Every CRM vendor now advertises AI: forecasts, lead scoring, automatic summaries, copilots. And yet daily life in many sales teams looks unchanged: the forecast is gut feeling, half the deals are outdated, and what was discussed in the last customer meeting is written down nowhere.
This article sorts out what AI in CRM actually delivers today, where it fails in practice, and why the most effective lever is not the smartest model but the most inconspicuous task: data maintenance.
What AI in CRM can actually do today
Behind the umbrella term sit three classes of features at different levels of maturity. Keeping them apart makes vendor promises much easier to judge.
Capturing data: the underrated part
Documenting conversations, emails and appointments automatically and attaching them to the right contacts. It sounds mundane, but it is the class with the highest practical value, because it replaces exactly the manual work on which CRM discipline fails in daily business. Maturity: high, speech recognition and matching work reliably today. And as we are about to see, the value of every other class depends on this first one.
Recognising patterns: scoring and forecasts
Lead scoring, revenue forecasts, churn risks, next-best-action suggestions. The showpiece discipline of CRM vendors and what impresses most in demos. The models are good, but they have a silent prerequisite: a complete, current data foundation. Scoring on well-maintained data separates hot from cold leads with remarkable accuracy. The same scoring on patchy data sorts at random, just with a confident percentage next to it.
Automating routine: triggers and workflows
Follow-up reminders, activity logs, proposal tracking, workflow triggers such as "when the deal stage changes, notify the team". Saves click work and reduces forgotten next steps. Maturity: high, but the same rule applies: a workflow that triggers on an outdated deal status automates the mistake along with it.
We looked at the long-term trajectory earlier in our article on how artificial intelligence is changing the future of CRM systems. This one is about today, and today there is a catch.
The catch: AI computes on data that does not exist
How the gaps appear
All forecasting and scoring features share one prerequisite: they work on the data that is in the CRM. And that is exactly the problem. In practice, CRM data is chronically incomplete: deals without a current stage, contacts without the latest state of the conversation, required fields filled with placeholders, notes that exist in the sales rep's head and nowhere else.
This is not laziness, it is a structural conflict of goals. Spending 10 minutes maintaining fields after an hour-long customer meeting, while the next call is already waiting, is the task that gets sacrificed first in daily business, and rationally so: the customer pays for conversations, not for records. Required fields and admin reminders do not change that, they just produce more creative placeholders. Why sales teams often experience their CRM as a burden is something we described in detail in why sales reps hate CRM systems.
Forecasts built on sand
The consequence for AI: a model that scores and forecasts on patchy data produces precise-looking numbers on a wrong foundation. The forecast says 78% win probability because the deal has sat in "negotiation" for three weeks, except the last real conversation was a rejection that never got entered. That is more dangerous than no forecast at all, because the number suggests a certainty that does not exist, and managers make decisions on top of it. Anyone evaluating AI features in a CRM should therefore start with one uncomfortable question: how current is our data, really?
The most effective use of AI: start at the source
Conversations are the source of almost all CRM data
If data quality is the bottleneck, then the most valuable AI in CRM is not the one that computes at the end, but the one that documents at the beginning. Because where does CRM data actually come from? Almost everything a CRM is supposed to know was said at some point: in the discovery call, in the proposal meeting, in the phone call about the contract renewal. Between the conversation and the record sits a human who has to transfer it, and that is exactly where the chain breaks. Sally starts at this point.
Natively integrated instead of hand-built
Sally attends customer conversations, in Google Meet, Zoom, Microsoft Teams and Webex, and covers calls and in-person appointments via the app. After the conversation, the summary, agreements and recognised tasks are created, and they land directly on the right contact or deal in the CRM. 7 systems are supported natively: HubSpot, Salesforce, Microsoft Dynamics 365, Pipedrive, Zoho, Odoo and Bitrix24. The difference to hand-built Zapier chains: native integrations know the CRM's object structure, create activities on the deal and survive schema changes instead of breaking with every adjustment. On top come 4 automation platforms (Zapier, Power Automate, make.com, n8n) and 8,000+ tool connections for everything beyond.
What flips in daily business
This flips the logic:
- Maintenance happens automatically, not when someone finds time. Every conversation leaves a complete, current record.
- The state of the conversation is readable, verbatim and with a source, instead of a three-word note.
- Follow-ups emerge from the content: the next steps from the meeting are recognised as tasks and tracked. How that turns into the email after the meeting is covered in our article on the follow-up email after a meeting.
- And only now do the computing features pay off: scoring and forecasts on complete conversation data are a different product than on empty fields.
Looking ahead, Sally goes one step further: the assistant that knows your conversations takes instructions directly, from CRM entries on demand to email drafts built from the conversation context. The direction is clear: less maintenance work, more selling time.
The GDPR point many overlook
Conversation data is personal data, often the most sensitive in the entire CRM. Many AI features of large CRM vendors run on US clouds, which requires a transfer assessment after Schrems II and is often a knockout criterion in regulated industries. Sally is operated by Aliru GmbH and processes exclusively in Germany. For the recording itself: only with the consent of everyone involved, and the visible bot in the meeting creates that transparency. Details on the page about GDPR and security.
Conclusion
AI in CRM is not a single feature but a chain: capture, understand, predict, automate. The chain breaks at its weakest link, and in almost every company that is manual data maintenance. If you deploy AI first where the data is created, in the conversation, you get a CRM that stays current without anyone maintaining fields. Everything else builds on top of that.
Which CRM systems are worth considering in the first place is covered in our overview of the best CRM systems. And if you want to see your conversations land in the CRM automatically: test Sally free for 30 days, native integration included.
