Guide / GTM Engineering

The GTM engineering stack we actually use

Every other week somebody asks for the list. So here it is, without the affiliate links and without the politics. This is the stack our team runs in 2026, broken down by layer, with what each tool does, what breaks when it is missing, and the common ways we have seen teams misuse it.

One principle before the list

A stack is not a strategy. Two teams can buy the exact same tools and one will produce pipeline while the other produces noise. The reason is almost always the same: the second team picked tools before they picked signals and workflows. Pick the workflow first, then ask which tool is the cheapest way to run it.

With that out of the way, here is what we use.

Layer 1: data and enrichment

This layer turns "we want to talk to mid-market data platforms in DACH" into a clean list of accounts and contacts with enough context to write something non-generic.

• usegrit.io. Our own enrichment and workflow platform. We built it because we wanted tighter control over agents, signals and how data moves between layers. It is what most of our production workflows run on day to day.
• Clay. The other serious option in this category, and one we still use. Great for TAM building, waterfall enrichment, scoring and message drafting with embedded AI columns. If you are not using our platform, Clay is what we would point you at.
• Apollo. A baseline source of contacts and firmographic data. We rarely use Apollo sequences for sending, but as a data source plugged into the enrichment layer it earns its seat.
• LinkedIn Sales Navigator. Still the highest-signal source for filtering people, especially around recent job changes, posting activity and company growth filters. Scraping is fragile, so we keep this manual or run it through tools that respect the platform.
• Specialist providers. For specific signals we layer in narrower providers: BuiltWith and Wappalyzer for tech stack, public hiring data for hiring signals, news APIs for funding and product launches.

What breaks without this layer: every message you write is generic, because the agent has nothing to work with. We have seen teams skip enrichment to save credits on the enrichment surface and then wonder why reply rates flatlined. The credits were the cheapest line item they had.

Common mistake: enriching everything to the maximum on day one. You do not need a 40-column waterfall for an MVP. Start with three columns that matter for your message, ship the workflow, then add columns when a real workflow demands them.

Layer 2: sending infrastructure

This layer puts the message in front of the human. Email and LinkedIn are still the two channels worth automating against in 2026. Calls are valuable, but they are a human channel, not an automated one.

• Instantly for email. We send the majority of cold email through Instantly. Domain management, warmup, sending pools, inbox rotation, basic deliverability monitoring. It is unglamorous infrastructure, which is exactly what you want from this layer.
• HeyReach for LinkedIn. For LinkedIn outreach at scale, HeyReach has been the most stable option we have used. Multi-account, sane rate limits, replies in one inbox. It respects the platform enough not to get accounts banned every other week, which is a lower bar than it sounds.
• Domain and inbox setup. Separate sending domains from your main brand domain, SPF, DKIM, DMARC, warmup. Boring. Non-negotiable. If you skip this you are not running a system, you are running a deliverability experiment.

What breaks without this layer: messages get sent, nothing arrives. We have inherited campaigns where the open rate looked fine and the reply rate was zero because the inboxes were silently in the spam folder for half the planet. The fix is unglamorous: split the sending pool, warm up properly, monitor placement.

Common mistake: sending from the main company domain. One bad week tanks your transactional email too. Use dedicated domains.

Layer 3: signal monitoring

Signals are what turn a static list into a triggered workflow. Without signals, you are sending the same message to everyone in your TAM on the same day, which is the old way and it stopped working.

• Hiring signals. Job posting feeds, scraped from the public web and filtered for roles that indicate a buying window for our client. A SaaS that sells QA tooling cares when a target hires their third QA engineer.
• Funding and corporate events. Pulled from a mix of public sources and specialist databases. A Series A on Tuesday is a different message on Wednesday.
• Social activity. LinkedIn posts, comments, podcast appearances. We use these to find Champions and Power Users inside accounts, not just Decision Makers.
• Product events, for SaaS clients. Signups, trial starts, integrations connected, usage milestones. These are the highest-signal triggers in the whole system. If you are SaaS and you are not wiring these in, that is the first thing to fix.

What breaks without this layer: you run volume against a market that has no reason to care today. You can still book some meetings, but the cost per meeting is multiples higher than it has to be.

Layer 4: AI models for research and personalization

We use AI models in three places in the stack: research, drafting, and reply classification. We do not have a religious commitment to a single provider. We pick whichever model is currently cheapest for the quality we need, and we revisit that every couple of months.

• Research. Long-context models that can read a company website, a few LinkedIn posts and a recent news article, then produce a short brief in the voice we want. We feed the brief into the drafting step, not directly to a human.
• Drafting. A separate prompt that takes the research brief plus a tight style guide and writes a first-touch message. The style guide is more important than the model. Most "AI personalization" reads like AI personalization because nobody wrote a style guide.
• Classification. Cheaper, faster models triage replies into buckets: positive, objection, out-of-office, wrong person, unsubscribe. The buckets route the conversation. Positive and objection go to a human immediately.

Common mistake: using one model for everything, with one mega-prompt. You end up paying frontier prices for triage and getting research-quality drafts from a model tuned for chat. Split the steps.

Layer 5: meeting intelligence

Once a meeting happens, the system needs the transcript. We use Fireflies for recording, transcription and summary across the team. The transcript is not for the AE who took the call, it is for the system: which objection came up, what stack the prospect mentioned, what next step was promised.

We pipe relevant fields back into the CRM and into the next campaign. A "we evaluated Competitor X" line in a call last quarter is a signal for a workflow this quarter when Competitor X has a bad week. Without meeting intel, that knowledge dies on the call.

Layer 6: CRM

CRM is the system of record. HubSpot for most of our SaaS and software house clients, Pipedrive for smaller setups, Salesforce when the client already has it. The choice matters less than the discipline: every workflow writes back to the CRM, every reply gets logged, every account has an owner.

What breaks without this layer: parallel sources of truth. The sequence tool says one thing, the calendar says another, the AE remembers a third. By the second quarter, nobody trusts the numbers.

On enrichment platforms and tool swaps

We get asked about enrichment platforms weekly. The honest answer is that this category has two serious options in 2026: our own platform usegrit.io, which we built because we wanted tighter control over agents and signal handling, and Clay, which is the strongest general-purpose option on the market and one we still use on certain workflows. Most other tools in this category are fine for narrow use cases but do not cover the breadth that a real GTM motion needs. We re-evaluate every six months.

For sending, Instantly and HeyReach are the defaults but they are not religion. Smartlead is a reasonable email alternative. La Growth Machine works for some LinkedIn flows. What matters is that you control the domains, the warmup and the rotation, not which logo is on the dashboard.

What this stack is not

This is not a magic combination. We have inherited every one of these tools from teams that produced nothing with them. The stack is necessary, not sufficient. The system that runs on top of it is the actual product. If you want the system side of the story, read how to build a GTM system. If you want to see the kitchen in production, the case studies are the most honest view we publish.