How do I get started with AI?

Sales

Research prospects instantly, automate post-call admin, and flag deals that are slipping.

Updated 4 days ago · 10 sources

Proof over hype

  • Enterprise

    Microsoft

    Microsoft's own sellers spent their days jumping between MSX, Dynamics 365, Outlook, Teams and half a dozen other systems just to log a note or check on a deal, described internally as a 'sales tax' on time that should have gone to customers.

    Per-seller revenue up 9.4%, opportunities per seller up 5%, and individual win rates up 20%, measured across Microsoft's own sales teams in the months after rollout.

    The takeaway The software didn't do the hard part. The listening circles and change management that got sellers to actually stop using personal notebooks did — a tool nobody adopts doesn't move any of these numbers.

    Microsoft Inside Track, Microsoft's own IT blog · Aug 2025

    How they did it

    Microsoft rolled Copilot for Sales out to its own sales organization, starting with a pilot in its small, medium and corporate sales group before scaling company-wide. Inside Outlook and Teams, it surfaces deal context and recent emails before a call, drafts replies using CRM data and BANT analysis (budget, authority, need, timing — the standard framework for qualifying a deal), and during the call itself, captures action items, sentiment and who said what, then prompts the seller afterward to log notes and create tasks without ever opening the CRM directly. A seller can just ask, in plain language, 'what's the latest with this opportunity?' and get an answer instead of digging for it. The rollout leaned as hard on change management as on the software itself: learning days, live demos, and role-specific listening circles to surface what was actually stopping sellers from adopting it, since plenty had spent years defaulting to personal notebooks and spreadsheets instead of the CRM. Principal technical specialist Bob Lincavicks summed up what changed: 'The CRM used to feel like a tax. Now it delivers value.'

    Tool used Microsoft

  • Enterprise

    monday.com

    monday.com's inbound demand was arriving faster than its roughly 1,000-person go-to-market team could work through it, and a demo request could sit for a full day before anyone replied.

    Demo request replies fell from 24 hours to under two minutes (Amanda). Trial users who interacted with Jax converted at 2.5x the rate of a control group, across 3,000-plus monthly calls. Account research that took 1 to 2 weeks by hand now takes about 5 minutes with Oscar.

    The takeaway Speed to lead is where this paid off fastest and most measurably — the 24-hours-to-2-minutes number is the kind of gain that shows up in pipeline within a quarter, before you'd see anything from the harder, judgment-heavy parts of the funnel.

    Growth Unhinged, Kyle Poyar · Jun 2026

    How they did it

    VP of AI for GTM Oran Akron, who had spent eight years building monday's RevOps function from zero to 80 people, set up an internal team called RevAI to build three purpose-built agents rather than buy one generic tool. Amanda handles inbound: she runs a roughly five-minute qualifying voice call using the BANT framework, matches the caller's accent and picks a local phone number within 60 seconds, discloses upfront that she's AI, and hands qualified leads to a rep with full context already pushed into Slack and the CRM. Jax lives inside the free trial product itself as an in-app avatar, reading what a new user is trying to do and configuring their trial instance to get them to value faster, effectively a sales engineer available at 2am. Oscar handles outbound account research, pulling from Clay, LinkedIn, financial reports and even podcast appearances to build a full account plan and suggested outreach sequence automatically.

    Tool used built in-house

  • Enterprise

    Microsoft

    The chat agent on Microsoft's own product pages, Ask Microsoft, was built on rigid, pre-scripted topic flows, and as traffic and the number of things it needed to know about grew, response times got noticeably slower for buyers waiting on a simple presales question.

    61% lower latency on the Microsoft 365 site, a 70% drop in chats that needed a human, a 16% increase in Azure trial signups, and visitors who used the agent were ten times more likely to move forward with signing up than those who didn't.

    The takeaway The 70% drop in human-handled chats is the number worth watching, not the headline 10x signup figure — it's the one that tells you the agent is actually resolving questions, not just being present.

    Microsoft Customer Stories · Feb 2026

    How they did it

    Microsoft rebuilt Ask Microsoft on Copilot Studio using multi-agent orchestration: a coordinator agent routes each question to specialized sub-agents that can be consulted individually or together for questions spanning multiple products, and answers are shaped by where on the site the visitor actually is — broader answers on a general product page, narrower and signup-focused answers on a trial page. For pages too long for Copilot Studio's own indexing to handle, it hands off to Microsoft Foundry agents to keep working. When the agent genuinely can't help, it transfers the visitor to a live chat rep instead of leaving them stuck. Director Alyse Muttera's framing was direct: 'Building a more advanced assistant with Copilot Studio has meaningfully raised the bar for our customer experience.'

    Tool used Microsoft Copilot Studio

  • Learn from thisMid-market

    Otter.ai

    Otter's AI notetaker joins Zoom, Google Meet and Microsoft Teams calls as a participant and transcribes everyone in the room, including people who never signed up for Otter and never agreed to anything.

    So: Get consent, spoken or written, from everyone on the line before a notetaker joins the call, not just your own team, and ask any AI notetaker vendor point-blank, in writing, whether they train models on your audio.

    The takeaway The exposure here isn't limited to your own reps and prospects. Anyone your rep is on a call with, including people who never touched your CRM or your tools, can be a plaintiff if the vendor recorded them without asking.

    The National Law Review · Aug 2025

    How they did it

    In Brewer v. Otter.ai, filed in California federal court in August 2025, the lead plaintiff, who has no Otter account at all, alleges Otter recorded and transcribed a meeting they were on without their consent, and that Otter then used those recordings to train its own speech-recognition and machine-learning models. That second part is what turns a consent problem into a much bigger one: it isn't just that a non-user's conversation got transcribed without asking, it's that the recording was allegedly repurposed to improve a commercial product neither the non-user, nor in some cases even the meeting host, had agreed to. The case is a class action, still in its early stages.

    Tool used Otter.ai

  • Learn from thisNewEnterprise

    Salesforce

    Salesforce sold Agentforce hard to its own sales customers as the fix for pipeline busywork, and told the market it was the fastest-growing product in company history.

    CIOs surveyed said the product wasn't ready to buy, and cited unpredictable pricing and unclear ROI as blockers, directly contradicting Salesforce's own claim that Agentforce is the fastest-growing product in company history, with named customers like Engine, Falabella and AAA live in weeks.

    So: Ask your CRM vendor to run a trial on your own data with your own usage pattern, get the pricing ceiling in writing before committing to anything consumption-based, and call two of their reference customers who aren't on the vendor's own case-study page.

    The takeaway Two things can both be true: Agentforce may genuinely be growing fast in signups, and the CIOs actually being asked to renew or expand it can still say it isn't ready. Growth numbers and buyer confidence are different metrics — ask for both.

    Salesforce Ben · Jul 2026

    How they did it

    A KeyBanc survey of chief information officers found the buyers Salesforce was selling to weren't convinced: they said the product 'isn't there yet,' pointed to unclear ROI, real implementation complexity and technical debt, and not enough real-world enterprise success stories to justify the spend. The pricing model drew specific criticism too: usage-based consumption pricing, especially around a feature called Headless 360, raised fear of runaway costs from agents acting autonomously with no hard ceiling. One analyst's read on the reaction: 'CIOs have become highly sensitive to unpredictable consumption pricing after a decade of cloud cost overruns.' There's also a structural objection underneath all of this that has nothing to do with product quality: buyers said they simply don't want to buy their AI capability from the same vendor that already runs their CRM, on principle, regardless of how good the product is.

    Tool used Salesforce Agentforce

Try this today

  • Brief yourself before tomorrow's customer call

    15 min

    You walk in knowing what this buyer cares about and which two objections are coming.

    Copy the prompt

    You are a sales manager preparing me for a first meeting. My notes and the prospect's own public material are below. Give me: three things this buyer likely cares about, each with the line that made you say so; the two objections most likely to come up, each with a one-sentence reply; five questions I should ask; and one thing I still do not know and should find out on the call. Use only the material below. Mark anything you inferred as ASSUMPTION. Keep every answer under 25 words. My notes: [PASTE YOUR NOTES] Their public material, such as a job post, press release or results announcement: [PASTE]

    One check first. The model only knows what you paste in. Treat every assumption as unproven until the call. Nothing goes to the buyer unread.

  • Pressure-test your open deals before the forecast call

    30 min

    You see which deals are stalling before the forecast call, and what to ask each rep.

    Copy the prompt

    You are a sales manager who does not believe the forecast. Below are my open deals, one per line, with stage, amount, close date, date of last contact, and my notes. For each deal give me: a rating of red, amber or green; the one fact in the row that drove that rating; and one question I should ask the rep on Monday. Rank the list worst first and return it as a table. Use only what is in the rows. If a row is too thin to judge, write NOT ENOUGH INFORMATION rather than guessing. Deals: [PASTE YOUR DEAL ROWS]

    One check first. The model does not know your pipeline. It only sees the rows you paste. Check the record yourself before you challenge a rep.

  • Turn one call recording into rep coaching

    20 min

    Your rep gets three specific changes, each tied to a line from their own call.

    Copy the prompt

    You are a sales coach reviewing one recorded call. The transcript is below. Give me: two things the rep did well, each with the line that shows it; three things to change, each with the line that went wrong and a better way to say it; the point where the buyer's interest dropped, with the line it happened on; and one thing for the rep to practise this week. Quote only lines that appear in the transcript. Do not guess what the buyer was thinking. Keep the whole thing under 300 words. Transcript: [PASTE TRANSCRIPT]

    One check first. Everyone on the call must have agreed to the recording. Consent rules differ by country and state. Read the feedback before you send it.

Specialized tools, and what to ask vendors

Specialized tools for this function
ToolWhat it doesSetupBest fit
Gongmid-market to enterpriseRecords your sales calls, writes the notes, and shows you which deals are quietly going cold.WeeksIT sign-offYou want conversation intelligence (software that records and scores sales calls) because deals die without warning.Skip it ifYour buyers refuse recording, or you sell where every person on the call must consent.
Clarimid-market to enterprisePulls deal activity out of your CRM and email to build a forecast your reps did not hand-type.MonthsIT sign-offYour pipeline coverage (open deals divided by your target) looks fine on paper and still misses.Skip it ifNobody logs activity in your CRM. Fix that first or you will forecast noise.
Claylow to mid-marketBuilds target lists and does enrichment (filling in missing company and contact details) from many data sources.WeeksIT sign-offYou have a clear ICP (the customer profile you sell to best) and messy list data.Skip it ifYou do not know who you sell to yet. Better data will not fix a vague market.
Apollofree tier to mid-marketGives you a contact database plus the email and calling tools to work through it.Under a weekYou are a small team that needs contacts and outreach in one place for one price.Skip it ifYou sell mainly into Germany or Japan, where its contact records are thinner.
Regie.aimid-marketWrites outbound messages and runs the sending, with a person reviewing before anything goes out.WeeksIT sign-offYou want more outbound volume and you still want a human reading what leaves.Skip it ifYour deliverability (whether your email reaches the inbox) is already damaged. Repair that first.

Questions to ask before you buy

Gong — 7 questions to ask them

Also used by. Rippling, per Gong's own customer pages

  1. Where are our call recordings and transcripts stored, and in which countries do those servers sit?
  2. Which of our markets require consent from every person on the call, and how do you collect it?
  3. What happens to our recordings and transcripts if we leave, and how long does full deletion take?
  4. Is our call data ever used to train models that your other customers benefit from?
  5. Show me a customer whose win rate did not move, and tell me what was different there.
  6. What is the price at full seat count including managers, and what pushes it up next year?
  7. What happens when a buyer asks us to delete a recording after the call?
Clari — 6 questions to ask them
  1. Will you run your model against our own last eight quarters and show us the accuracy?
  2. How clean does our CRM have to be before your forecast beats the spreadsheet we use now?
  3. Who owns the number when your roll-up and my regional leaders disagree in week ten?
  4. What does quarter one look like while the model is still learning our sales patterns?
  5. What is the all-in first-year cost, and who on my team has to staff this?
  6. Name a customer who left you, and tell me what went wrong there.
Clay — 6 questions to ask them

Also used by. Saviynt, Figma and Intercom, per Clay's own case study index

  1. Which underlying data providers do you use, and what is your lawful basis for European contacts?
  2. What happens to the contact data we enriched through you if we stop paying next year?
  3. How does credit-based pricing behave in a heavy month, and what is our realistic annual spend?
  4. What is your measured accuracy for mobile numbers in our top three markets?
  5. How does a deletion request from a contact flow back through every provider you used?
  6. Who on my team maintains these workflows, and what breaks when that person leaves?
Apollo — 6 questions to ask them
  1. How do you protect our sending domain as volume rises, and what limits do you enforce?
  2. What spam complaint rate do you see across your customers, and how do you measure it?
  3. Where did these contact details come from, and can you show me the consent trail?
  4. What is the per-seat price at our full team size once the first-year discount ends?
  5. How do you remove a prospect who opts out, across every list we hold with you?
  6. How does your data coverage in Germany and Japan compare with the United States?

What everyone is asking

  • NewJul 2026

    Best AI for call recording and coaching?

    Gong is the default once you pass about thirty reps. Check your consent rules first, because recording law narrows the field more than features do.

    What to watch for

    The 2025 and 2026 court fights over notetakers were about consent from everyone on the call.

    Also worth a look. Gong, Clari Copilot, Microsoft 365 Copilot for Sales, Otter.ai

  • NewJul 2026

    Do AI SDRs actually book meetings?

    Some do, with a person reviewing the output. The firms selling these tools are still hiring humans for the SDR role (the rep who books your first meetings).

    What to watch for

    AI-native companies more than doubled their own human SDR headcount last year, while overall SDR job posts fell 21 percent.

    Also worth a look. 11x, Regie.ai, Apollo, human SDRs

  • NewJul 2026

    Best AI for sales forecasting?

    Clari and Gong both forecast from deal activity rather than rep opinion. Neither helps until your CRM records are complete enough to read.

    What to watch for

    A KeyBanc survey in July 2026 found chief information officers unwilling to buy AI through their CRM vendor.

    Also worth a look. Clari, Gong, Salesforce Agentforce

  • NewJul 2026

    Is AI worth it for my sales team, or is it hype?

    Microsoft measured its own sellers and reported higher revenue per seller. monday.com cut demo response time to minutes. Both spent months on it first.

    What to watch for

    Microsoft's own write-up says early trying was widespread but sustained use was not. Adoption is the hard part.

    Also worth a look. Microsoft 365 Copilot for Sales, agents built in-house, waiting another quarter

  • NewJul 2026

    Best tool for building prospect lists?

    Clay if you have someone to build and maintain the workflows. Apollo if you need contacts and email in one place next month.

    What to watch for

    Clay's published customer numbers are its own. Ask for a reference in your market before you believe them.

    Also worth a look. Clay, Apollo

Worth following

  • GTMnow

    The publication and podcast that Sales Hacker became, run by the GTMfund team.

    weekly · newsletter

    Why them

    Practitioner posts on what is working in sales now, rather than vendor marketing.

  • Kyle Poyar — Growth Unhinged

    Former OpenView partner who publishes data on how software companies actually grow.

    weekly · newsletter

    Why them

    He shows the numbers behind AI sales claims, including the ones that undercut them.

  • Pavilion

    A paid private community for revenue leaders, with courses and in-person events.

    daily · community

    Why them

    Peers running comparable teams will tell you what a vendor reference call will not.

  • Jason Lemkin — SaaStr

    Founder of SaaStr, the largest community for software founders and sales leaders.

    daily · publication

    Why them

    He publishes his own AI sales experiments, including the parts that did not work.

  • Gong Labs

    Gong's research arm, which studies patterns across millions of recorded sales calls.

    monthly · research

    Why them

    Call data you cannot get elsewhere. Remember that Gong sells the product it studies.

  • Salesforce Ben

    An independent publication covering Salesforce products and the people who buy them.

    daily · publication

    Why them

    It reports analyst surveys and buyer complaints that Salesforce does not publish itself.