Consultative Selling & Closing · Follow-up cadences and CRM hygiene · lesson 16 of 16 · 15 min

AI-assisted selling and call intelligence — useful, lawful and honest

Where AI now sits in the consultative sales process

AI is embedded across the sales stack: research summaries in LinkedIn Sales Navigator (Account IQ, Lead IQ), CRM assistants in HubSpot (Breeze) and Salesforce (Einstein and Agentforce), prospecting data and sequencing in tools such as Apollo, conversation intelligence in Gong and similar platforms, meeting summaries in Zoom and Microsoft Teams, and general assistants such as ChatGPT, Claude, Gemini and Microsoft Copilot. Features and plan names change often; check vendors' current documentation.

Mapped to this course:

| Stage | Useful AI assistance | Human responsibility | |---|---|---| | Prepare | Summarise public company information; draft hypotheses and questions | Verify facts; choose what to ask | | Discover | Transcribe and summarise (with consent) | Listen, probe, judge; capture exact buyer words | | Qualify | Flag missing MEDDICC fields from notes | Confirm with the buyer | | Propose | Draft sections from discovery notes | Check every claim and number; tailor | | Handle objections | Surface patterns across many calls | Respond to this buyer, in context | | Follow up | Draft recaps and value touches | Send personally; honour opt-outs | | Coach | Talk-ratio, monologue and topic analytics | Interpret fairly; coach humanely |

Conversation intelligence: what it measures

Call-recording and analysis tools typically report talk-to-listen ratio, longest monologue, question counts, topics mentioned (pricing, competitors, next steps), and sometimes deal-risk signals. These are useful prompts for coaching and pipeline review. They are not verdicts: a long monologue may have been a requested explanation; a "risk" flag may simply reflect missing data.

The limits: legal and ethical

Consent for recording and transcription. Follow the rules in the Remote and Video Selling lesson — announce, explain, ask, stop on objection, and apply the strictest applicable law (some US states require all-party consent).
Data protection. Recordings and transcripts are personal data. Under the UK/EU GDPR (and laws such as the UAE and Saudi PDPLs) you need a lawful basis, transparency, security, retention limits, and processor agreements with vendors. Don't paste transcripts or customer data into unapproved tools.
Emotion recognition. Be wary of tools claiming to detect emotions or deception from voices or faces. The EU AI Act has, since February 2025, prohibited AI systems that infer the emotions of people in workplaces from biometric data (with narrow exceptions), and the science behind such inferences is contested. Using such features on your own sales team in the EU is likely to be unlawful; check with legal before using any "sentiment" feature that goes beyond text.
Transparency. If a customer interacts with an AI agent or chatbot, tell them. In the EU, AI Act transparency obligations (Article 50) have applied since 2 August 2026. Never let an AI impersonate a named human rep without disclosure.
Accuracy and accountability. AI summaries can omit nuance or invent commitments. The rep owns what goes to the customer and into the CRM.
Fairness in coaching. Analytics can penalise reps whose buyers speak differently, or who sell in cultures where longer relationship-building conversations are normal. Calibrate benchmarks by market and deal type.

Hands-on: a team AI-use policy (one page)

AI IN OUR SALES PROCESS — POLICY (v1)
Approved tools: [CRM AI], [conversation-intelligence tool], [assistant with business terms]
Never allowed: unapproved tools for customer data; emotion/deception detection;
  AI messages sent without human review; AI pretending to be a named person.
Recording: announce + consent on every call; stop on objection; delete after [N] days.
Research: public sources only; verify before using; no scraping.
Customer-facing AI: always disclosed as automated.
Coaching analytics: used for development conversations, not automated ratings;
  benchmarks set per market and deal type.
Owner: [name]. Review: every 6 months.

Hands-on: a prompt for turning your notes into a CRM update

From my call notes below, fill these fields ONLY with information in the notes:
Pain (buyer's exact words), Metrics, Economic buyer, Decision process,
Paper process, Champion, Competition, Next step + date.
Write "unknown" where the notes don't say. Don't infer or guess.
[paste notes — remove personal data not needed]

Worked example

A Karachi software house selling to UK and Gulf clients rolls out a conversation-intelligence tool. They publish a one-page AI policy, add a consent line to every calendar invite and a spoken consent step at the start of calls, disable any voice-based "sentiment" features, and set talk-ratio benchmarks separately for UK and Gulf deals after noticing that Gulf calls naturally include longer relationship conversation. Managers use the analytics to choose which moments to review together, not to rank reps. Within a quarter, reps' notes are more complete and next steps more consistent — and no customer has complained about recording.

Measuring AI's contribution

Track time saved on admin, CRM completeness (for example, share of deals with all MEDDICC fields), next-step consistency, and quality signals (customer complaints, opt-outs, errors found in AI summaries). If speed rises but quality falls, adjust.

Next step

For a full, hands-on treatment of AI across prospecting, research, call analysis, forecasting and governance, continue with AI for Sales Teams (ai-for-sales-teams) in the Optimize All Academy.

Video lecture: AI-assisted selling and call intelligence — useful, lawful and honest

Lecture coming soon · 11 chapters · about 8 minutes. Read the full transcript below.

AI-assisted selling
Why now
AI by stage
What call intelligence measures
Six limits
Example 1: an invented commitment
Example 2: Karachi software house (illustrative)
Watch me do it: notes → CRM
One-page AI policy
Mistakes + measures
Recap + try this now

Lecture transcript

AI-assisted selling

Imagine every sales call your team makes is recorded, transcribed and analysed. Every question counted, every competitor mention flagged, every next step logged automatically. That's not the future. For many teams, it's this quarter. Used well, it makes sellers better and buyers better served. Used badly, it breaks the law, erodes trust and turns coaching into surveillance. In this lecture you'll learn where AI fits across the consultative sales process, what call-intelligence tools actually measure, six legal and ethical limits, and how to write a one-page policy your team can follow.

Why now

Why does this matter now? Because AI is already built into the tools sellers use every day. Research summaries in LinkedIn Sales Navigator. CRM assistants in HubSpot and Salesforce. Prospecting and sequencing in tools like Apollo. Conversation intelligence in Gong and similar platforms. Meeting summaries in Zoom and Teams. And general assistants like ChatGPT, Claude, Gemini and Copilot. Features change constantly, so always check the vendor's current documentation. The question isn't whether to use AI. It's which parts of the job to hand it, and which parts must stay human.

AI by stage

Here's the helpful mental model. AI is a very fast assistant who never gets tired of admin, but who sometimes misremembers what was said and has no sense of the relationship. So map it to the process. Prepare: AI summarises public information and drafts hypotheses; you verify and choose what to ask. Discover: AI transcribes, with consent; you listen, probe and capture the buyer's exact words. Qualify: AI flags missing fields; you confirm with the buyer. Propose: AI drafts sections; you check every claim. Follow up: AI drafts recaps; you send them. And coach: AI surfaces patterns; managers interpret fairly and coach humanely.

What call intelligence measures

What do conversation-intelligence tools actually measure? Typically: talk-to-listen ratio, longest monologue, how many questions you asked, which topics came up, like pricing, competitors or next steps, and sometimes deal-risk signals. These are useful prompts for coaching and pipeline reviews. But they're not verdicts. A long monologue might have been an explanation the buyer asked for. A risk flag might just mean data is missing. Think of them like a car's dashboard warning lights. Worth checking every time, but you still look under the bonnet before you decide what's wrong.

Six limits

Now the six limits. One: consent for recording and transcription. Announce, explain, ask, stop on objection, and apply the strictest applicable law. Two: data protection. Recordings and transcripts are personal data, so you need a lawful basis, security, retention limits and proper agreements with vendors, and you never paste them into unapproved tools. Three: emotion recognition. Since February twenty twenty-five, the EU AI Act has prohibited AI systems that infer the emotions of people in the workplace from biometric data, with narrow exceptions, and the science behind reading emotions from voices or faces is contested. Four: transparency. If customers talk to an AI agent, tell them; in the EU that's been a legal duty since August twenty twenty-six. Five: accuracy and accountability. Six: fairness in coaching.

Example 1: an invented commitment

First example, the simple one: accuracy. After a call, the AI summary says: the buyer committed to sign by Friday. Your own notes say: the buyer will review the proposal by Friday. That's a big difference. If the AI version goes into the CRM, your manager forecasts a deal that isn't closing, and if it goes into your follow-up email, the buyer thinks you weren't listening, or worse, that you're pressuring them. So you correct it. Review by Friday. The rule is simple: the rep owns what goes to the customer and into the CRM, however it was drafted.

Example 2: Karachi software house (illustrative)

Now a realistic business scenario, with illustrative details. A software house in Karachi sells to UK and Gulf clients and rolls out a conversation-intelligence tool. They do five things. They publish a one-page AI policy. They add a consent line to every calendar invite and a spoken consent step at the start of every call. They disable any voice-based sentiment features. They notice that Gulf calls naturally include longer relationship conversation, so they set talk-ratio benchmarks separately by market. And managers use the analytics to choose which moments to review together, not to rank reps automatically. Illustratively, within a quarter, CRM notes are more complete, next steps are more consistent, and no customer has complained about recording.

Watch me do it: notes → CRM

Watch me do it. I'll use an assistant to turn my call notes into a CRM update, safely. I paste my notes, with personal details I don't need removed. Then I write: fill these fields only with information in the notes. Pain, in the buyer's exact words. Metrics. Economic buyer. Decision process. Paper process. Champion. Competition. Next step and date. Write unknown where the notes don't say. Don't infer or guess. Here's the output. Pain: quoted correctly. Economic buyer: the finance director. Paper process: unknown. Good, that's honest, and it tells me what to ask next. Competition: it wrote, likely a larger vendor. My notes never said that. So I change it to unknown. Then I paste the checked version into the CRM.

One-page AI policy

Here's the one-page policy from the lesson, in brief. Approved tools, named. Never allowed: unapproved tools for customer data, emotion or deception detection, AI messages sent without human review, and AI pretending to be a named person. Recording: announce and get consent on every call, stop on objection, and delete after a set number of days. Research: public sources only, verified, no scraping. Customer-facing AI: always disclosed. Coaching analytics: for development conversations, not automated ratings, with benchmarks by market and deal type. And an owner, with a review every six months. One page. Everyone knows the rules.

Mistakes + measures

Common mistakes. Trusting AI summaries without checking. Letting AI send messages on your behalf. Recording without clear consent. Pasting transcripts into consumer tools. Turning analytics into a league table that punishes reps selling in relationship-first markets. And buying tools that promise to read buyers' emotions. How do you measure AI's contribution? Time saved on admin. CRM completeness, like the share of deals with all qualification fields filled. Next-step consistency. And quality signals: complaints, opt-outs, and errors found in AI summaries. If speed rises but quality falls, adjust.

Recap + try this now

Recap, and the end of this course. AI helps you prepare, summarise, draft and coach, while you verify, decide and stay accountable. Treat call analytics as prompts, not verdicts, and calibrate them fairly. Get consent, protect data, avoid emotion recognition, and disclose customer-facing AI. Your try-this-now action: draft a one-page AI-use policy for your team using the template, and share it with your manager. For a deeper, hands-on course on AI across the whole sales process, continue with AI for Sales Teams. Thank you, and go and diagnose before you prescribe.

Key takeaways

AI helps prepare, summarise, draft and coach — humans verify, decide and stay accountable.
Conversation-intelligence metrics are coaching prompts, not verdicts; calibrate them by market.
Recording needs consent; transcripts are personal data; approved tools only.
Avoid emotion-recognition features (prohibited in EU workplaces since Feb 2025) and always disclose customer-facing AI.

Try it

Draft a one-page AI-use policy for your sales team using the template, including approved tools, recording consent and prohibited uses, and share it with your manager.