AI for sales teams: a practical guide

This guide is for sales leaders, account executives, and sales managers at B2B companies. By the end you'll have one shared Sales Space in type.com running seven workflows, each with copyable instructions, a schedule, a review rule, and the failure modes to watch.

AI reads and checks: CRM history, call transcripts, email threads, the answer library, and the pipeline. People decide and act: what to say on the call, what to forecast, what changes in the CRM, and what reaches the customer. Nothing reaches a buyer or overwrites a field until a named person says yes.

This guide starts when a meeting is booked and ends at signature. For prospecting and outbound, see AI for SDRs. For onboarding, renewals, and expansion, see AI for customer success. The examples follow a fictional company, Lumenfield, whose six AEs sell procurement software to mid-market manufacturers using Salesforce, Gong, Gmail, and Google Drive. Today in the examples is Wednesday, Oct 7, 2026, one week into Q4.

A common pattern is one bot per job (a prep bot, a forecasting bot, a slides bot) plus a coordinator bot to manage them. Each has its own logins, rules, and memory, and the sales manager can't see what any of them did. In type.com you create one Space for the department instead. Everyone shares the same connections, skills, memory, and AI setup. Specialist jobs become shared skills, and routines become automations that post where the team can see them.

Connect the CRM and call recorder as organization connections, and keep email and calendars personal so each AE's AI reads only their own mailbox. A Read-only setting tells the agent not to make changes, but it doesn't block changes in the connected service, so connect Salesforce with an integration user that can't edit until you turn on approved writes in week three.

When it runs: on demand, when an AE types /account-brief and a company in the accounts channel, usually before a first meeting or a stage change. Owner: the opportunity's AE. Inputs: the CRM, transcripts, the AE's email, and the company's own newsroom and careers pages.

This is not a prospecting packet. The SDR question, "why reach out now?", is covered in the SDR guide. The AE question is "who decides, what have they told us, and what don't we know?" So the brief puts the buying committee and past conversations ahead of news.

Before Hannah's first meeting with Harlan Castings on Thursday, Oct 8, the brief found a 2025 evaluation that ended in no decision after their CFO left, a June 2026 press release naming a new CFO, and a Sep 21 job post for a procurement systems analyst. It also listed what nobody knew: who owns the budget now. That became Hannah's first discovery question.

Review rule: for two weeks, the AE opens every source link. After that, spot-check one claim per brief. If a brief invents a person or a title, fix the skill that day.

Shortcut for agencies and services firms: when a prospect has shared real data, the New Business Pitch Builder skill in the type.com Skills Library turns that audit into a prioritized gap list with confidence-discounted sizing and a 90-day engagement outline.

When it runs: weekdays at 7:00 AM PT, posting to the calls channel. Owner: each AE, who creates their own copy, because scheduled runs use the creator's identity and credentials. Nadia's run reads only Nadia's calendar and mailbox. Inputs: calendar, CRM, transcripts, and email.

Most reps prep by skimming the opportunity a minute before the call. That misses change: a new name on the invite, an amount above the budget the buyer gave, a promise the team didn't keep. A good brief puts those first.

Nadia's 10:00 AM call with Brennan Tool & Die got four flags. CFO Maria Santos was joining for the first time. The $84K amount sat above the $70K budget the champion, Owen Price, mentioned on Sep 16. A security questionnaire promised for Sep 30 hadn't gone out. The CRM next step hadn't changed since Sep 18. Nadia opened by owning the late questionnaire and asking Maria how approvals work.

Review rule: the AE checks any flag before raising it with the buyer. The brief never goes to the customer.

Shortcut: the Pre-Call Briefing skill in the type.com Skills Library packages this with tested scripts. It flags stale records, budget conflicts, unanswered questions, overdue commitments, new stakeholders, and competitor landmines, and can run on a short schedule that briefs each call just before it starts.

When it runs: on demand, when the AE types /call-debrief-review and the account after a call. A second automation at 4:30 PM PT lists today's recorded external calls in Gong that have no debrief, or a debrief with no approval. Owner: the AE. Inputs: the transcript, the opportunity, and the email thread.

Unreviewed CRM automation corrupts a forecast quietly. A model that writes "budget approved" because the buyer said "budget shouldn't be a problem" creates a fact nobody checked. So the output is a proposal: each row shows the field, current value, proposed value, and the quote behind it.

On the Brennan call, the buyer moved the decision to Nov 13 because IT wanted a security review first, and the CFO asked to start with two plants rather than three. The debrief proposed five changes. Nadia approved four, corrected the amount from a naive $56K to the $72K two-plant quote, and the AI wrote exactly those rows. Until week three, while the CRM is read-only, the AE copies approved rows by hand.

Review rule: the owner approves rows by number. Stage and forecast category changes are never inferred from tone; they go in an "Ask the rep" list.

Shortcut: the Call Debrief skill in the type.com Skills Library turns a transcript into a debrief, a CRM update, and a drafted follow-up for HubSpot, Attio, or Salesforce. Its defaults confirm before any CRM write and keep the follow-up as a draft. Keep those defaults.

When it runs: Mondays at 7:00 AM PT in the forecast channel, and on demand before the Tuesday 1:00 PM PT deal review. Owner: the sales manager. Inputs: open opportunities, stage history, activity dates, and transcripts for deals in review.

A deal with no dated next step, or a close date already past, is the first place a forecast lies. The rules below are mechanical on purpose, so every line on the list is defensible.

Lumenfield's Monday, Oct 5 run found 14 issues across 46 open deals: 6 with no next step, 4 with a past close date, 3 with no activity in 21 days, and 1 with no amount. For the deals going to Tuesday's review, the packet added qualification gaps. Brennan Tool & Die, an $84K commit deal, had no contact with the economic buyer and an overdue security questionnaire.

Review rule: AEs fix their own records. The manager picks which review questions to ask.

Shortcut: the Pipeline Hygiene Audit skill in the type.com Skills Library runs these checks as deterministic rules over a CRM deal export, including deals stuck beyond their stage median and duplicate accounts. It judges the record, not the deal, so pair it with the qualification questions below.

When it runs: on demand in the deals channel after a call that moved dates, owners, or the path to signature. Owner: the AE. Inputs: the debrief, transcript, opportunity, and email thread.

A mutual action plan is the most useful document in a complex deal and the one most often skipped, because it takes an hour by hand. The draft starts from the dates the buyer actually said and works backward from their deadline, which shows whether security, legal, and procurement fit before the close date.

Brennan's plan ran from Oct 7 to signature on Fri, Nov 13: seven steps, each with an owner at both companies. The email draft was four short paragraphs: what Maria asked for, the next two steps, a pricing placeholder for the deal desk, and one question. Nadia rewrote the opening line and sent it from Gmail at 11:40 AM.

Review rule: the AE edits and sends everything. Dates the buyer didn't confirm are marked "proposed", and pricing and legal terms stay as placeholders for the deal desk and legal.

For how approvals work in a shared thread, read AI agent approval workflows.

When it runs: when an RFP, security questionnaire, or proposal request arrives and the AE posts it in the proposals channel. Owner: the solutions engineer, with the AE. Inputs: the file, the approved answer library and security documents in Drive, and the opportunity.

The first question of any RFP is whether to bid. On Oct 2, Ostrava Metals sent a 142-requirement RFP due Fri, Oct 23. The first pass matched 96 requirements to approved answers, found 31 needing SE edits, and found 15 with no answer, 8 of them mandatory. That gap list drove the bid decision and then became Ivy's work list.

Review rule: the SE approves technical answers, legal approves terms and privacy, and the deal desk owns pricing. The AI never fills a mandatory gap with a plausible guess, because a wrong "yes" on a security questionnaire can become a contract obligation.

Shortcut: the RFP Response Builder skill in the type.com Skills Library scores each requirement against your answer library and reports coverage, weighted points at risk, and unanswerable gaps. Its matching is keyword-based, so library answers need real substance.

When it runs: Mondays at 7:30 AM PT, after the hygiene run, in the forecast channel. Owner: each sales manager, for their team. Inputs: open opportunities with forecast categories, last Monday's snapshot, activity, and the last call for each commit deal.

Here AI is an evidence checker, not a forecaster. Reps forecast from conviction, and the manager's job is to test it. The review lines each commit deal up against the record (economic buyer met, a dated next step, a plan that fits the close date, recent activity) and lists what doesn't fit.

The Monday, Oct 5 review showed a $1.8M Q4 target, $96K closed, $640K commit, $1.05M best case, and $3.2M of open pipeline, 1.9x coverage of the remaining $1.7M. Three commit deals had gaps, including Brennan. That flag is why Nadia's Wednesday brief led with the CFO.

Review rule: the manager reads it before 1:1s and decides what to ask. The AI never changes a category or the submitted number.

The setup is the same pattern used to automate a weekly team report: a narrow scheduled job, a skill for the rules, and one channel the team reads.

When it runs: the first Thursday of each month at 9:00 AM PT in the win-loss channel, plus once after each quarter closes. Owner: the VP of Sales or enablement. Inputs: opportunities closed in the period, their closed reasons, transcripts, and email threads.

CRM loss reasons are picked from a dropdown in ten seconds, and "price" is the easiest pick. On Thursday, Oct 1, the Q3 review read 31 closed deals: 12 won, 19 lost. The CRM said price for 9 losses. The calls said no decision in 7, an ERP add-on module in 5, price in 4, and timing in 3.

Won deals had met the economic buyer before the proposal in 9 of 12; lost deals in 5 of 19. Rachel, the VP of Sales, made "economic buyer met" an exit criterion for the Proposal stage, which is also why it's in the forecast evidence check.

Review rule: team patterns go to the win-loss channel. Notes on one rep's calls stay in a private thread between that rep and their manager. With samples this small, label findings "directional" and confirm with buyer interviews before changing pricing or the pitch.

Each run lands about an hour before the meeting it feeds, in a channel the team already reads. All times are America/Los_Angeles.

Write these rules into the instructions and skills, then enforce them with permissions in the tools themselves. Instructions guide the model; they aren't a security control.

In type.com, AI acts with the effective access of the person who started the work, and a scheduled automation runs as its creator. If the manager who created the forecast review loses CRM access, it fails and pauses after three failed runs, so give each important automation a second owner.

Data handling: keep the Space private, assign the CRM and call recorder only to it, keep secrets out of prompts, follow your call-recording consent rules, and keep pricing exceptions and other customers' names out of anything drafted for a buyer.

Pilot with one busy AE and one manager. A skeptical top performer is the best pilot, because their corrections tighten the skills fastest.

Take the baseline in week one: four weeks of CRM data, plus two AEs timing prep and debriefs for ten calls. Win rate and cycle length move over quarters, so track them, but don't judge a 30-day pilot on them.

When a workflow drifts, fix the skill, not the thread, so every rep gets the correction. Read how shared AI memory works for teams, and for a full list of skills, see the best AI skills for B2B teams.