Skip to content
tl;dr on AI Pre-call Account Research
- AI pre-call account research means asking one question ten minutes before a call instead of opening your CRM, your inbox, and your call library separately.
- There are four questions that get you a brief: who’s attending and who was added late, what was actually said across every past call, where the CRM and the email thread disagree, and what the assistant couldn’t find.
- Research the account and the buying group, not the person. Gartner found individual-level content has a 59% negative impact on group consensus, against 20% positive for content aimed at the organization.
- The tl;dv MCP has five read-only tools. Nothing it does can change your library.
- An empty source and a genuine “nothing there” look identical in the output. Always ask what it couldn’t find.
AI pre-call account research means that a few questions pointed at a connected AI before a call can save time in trawling through a mess of notes, CRM, calls, and documentation beforehand.
This is now possible thanks to MCPs, whereas before you’d need to check the CRM to see where the deal is in the pipeline, read through emails to see what the latest situation is, you can just type in a single prompt and voila, you have everything you need.
Setting up takes no time at all, but working out what to ask is a bit of a skill to master. I worked in sales myself for many years, and still do to a degree. I have also spent a lot of time working with AI, MCPs, and thinking about how this can help the average sales rep. Below you’ll find a solid workflow that I use myself before jumping on a call with customers and clients, queries, prompts, and how I use that information to make sure each call I make is useful and gets the job done for both parties.
What You Need Before You Start
Before you start you’re going to need a few things. For this situation, you’ll need:
- A tl;dv account, on Pro or above, with the calls you need in your library. If a colleague had a call, but didn’t share it with you, you’ll need access to that. This is easy to do with just a few clicks.
- A paid Claude or ChatGPT account. Both work, and tl;dv is listed on the connector directory for both.
- The connection itself. You’ll need to have actually set this up, which can be done with just a few clicks from the connector directory. You log into tl;dv to authorize it. There’s no custom connector needed, no API key required, no config file, nothing else to install.
- A spot check. Things can go wrong sometimes, so before you put all the pressure on this call, do a quick sense check with the connected MCP. Ask a question that you know the answer to, “What is the name of the person from Acme Corp?” or “What did I say to [Client] on [Date]?”. That way you can make sure everything is pulling through as needed. You can also ask for the verbatim transcript.
That’s all it takes. Once that’s set up, then you’re good to go.
What a Pre-Call Brief Actually Needs
A really good pre-call brief answers four specific questions.
Who is going to be on the call? Your calendar. The attendees, their roles, the meeting type and timing, and whether anybody has been added or dropped since the invite first went out.
What have we already said to these people? Your call library. Previous calls, the promises made on both sides, and the objections raised. These get overlooked in favor of summaries, but searching the transcripts themselves uncovers the small details that matter when an objection comes back around.
Where does the deal actually stand? The CRM. Deal stage, amount, owner, account history, and when activity was last logged. This is the official internal view of the account.
What has been said in writing? Email. The latest thread, the questions asked, and anything that has changed since. This is usually where the truer and messier version of the account lives.
What needs to come from the call today? This is the goal of the meeting, when you have everything laid out in front of you what can be achieved and what decisions can be pushed through easily with the conversation.
Everything else is context around the edges. Funding rounds, hiring signals, anything that builds a picture of where the company is right now. Useful, and secondary to the four above.
Why Ten Minutes of Pre-Call Research Pays
There are solid, documented reasons why pre-call research is worth the time it takes. Sellers spend around 40% of their time selling, according to the Salesforce State of Sales. The remaining 60% was reported to go toward things like data entry, admin, and research.
That means more than half a rep’s time is taken up by this kind of manual work. That’s time that could be spent on calls with potential customers, getting people across the line, and generally generating more revenue.
So the case for automating the research part is strong. A brief put together from your own transcripts, your own CRM record and your own inbox is built on what was actually said. Very little guesswork, and no reliance on you remembering every detail from a call six weeks ago.
There’s another element to this. Gartner’s 2025 survey of B2B buying groups found that content personalized to individual buyers is not as powerful as content aimed at the buyers as a group. Tailor your pitch to the one person on the call and they’ll likely be enthusiastic, but that enthusiasm won’t necessarily get the deal over the line. So point the research at the account rather than the person. What has this company said about the problem? Where did people on the call disagree with each other? What does the business get out of fixing it? That’s material a champion can repeat without you there.
The Four-Step Pre-Call Research Workflow
To get started on using AI to create solid pre-call account research just takes a few steps. I’ll use Claude here, but you can also use ChatGPT to do this.
Step One: Start From the Calendar, Not the Account
Before you do anything, ask Claude what is in your calendar tomorrow and who will be attending. If you have a busy inbox, then there’s a chance that other people may have been added to the call, or even declined the meeting invite. I’ve had this happen where I’ve gone into a call expecting to speak with my contact, and found that the VP of Marketing was attending and I’d missed the notification.
Having the attendee list also lets you get specific about what you need to pull from the data. If someone is coming in from finance, you can have the monetary figures ready to go. If there’s someone technical, you can prep the specifications beforehand so you can answer their questions. It might also give clues that the deal is progressing behind the scenes.
Step Two: Ask Your Call History a Real Question
Having Claude connected to your tl;dv library lets you cross-examine every transcript available to you. The trick is making sure you’re specific rather than broad, so you don’t just get a top-level summary.
The question you want brings together strands across multiple calls and stages. Some examples:
- Across every call with this account, what have they said about pricing, and has their position moved?
- Which objections have come up more than once, and did we ever actually answer them?
- What did we commit to, and did we do it?
- Who has been on every call, and who dropped off after the first one?
To look these up manually would take close to an hour. Using the connector, seconds. One good habit to get into is asking for specific quotes: “Tell me what they said about budget, quoting the transcript with dates.” That keeps the answer grounded in what was said rather than a generalization smoothed by the model.
Step Three: Reconcile CRM Against Email
Next, pull the CRM record and most recent email thread in the same question, then ask if there are any differences or discrepancies between them. The CRM might tell you the deal is in negotiation, closing this month, but the email says their procurement lead is out of the office until September and nobody has responded in a week. Both are correct, to a degree, except the email has the most up-to-date and defensible information.
One thing you can ask directly: “Is the CRM stage still accurate when compared with the email thread, and what has changed that isn’t reflected in the CRM?”
Step Four: Ask for the Brief, Then Ask for the Gaps
Now you can ask for all that information to be pulled together in a format that works for you, and for the call you are about to have. I personally like to have:
- Who is on the call
- What’s changed since the last call
- Where the deal is currently
- Where the CRM looks stale
- The open commitments on both sides
- The likely objections based on the history
- A good question to ask to move this deal forward
- Where people on their side disagreed with each other
Then once I have all that, I ask “What is missing?”
That last question earns its place when multiple stakeholders are involved. Closing gaps, filling in missed details, and spotting blockers before the call is what lets you walk in prepared rather than hopeful.
tl;dv’s Five Tools
tl;dv’s MCP has five tools available to use once connected up. They are:
- Search Meetings — finds calls by date range, specific text, internal or external, or whether you were attending. This is a solid place to start your brief research.
- Get Meeting Metadata — returns the title, date, duration and all the other details of a meeting from its ID.
- Get Meeting Notes — returns the AI-generated and manual notes attached to a call. A good way to check something quickly before you open the full transcript.
- Get Meeting Transcript — returns the entire transcript in markdown. This is the one you’ll reach for most, and the one to use when you’re after specific quotes. It’s also useful when you’re sending documentation: ask the AI to check what you’re about to send against what was promised on the call, and flag any gaps.
- Get User Profile — returns your own profile, which confirms it’s searching your library.
The key thing with all five is that they are read-only. The AI can recall and share whatever is in your library, and it cannot write to it or change anything.
The Claude Projects Layer
The best place to get this set up within your AI, if you are using Claude, is Claude Projects. You can create a dedicated space to query everything you need to, based on the context that’s required.
So for example, you could create a project for a particular role, or even for the company project itself. You can then add context into the Project, such as your qualification framework, ICP profile, objection library, technical data, even the format you want the brief to be in, and the results come out shaped to what you need.
One limitation decides whether this compounds or resets. Anthropic states it plainly: “Context is not shared across chats within a project unless the information is added into the project knowledge base.” So if you research an account on Monday and open a fresh chat before Thursday’s follow-up, you start from nothing and pay for the same reasoning twice.
The fix takes fifteen seconds. After the brief, ask for a short account summary written for future reference, and save it into project knowledge. Next time the assistant starts warm, and the live sources only fill in what’s changed.
Projects are available on every Claude plan, with free accounts capped at five. Check out our piece on Claude Projects for a deeper look at what you can do with the workspace.
The Pre-Call Research Prompt
I have a call with [account] at [time] today. Work through this in order.
1. Pull the calendar event and list who's attending, with titles. Flag anyone added after the invite first went out, and anyone who has declined.
2. Search my tl;dv library for every past call with this account. Tell me what was actually said about pricing, timelines and any competitor, quoting the transcript with dates. Not a summary of the summaries.
3. Pull the CRM record: stage, amount, close date, owner, last logged activity, open tasks.
4. Search my email for the most recent thread with this account. Compare it against the CRM record and tell me what has changed that the CRM doesn't reflect.
5. Give me a brief in this format: who's on the call and what's changed since last time; where the deal stands and where the CRM looks stale; open commitments on both sides with who owes what; the two objections most likely to come up based on our history; where people on their side disagreed with each other; one question worth asking that we haven't asked before.
6. Then tell me what you couldn't find, and which parts of the above rest on thin or missing data.
Keep it under 400 words. Don't fill gaps with plausible assumptions. If you didn't find something, say so.What to Check When the Brief Comes Back Wrong
No prompt is perfect, no email chain or CRM captures everything, and the above will get you what you need most of the time. When it doesn’t, it’s almost always one of these five.
1) The calls aren’t in your library
Easily the most common issue when the answer isn’t correct. If a colleague has taken a meeting on their own account and not shared it, it’s not yours to search, and the AI has no way of accessing it or even knowing it exists. If this is the case, check your tl;dv library directly and make sure you have the call.
2) An empty result and a true negative look the same
When a connected source returns nothing, for any reason, the AI reports that it found nothing on the topic. That sentence is true from where it’s standing and completely wrong from where you are, and nothing in the output tells you which one you’re looking at. It’s why the last line of the prompt asks what it couldn’t find.
3) The CRM is stale, and the AI doesn’t know
Despite tl;dv being able to integrate with CRM systems, many people still forget to update the record. They don’t add notes, they don’t log an entry when something happens, and the important context stays in somebody’s inbox rather than anywhere you can reach it.
4) Email search only has what you can see
Great from a security point of view, limiting from yours. If you’ve been left off an important thread, or the prospect messaged your colleague instead, that data isn’t available to you and nothing will tell you it’s missing.
5) You requested a summary instead of asking a question
When the output comes back looking like a set of notes from a call rather than definitive answers, the wording of your prompt was probably off. Go back to step two and rework it to ask for something specific.
It’s unlikely you’ll ever get a full error message. Spot checks are what confirm, and fact-checking against what you already know is what stops it failing at the wrong moment.
What This Workflow Does Not Do
This won’t magically turn every call into a sale. All of it shortens the preparation time, and it’s still up to the rep to use solid sales techniques to get the deal over the line.
You still need to know how to open the conversation, read the room and spot when someone has gone quiet, and tell the difference between an objection and a buying signal that just needs a little more information.
AI can help you fill in the CRM and draft the follow-up email. You still need to be a salesperson.
What the workflow above does do is shrink that 60% of non-selling time, so you can focus on the craft of selling, listen properly to your customer, and have the energy and space to get better at it.
How to Start Without Overbuilding It
Getting started is simple, and you don’t need to do it all at once. An AI assistant with the tl;dv MCP connector set up gets you most of the way there. Your CRM and email connections can wait until that’s earning its place and giving you clear answers to the questions that matter most before a call.
Test it for yourself today by getting tl;dv connected to your AI meeting assistant, and running it before your next call.
FAQs About AI Pre-Call Account Research
Can Claude or ChatGPT connect to my CRM, email and meeting recordings?
Yes, through connectors. In Claude, you add them from the connector directory and authorize with your account, with no API key and nothing to install. The tl;dv connector is listed there and is read-only. ChatGPT works the same way through Apps. Both are available on tl;dv Pro and above.
What should I ask AI before a sales call?
Compare the CRM record against the most recent email thread and ask what has changed that the CRM doesn’t reflect. The CRM holds what somebody typed at the time. Email holds what has happened since. Also cross-reference against the most recent calls.
Is it safe to connect AI to my meeting recordings and CRM?
The tl;dv connector is read-only. Its five tools, search_meetings, get_meeting_metadata, get_meeting_notes, get_meeting_transcript and get_user_profile, all fetch and none write. It cannot create, edit or delete anything. The worst outcome from a badly worded query is an unhelpful answer rather than a changed record.
Can AI access meetings or CRM records I don't have permission to see?
No. It inherits your existing permissions, so it reaches what you could already open yourself. That cuts both ways: if the email thread you need is in a colleague’s inbox, or a call was recorded on their account and never shared, it won’t appear in your brief and nothing will tell you it’s missing.
How accurate are AI-generated sales call summaries, and can it invent things?
A brief is accurate to the extent its sources are. Ask for quotes with dates rather than characterizations, and you can verify every claim against the transcript. If you give a vague request without asking for it to be grounded in the material then AI may sometimes try and fill in what it does not know.
Can AI compare what the customer said against what's in the CRM?
Yes, and it’s an incredibly useful query in the workflow. Pull the CRM record and the most recent email thread in the same question, then ask for the discrepancies. The CRM says the deal is closing this month. The thread says nobody has replied in two weeks. Both are accurate.
