Pick any CRM homepage in 2026 and count how long it takes to reach the word "AI." Ours included — we're not immune. Salesforce, HubSpot, Dynamics, Pipedrive, Zoho, all of us. Under the branding, the actual capabilities range from genuinely useful to pure decoration. This is a plain-English map of what today's tools really do.
What "AI CRM" actually covers
The features that ship in most serious tools fall into six categories:
- Data entry automation — pulling contacts, meetings and notes into records without a human typing.
- Enrichment — filling in job titles, firmographics and org charts from third-party data.
- Summarisation — turning meeting transcripts and email threads into a paragraph a partner can skim.
- Scoring — lead scoring, deal scoring, health scoring, all with an explanation attached.
- Drafting — first-draft follow-up emails, briefings, executive summaries.
- Question-answering — natural-language search across your CRM data ("show me every open opportunity where the champion hasn't emailed us in 30 days").
What actually works today
Ranked, honestly, from where the field is now:
- Summarisation is now reliable. A good AI CRM turns a 60-minute meeting transcript into a briefing paragraph and a list of action items with owners, and gets it right the great majority of the time.
- Data entry automation is essentially solved once you feed the system Microsoft 365 or Google Workspace. The activity log fills itself.
- Question-answering works well for scoped, well-structured questions against your own data. Ask it something that crosses many systems or asks for judgement and it still fumbles.
- Drafting works for first drafts. The partner still edits every one, and that's fine — the win is time to first draft, not zero effort.
- Scoring works when it's built on real behavioural signal (mailbox, calendar, tickets). It's decoration when it's built on activity counts a rep enters manually.
- Enrichment is mixed. Firmographics — fine. Org charts, personality profiles and "warm intro paths" derived from public data — mostly imaginary, and worth treating with scepticism.
What doesn't work yet
- Autonomous agents that "run the account" end to end. The demos are impressive; the real deployments are rare and heavily supervised.
- Predicting close date more accurately than a good sales manager. AI is not obviously better here yet. Maybe next year, but I'd bet against it.
- Cross-tenant "who at my firm knows anyone at Company X" without ingesting real communications from your firm. Nobody can conjure that from public data alone.
How to tell a real AI CRM from a decorated one
Three questions:
- What primary data does the AI use? Mailbox, calendar, transcripts — real signal. Activity counts entered by reps — decoration.
- Does it show its reasoning? A score without an explanation is a magic 8-ball. A score with three specific facts behind it is a decision aid.
- Can you turn features off? The serious vendors let you disable individual AI features per user or per feature. The decorators bundle everything into "AI mode."
Where relationship intelligence fits
Relationship intelligence platforms (Kith among them) are AI CRMs where the core AI job isn't drafting emails. It's inferring the relationship graph and the health of the portfolio from real communications data. The other AI features come along for the ride, but the primary value is understanding the account, not automating outreach.
Which one you need depends on the job. Transactional SaaS pipelines love drafting and scoring. Consultancies and technology partners live and die by the relationship graph. Buy for the job, not the branding.