Fewer than most lists suggest. AI and automation do four jobs for a revenue team: keep a record of customers and deals, capture what happens on calls and in email, move data between tools, and read that data to summarize, extract or flag. Which product fills each job matters less than whether the tools are connected and someone owns them.
The tools named below are common examples. New ones appear every few months, so treat this as a list of slots to fill.
Where should customer and deal data live?
Usually in a CRM such as HubSpot or Salesforce. AI tools read from it and write to it, so the quality of what’s inside sets the ceiling for everything else. I cover that in how to fix CRM data quality with AI.
Some teams need less than a full CRM. A small team with a short sales cycle can run on a simpler database or a shared table, as long as one place holds the truth and the automation writes to it. Buying a CRM first and working out what to track afterward is the expensive order.
How do you capture what customers say?
Record and transcribe calls. Gong, Fathom, and the recorders built into Zoom and HubSpot all produce transcripts. The transcript is the raw material: the qualification details, objections and next steps are already in it, so reps shouldn’t have to retype them. Automating MEDDIC qualification shows what that looks like.
Choose by what you’ll do with the transcript. If you want analysis inside the tool, the larger platforms do more. If you want transcripts to feed your own workflows, any recorder with a usable export or API will do.
Do you need an enrichment tool?
Only if you score or route leads on details you don’t collect yourself, such as company size, industry or tech stack. Clay and Apollo are common choices, and they work differently; Clay vs Apollo covers which suits which team. If most of your leads come from referrals and conversations, this can wait.
What connects the tools?
An automation tool such as n8n, Make or Zapier. It runs the steps: a lead arrives, gets enriched and routed, and the CRM is updated; a call ends, the transcript goes to a model, and fields are drafted for the rep to check.
Most of the work happens in this layer, and it’s where things break when someone renames a field or adds a deal stage. In the automations I look after, the first thing to break is usually a field mapping, after someone tidies up the CRM and renames a property a workflow depends on. It needs an owner. Setting up a first AI sales agent walks through one build.
Which AI model should do the reading?
Any of the major models can read a transcript, summarize a deal or draft an email. Claude and ChatGPT are the usual choices, and Claude vs ChatGPT for sales workflows suggests how to test them on your own material. The model is the easiest part to swap later, so don’t let the choice hold up the rest. When I’ve changed the model in a working automation, it took a new API key and some prompt adjustments, and the connections and data stayed as they were.
What can usually wait?
- Tools that write and send outbound email for you. They change what your buyers see. Get the operational basics running first, then judge these on their own.
- A separate forecasting tool. Clean CRM data and a regular summary of what moved usually tell leadership more than a new forecast layered on the same data.
- Anything bought to fix a problem nobody has written down yet.
Where should you start?
Write down the one manual task that costs the team the most time each week, then work out which of the slots above it touches. That tells you which tool to add or fix first. When I look at a team’s stack for the first time, I usually find a call recorder and a CRM that both work and have never been connected to each other.