AI for Sales Teams: A Practical Daily Playbook

Most of a sales rep's day is not selling

Ask a rep at a 30-person B2B company to track a week honestly and the pattern is familiar. Perhaps a third of the time is spent talking to prospects and customers. The rest goes on researching accounts, updating the CRM, writing follow-up emails, chasing internal colleagues for a quote and trying to remember what was promised on a call three weeks ago.

That second pile is where AI for sales teams earns its keep. Not in replacing the conversation, which is the part buyers still want from a person, but in shrinking the work around it. If a team of eight reps each gets back 45 minutes a day, that is six hours of selling time daily, which is close to an extra rep without the recruitment.

The trick is picking the tasks where a wrong output is caught quickly and costs little. Here is how we would lay out a rep's day.

Before the call: account research in two minutes

Pre-call research is the easiest win. A rep about to speak to a logistics firm wants to know what the company does, roughly how big it is, what has changed recently, what our history with them is and who else in the account has spoken to us.

An AI assistant can pull that together from the CRM, past email threads, support tickets and the company's own website, then produce a short brief. We covered the build side of this in our piece on an AI application for sales research. The key design choice is that every claim in the brief links back to its source, so a rep can check anything that looks odd before repeating it on a call.

• Company summary from their website and public filings
• Our own history: past deals, open quotes, support issues in the last 90 days
• Contacts we know, with the last time each one replied
• Three questions worth asking, based on the gaps in what we know

During and after the call: notes that write themselves

Call summaries are the second win and, in our experience, the one reps like most. The call is recorded and transcribed, a model extracts the pain points, budget signals, next steps and named stakeholders, and the result is proposed as a CRM update for the rep to accept or edit.

The word proposed matters. A model that writes straight into the CRM will eventually record a next step nobody agreed to, and then forecasts are built on fiction. One click to accept keeps the rep accountable and still removes ten minutes of typing per call.

Consent is not optional here. Tell people the call is being recorded and transcribed, and check your obligations in each country you sell into, because the rules differ between the UK, the US states and the Gulf.

Follow-ups and proposals: drafts, not sends

A good follow-up email references what was actually said, restates the next step and attaches the right material. With the call summary already in hand, drafting that email is a small, reliable task. The rep reads it, fixes the tone and sends it.

Proposal and quote drafting follows the same pattern, with one hard rule: prices, discounts and terms come from your pricing system, never from the model. We wrote more about that boundary in our post on AI proposal and quote generation.

Pipeline hygiene: flagging deals that have gone quiet

Every sales manager knows the deal that has sat at 'proposal sent' for 60 days with a confident close date. AI is decent at spotting these, because the signals live in text: no reply to the last three emails, a champion who has left the company, a competitor mentioned in the last call.

A weekly digest that says 'these five deals look stale, here is why' is more useful than a score nobody trusts. Keep the reasoning visible. A manager who can read 'no inbound reply since 2 August, last call mentioned budget freeze' will act on it. A manager handed a number between 0 and 100 will not.

Where AI makes sales worse

The loudest sales AI pitch right now is fully automated outbound: an agent that finds prospects, writes personalised emails and books meetings. Some of it works. A lot of it fills inboxes with plausible, slightly wrong messages that buyers have learned to spot within a line.

• Personalisation at volume. 'I saw your recent post about growth' fools nobody now. Sending thousands of these burns your sending domain.
• Deciding who to call. Lead scoring can help, but a model trained on a small, messy CRM mostly learns your past biases.
• Negotiation and objections. Buyers want a person who can make a commitment. A model cannot.
• Anything that makes a promise. Delivery dates, contract terms, integration claims. These need someone who will be there when the promise comes due.

If you do want an agent booking meetings, read our playbook on AI sales agents first and start with warm, inbound leads rather than cold lists.

How to roll it out without annoying the team

1. Pick one task, usually call notes, and one team of three or four reps willing to try it
2. Run it for four weeks and measure time saved and edit rate on the drafts
3. Fix the complaints, which will mostly be about CRM field mapping rather than the AI
4. Add account research once notes are trusted
5. Only then look at follow-up drafting and pipeline flags

When we build this kind of thing at SpiderHunts, most of the work is connecting the CRM, the calling tool and email cleanly. The model is the easy part. If you would rather have it built properly than stitched together from five subscriptions, that is what our AI integration service covers.

Frequently asked questions