AI Call Analysis: every recorded call scored, every lead given a next step

If your team sells over the phone, you already know the problem. A rep makes dozens of calls a day. The manager can listen to a handful. The rest turn into a one-line note, or nothing at all, and the lead’s fate depends on how the rep felt about the call at 6 pm.

So good leads go cold because nobody flagged them, and bad leads get chased for weeks because nobody said they were never going to buy. We built AI Call Analysis to fix that: every recorded call is transcribed and scored, and each lead gets a plain recommendation the whole team can see.

How it works

Reps call from MicroSIP on Windows or Nerd Dial, our own softphone, on Mac, and the NerdCRM recorder uploads each recording, which is linked to the right contact by phone number. If you have not set that up yet, the call recording setup guide walks through it.

From there:

1. The call is transcribed. Speech to text runs with Bangla and English hints and separates the speakers, and the model works out which one is your agent.
2. The call is scored on six signals, each from 0 to 5: fit, need, decision maker, budget, timing and engagement. Each comes with a short note and, where there is one, a quote from the call. No evidence means a low score, not a guess.
3. The score and verdict are computed in code, not by the model. The signals are weighted into a 0 to 100 score and fixed rules turn it into a verdict, so the same signal scores always give the same verdict. You can tune the thresholds in settings.
4. The latest conversation decides. A contact’s status follows its most recent real conversation, by when the call happened. Voicemail and calls that did not connect do not count.

The contact gets one of four recommendations:

• Qualified: prioritise, push for a demo and close.
• Nurture: keep warm with occasional follow ups.
• Needs another call: shown as “Call again”, with the open questions to ask next time.
• Disqualified: do not spend sales time on this lead.

After two real conversations (you can change the number), a lead is always decided, so nobody stays in “call again” forever.

Where the team sees it

Every deal card on the pipeline board carries its contact’s recommendation and score, with the reason on hover and an AI filter on the board. The deals list has an AI recommendation column, a filter and two extra CSV export columns. The contact and deal pages show the score, the reason and the questions for the next call, and each recording has its own panel with the summary, signals and transcript.

A person can set the status by hand on the contact page. While it is overridden, the AI keeps scoring in the background but never changes the status. One click on Use AI status hands it back.

Notes count too

A lot of selling happens outside recorded calls: a WhatsApp confirmation, a demo, a promise to pay next week, written up in a note. So when a contact has at least one analysed call and at least one note, on the contact or any of its deals, the lead is judged again from the latest call, short summaries of earlier calls and the most recent notes. That judgement takes the place of the latest call in the rules above, and the card says “Based on the latest call and N notes”. Notes alone never create a status. A contact needs a call first.

Rolling it out safely

We did not want the AI quietly rewriting your contacts on day one, so it starts in shadow mode: it scores and stores everything and changes nothing. A sensible rollout:

1. In Settings → Call Analysis, turn on automatic analysis and keep shadow mode.
2. Fill in What you sell: your product, your ideal customer, your packages and prices, and who is not a fit. Every call is judged against this text, so be specific.
3. Run Analyse missed calls on recordings you already have, for example the last two weeks.
4. Open Analytics → Lead Quality and look at contacts your team knows well. Do the verdicts match what your best rep would say?
5. When they do, switch to live.

In live mode, each contact keeps exactly one lead tag: lead:qualified, lead:nurture, lead:in-progress or lead:disqualified. Because they are ordinary tags, your existing contact filters and WhatsApp campaigns can target them. If you want, live mode can also set contact priority to high for qualified leads and low for nurture and disqualified ones.

What it costs

AI Call Analysis is an add-on at $5 per user per month on top of the $4 CRM plan. It includes 300 analysed calls per user per month, pooled across the team. If you need more, extra calls are $3 per 100, only up to a monthly limit the owner sets, or none at all. Paid plans without the add-on get 20 analysed calls per user per month to try it on real calls. The free plan has no AI. Full details are on the pricing page.

Recordings are kept for 45 days. The scores, summaries and recommendation stay on the contact after the audio is gone. The call recording feature page covers the recording side.