, a LinkedIn DM thread — none of it counts unless someone manually logs it, and people don't. So the partner who does their best relationship work on the phone and over dinner reads as a cold contact. The algorithm isn't measuring the relationship. It's measuring the exhaust from one or two channels, and calling that the relationship.
VIIThe reliability problems underneath the algorithm
Separate from whether the score is accurate is whether the tool underneath it is solid, and here I'll be even-handed rather than cherry-pick. A first-hand account from someone whose firm runs Affinity reports teammates losing many of their meeting notes with no way to recover them — enough that staff started drafting notes in Apple Notes first — and List views taking 20 to 30 seconds to load, observed even in screen-shared sessions with Affinity's own customer success team.
Balance that against the aggregate. As of September 2026, Affinity's G2 page showed 4.4 out of 5 across 73 reviews, with VC reviewers praising the Gmail connection, the Kanban-style pipeline, and team coordination, while flagging price and asking for more granular user permissions. That's a well-liked product with real rough edges — not a disaster, not the flawless platform the demo implies. If you're leaning on the relationship score for anything important, a tool that occasionally eats a meeting note is a tool whose manual-log escape hatch you can't fully trust either.
VIIIWhat it costs to unlock, and what you can't take with you
Relationship intelligence isn't on the entry tier. Per list pricing verified against affinity.co in May 2026, the feature lives on Scale and above, not Essential.
List price is only the starting line — what funds actually pay drifts well above it once seats and add-ons stack up, which I've broken down separately in what Affinity actually costs by fund size.
The part that should shape the buying decision is the lock-in. The computed strength score is proprietary and does not migrate. A CRM-migration specialist confirms that you can move contacts, notes, and emails out via API or a migration partner — but the score itself cannot be ported. Switch tools and you rebuild an equivalent formula from scratch, for example in Attio via custom fields keyed on last-contacted date or total emails sent. So the single most-cited reason to buy Affinity is also the thing you can never take with you. That's worth sitting with before you standardize on it — I get into the switching math in Affinity vs. Attio.
IXHow this compares to the rest of the category
None of this is unique to Affinity — it's inherent to scoring relationships from metadata. SAP ships its own relationship-intelligence module (internally nicknamed "Hugrank") built into SAP Sales Cloud, and it computes strength the exact same way, from frequency and recency — along with the same structural blind spot. SAP's version only integrates with Office 365, so any activity in Gmail is simply invisible to the score. Different vendor, identical shape: the algorithm is only as complete as the one mailbox it's allowed to watch.
4Degrees is the closest competing take aimed squarely at venture, and it makes many of the same bets. I won't turn this into a head-to-head — that's its own piece. The point is that "relationship intelligence" across the whole category rests on the same frequency-and-recency foundation, which means it inherits the same two weaknesses everywhere: it mistakes volume for influence, and it goes blind on any channel it isn't plugged into.
XThe honest verdict: when to trust the number, and when to call the partner instead
Use the score as a triage signal. When you need an intro to a company, it's a fast, genuinely useful way to see who on the team has a live thread worth leaning on — and the resurfacing (who was close last year versus this year) is real value no spreadsheet gives you.
Don't use it as a verdict on relationship quality, and never trust it on anything time-sensitive without a human check. If a target has gone quiet, the score lags by up to six hours and won't flag the chill. If the person who owns the relationship works by phone, the score underrates them. If one inbox on the team isn't synced, the whole ranking is quietly off. Any of those, you ask a partner before you act — the number is a starting point for that conversation, not a replacement for it. In practice the fastest answer to "who actually knows this founder" often comes from searching your team's real network directly, which is why I've argued that a Slack bot that searches your fund's actual network beats relationship-intelligence software for that specific job.
XIFAQ
Do you build a custom scoring layer on top of Affinity's connection data?
Yes, when a fund wants one. Affinity's native score isn't configurable or weightable, so if your partners care about signals it ignores — a channel it can't see, a flag for relationships that just went quiet, a weighting that treats a board relationship differently from a newsletter reply — that logic has to live outside the native score. I build that layer against Affinity's underlying connection and activity data (and your LinkedIn network, which the CRM never sees), so the ranking reflects how your firm actually works rather than one fixed formula. We're an AI systems studio, not a CRM vendor — the goal is to make the tool you already pay for tell you the truth.
What does the Affinity Chat Bot add on top of the native score?
The Affinity Chat Bot lets your team query the CRM in plain English — companies, deals, co-investors, notes, and the network itself — straight from Slack or a Custom GPT in ChatGPT, without anyone logging into Affinity to click around. Where the native score hands you one number on a record, the bot lets you ask the question you actually have — "who here has a live thread with anyone at this company, and how recent is it?" — and get the warm-path answer back in the channel where the work already happens. It's a layer on top of Affinity's data, not a replacement for it, and it's the kind of thing we'll build to fit a specific fund's workflow.
— Michael Rouveure
