Presenting partner: Viktor
Viktor is an AI employee that lives in Slack and Microsoft Teams. Connects to 3,200+ tools. Does the work, not just answers questions. Hire Viktor for your team.
GTM partners:
Ahrefs Brand Radar - Make AI recommend your brand.
Attio - The agentic CRM built for modern GTM teams.
Clay - Build systems to grow revenue.
Hyperagent - Agents that ship real, powerful work.
Miro - The collaborative workspace where GTM teams and agents converge to build together.
Dear GTM Strategist,
Most CRM advice comes from people who sold you the implementation and then moved on to the next client. They are not there in month four when the scoring model turns out to be wrong.
Kyle Doherty runs RevOps at Attio - an agentic CRM that is used by companies like Parallel, Turbopuffer, Wordsmith, and Granola - and more than 30,000 others.
That means he configures Attio for Attio’s own revenue team. He lives with every decision he makes. If the routing is bad, his AEs tell him on Monday morning.
I asked Kyle to walk me through his setup and GTM stack. What stayed with me was how simple most of it is - and how he adds AI and agents on top of it: reading calls, updating records, explaining why deals were lost.
What you will learn:
How to turn prompts into shared team infrastructure, with the full “Update Deal” prompt to copy
The custom agent that reads every closed-lost deal and writes back why you lost
Why reps run their own Claude skills through MCP instead of one workflow everyone follows
Two ways to store signals in a CRM
Why Attio segments before it scores, and what their entire routing logic amounts to
Kyle’s full GTM stack - including the integration he built himself
Over to Kyle.
1) Shared prompts as team infrastructure
Keeping a CRM current used to be the most cumbersome task. That is exactly where AI now comes in.
Ask Attio is an effective AI assistant, accessible anywhere in Attio. What makes it so useful is a library of prompts anyone on the team can build, save and share directly in the CRM.
So good prompts are not just personal productivity hacks - a shared prompt becomes a team infrastructure.
The plays our AEs lean on most:
Update Deal is a favorite one - they are generally used after calls or at the end of the day and look at the call(s) and email(s) and suggest updates. Most AEs have their own versions tailored to how they want it.
Prep for call - many AEs have their tailored version of these as well, but generally they assemble a single pre-call brief by pulling the account together from everywhere it lives: CRM records and deal history, recent emails and calls, product usage and trial status, support conversations, and fresh web research on the company. It adapts to the moment, giving a light intro brief for a first meeting and a fuller one for a returning call, with a recap of their stated pain and timeline, gaps in qualification, and a proposed agenda. The rep opens one doc instead of eight tabs.
Prep all today’s external meetings - these types are similar but help AEs prep for the day instead of a single call.
Draft specific emails - revenue team members have these from everything from hand raiser follow-up emails to trial-ending emails to handoff emails for SEs and CSMs. It’s not just salespeople that use these but the entire revenue team.
Here’s an example of the “Update Deal” prompt:
**Context**
You are updating a deal record based on a specific sales call. You have access to the call transcript, any notes linked to the meeting, the deal record to update, and associated people and company records.
**Task**
Analyze the call and determine whether any deal record updates should be made, including:
- Deal stage: should it move forward, stay the same, or move backwards. If the deal stage is being updated to Lost, suggest an update for a loss reason attribute.
- Estimated close date: does it need to change based on timeline signals
- Deal value: does scope, seat count, or budget indicate a change
- Next steps: what actions were agreed, and who owns them
- Associated people or stakeholders: any new people records that should be added as contacts or decision makers
- Current solution: what is the prospect currently using as a solution
- Competitors: what are the other solutions the prospect is evaluating alongside our product
- Any other attributes on the deal record that might be relevant to the conversation
**Constraints**
Use the following output structure:
- Be concise and actionable, avoid long paragraphs
- If information is missing, write “Unknown”
- If there are no relevant call recording and deal record in context, ask the user to confirm what the relevant call recording and deal record is before proceeding
- If there are multiple deals in context, ask the user to confirm which deal is the relevant one
Output two parts:
- Recommended updates to the deal record: a record update suggestion for the deal
- Evidence from the call: For each update, include a short one sentence justification on what was said and by whom. Format each attribute update like this: Field: {name}, Current value: {current value if set}, Proposed update: {new value proposed}, Reason: {short reason for the proposal}. Split each attribute evidence into a new subsection, and use bullet lists to separate the details within a single attribute.Many reps create and share their own prompts. We want to empower them to be creative within our sales process and adapt to what leads and customers need, instead of forcing “one prompt to fit everyone”.
Pro tip: In Ask Attio, click “Browse prompts” for a selection of proven and effective prompts you can immediately use.
2) Custom agents
Most teams treat a lost deal as a closed task. The stage updates, the rep moves to the next opportunity, and nobody looks at the deal again.
I have built a custom agent in Attio that looks at closed-lost deals and analyzes the associated emails, calls, and everything that is already in the CRM. The agent then determines why the deal was lost and updates the record: it populates attributes such as closed-lost details, summary, category, and confidence score (how confident the agent is in its findings).
Two things we do with this agent:
Trend analysis. After some time, you can see whether losses are clustering on price, timing, a missing capability, or a competitor, and how that shifts quarter to quarter. This is great feedback for product, marketing, and other teams.
Re-engagement campaigns. A deal lost on timing will lead to a completely different follow-up than a deal lost on a feature that is in the development pipeline. With a category and a confidence score on every record, you can build a list for each and write to it properly.
This is just one example of how we use custom agents to analyze CRM data. Attio agents are incredibly powerful and reliable because they operate off of Attio's Universal Context (the actual CRM data, calls, emails, etc.) and thus don't hallucinate.
Another example is a custom agent that evaluates whether a deal has sales qualification criteria (based on the CHAMP framework).
3) MCP: Same data, different interface
Our AEs and CS folks also work through the MCP connector, running their own Claude skills customized for how they work or which segment they cover.
That is the part I would highlight for anyone building this internally. You do not need one canonical AI workflow that everybody follows. A rep covering marketplaces needs different context than one covering VCs, and a CS person post-sale needs something different again. Same underlying data, different skill on top.
So if reps just want to make a quick update or a check, they don’t need to open CRM: they just tell Claude and it does the lookup or updates Attio for them. (Plus, Attio’s MCP is actually purpose-built, not just a re-dressed API, which results in low token consumption.)
Maja has previously shared this infographic of the best ways to use Attio MCP that can give you more ideas:
4) Capturing and storing the signals
Everyone can produce a list of signals worth tracking.
The hard thing is landing them somewhere that produces an action. Only a signal you can act on is intelligence.
If a signal arrives in a place nobody looks, it does not matter how good the signal was. Most teams drown in signal volume and then conclude signals do not work.
Here are two ways that can help you tackle signals in Attio - depending on the problem you are solving.
Store signals as attributes on the object. Fine for simple, low-volume cases. It breaks the moment you have something like revealed web visitors, where one company generates a stream of events. You want to see the journey. An attribute gives you a single field that keeps getting overwritten.
Build a custom signal object. More flexible, but it gets messy fast. Every signal type carries different data, so you end up with an object holding 30 attributes where only two of them apply to any given record.
If you go the attribute route, the fastest way to get the data in is a ready-made app. The Clay integration below is a good example of what that looks like in practice.
5) Segmentation before lead scoring
People get lost in building scoring models.
I understand the pull. A weighted model feels like maturity. But scoring is a prioritization tool, nothing more, and the model you build on day one runs entirely on assumptions you have not tested.
We started with the simplest possible version: sales qualified, yes or no.
That was tied to a goal. We optimize for learning speed. A strict score early would have filtered out companies we still needed to see, and we would never have found out they were good. We only layered in tiering once we had attributes we could defend, because we had watched them play out.
What happens after the yes or no
Not sales qualified: they go into normal PLG sequences in Customer.io. No rep touches them.
Sales qualified: they enter a sales assist motion. Routing is location first. Our UK team covers Europe, our US team covers the Americas, South Pacific, and East Asia. Then some light segmentation decides whether it goes to senior AEs or the full pool.
That is the entire routing logic. No vertical ownership, no round robin with exceptions.
The reason is that we want our AEs to be athletes. Give them good data and let them decide where their hour goes. We work with them afterward to learn which attributes actually mattered, and that loop improves the system faster than any routing rule would.
Why segmentation matters more than the score
We sell a horizontal product, which makes a single score cutting across every customer type close to meaningless.
The clearest example I can give: a 20-person VC firm is a very attractive lead for us. A 20-person tech startup is an okay lead. A 150-person tech company is attractive again.
Two of those have identical headcount and completely different value. VCs use the product differently than AI startups, who use it differently than marketplaces, who use it differently than fintechs. Different behavior, different seat counts, different reasons to expand.
You can run simple scores that cut across segments. Once you go deeper than that, you have to score within them.
Agents give us better attributes than enrichment alone
For certain company and people attributes, agents get us better data than waterfall enrichment does. This is not either/or. We still use Clay. But for some fields, an agent wins, because it pulls real-time data and because you can shape the output to exactly what you need.
Industry is a typical example. There are two usual ways to get industry:
Waterfall enrichment pulls from several providers who each define industry differently, so it’s very hard to normalize.
Self-reported LinkedIn industry often isn’t how you’d want to classify a company, so it's not great either. For example, a SaaS company selling into healthcare will list itself as being in the healthcare industry, but for our purposes they’re a vertical SaaS company in the healthcare vertical.
Instead, we chain two agents. The first researches the company against the things we actually care about. The second classifies it strictly against our own taxonomy, with no freedom to invent a category.
The output is a custom industry value that is genuinely useful for segmentation and prioritization, because it was built against our definitions rather than a data provider’s.
Code blocks instead of nested logic
Attio workflows had long chains of if and switch statements. They worked, but nobody could read them. To change one branch, you first had to hold the full tree in your head.
We now replace those chains with one code block. The workflow becomes short, and a new person can follow it.
There is a second benefit to that: You can write unit tests for a code block in an AI coding tool like Claude Code. Run many tests against the logic before you put it in the workflow. Then do the manual test in Attio at the end. This is faster, and the quality is better.
6) What we connect Attio to
Most of our stack connects through the App store. There are a lot of integrations there, and for the majority of tools, that is the whole setup: find the app, connect it, done.
When there is no app for something, you are not stuck. The Attio SDK makes it straightforward to build your own, and that is a real difference from most CRMs, where a missing integration means a Zapier chain or a ticket to engineering.
That is how Lemlist ended up in our instance.
When I joined Attio in January, we needed to get Lemlist activities into Attio, but there wasn’t a native integration. So I vibe-coded an Attio SDK app and installed it in our instance to allow us to listen for Lemlist activities such as email sent, email reply, LinkedIn invitation sent, etc., and then associate these activities with people and their related companies and deals.
Here is what our stack looks like today:
Signals in
Segment: product and website events land on the company record, so a rep sees usage before the call
Market signals: funding news, job postings, and social engagement, so we know when something changes at an account
Luma: event signups and attendance, which we track on event lists
Attio call recorder: calls and transcripts on the deal, which the closed-lost agent reads
The system of record
Objects: people, companies, deals, workspaces, users, and custom objects. We model the business the way it actually works instead of forcing it into someone else’s shape, which is why we get a real 360 view of a lead or a customer
AI layer: Ask Attio, custom agents, and workflows running across all of it
Work out
Attio sequencer: inbound follow-up
Lemlist: outbound sequences, with activity synced back through the SDK app I built
Slack: alerts and handoffs to the rep who owns the account
Claude via MCP: each rep runs their own skills against the same data
Data layer
Clay: enrichment and research on new companies and people
Fivetran: pulls source systems into BigQuery
BigQuery: our warehouse
Polytomic: pushes computed attributes from the warehouse back onto Attio records
Omni: AI analytics on top of the warehouse
Thanks, Kyle, for these hands-on insights!
If you are looking for an AI-native CRM, one that is built for how work gets done in 2026, I seriously recommend that you check out Attio.
It’s easy to set up, but if you have a development background (or your colleagues do), you will also be amazed by how easy it is to build on top of it and customize it. You can find great use cases on Attio’s Engineering blog.
And if you’re eager to learn more, explore their latest features or dive into the background of their recent out-of-home campaign.
Until next week!
✅ Need ready-to-use GTM assets and AI prompts? Get the 100-Step GTM Checklist with proven website templates, sales decks, landing pages, outbound sequences, LinkedIn post frameworks, email sequences, and 20+ workshops you can immediately run with your team.
📘 New to GTM? Learn fundamentals. Get my best-selling GTM Strategist book that helped 9,500+ companies to go to market with confidence - frameworks and online course included.
📈 My latest course: AI-Powered LinkedIn Growth System teaches the exact system I use to generate 7M+ impressions a year and 70% of my B2B pipeline.
🏅 Are you in charge of GTM and responsible for leading others? Grab the GTM Masterclass (6 hours of training, end-to-end GTM explained on examples, guided workshops) to get your team up and running in no time.
🤝 Want to work together? ⏩ Check out the options and let me know how we can join forces.