ABM's old bottleneck
Classic ABM was labour-intensive by design. A strategist picked a short list of target accounts, a researcher dug through each company's site, news, and hiring pages, and a writer crafted a custom angle per account. The output was genuinely relevant — and impossible to scale past a few dozen accounts without a small army.
So most teams cheated. They kept the ABM language but dropped the research, and "personalisation" collapsed into Hi {{first_name}}, I saw {{company}} is in {{industry}}. Recipients can spot a merge field from across the room, and response rates show it.
What AI changes is the economics of the research-and-drafting step. A model can read an account's public footprint, summarise what matters, and draft a grounded angle in seconds. That makes the labour-intensive version of ABM viable at hundreds or thousands of accounts — provided you keep a human reviewing the output and feed the model real inputs, not guesses.
Building the target account list
Targeting rests on two questions. Fit: does this account look like an account you can serve and win? Intent: is there a signal they are in-market right now? Fit comes from firmographics and technographics; intent comes from behaviour — content consumption, hiring, funding, product launches, search activity.
AI helps on both. It can enrich and de-duplicate account records, infer fit from messy public data, and cluster accounts into tiers so your effort matches the prize. The goal is a ranked list, not a flat one: Tier 1 accounts earn deep human-reviewed personalisation, Tier 3 accounts get lighter, signal-triggered touches.
Treat the criteria above as an example, not a prescription. Your fit definition and signal sources are yours to set — AI just applies them consistently across a list no human would have time to score by hand.
Personalisation at scale
The useful unit of personalisation is the angle, not the salutation. From an account's public signals — a recent funding round, a new VP of Engineering, a product announcement, a stated initiative in a job posting — a model can generate a specific, defensible reason this account would care about what you do, and draft messaging around it.
Done well, the recipient reads something that could only have been written for them. Done carelessly, it tips into creepy. The line is roughly this:
- Personalised: references information the account has published or would expect to be public — funding, roles, launches, posted initiatives.
- Creepy: references behaviour they didn't knowingly share, or implies a level of surveillance no peer would.
- Grounded: every claim in the message traces back to a real, citable signal — no invented detail to make the angle land.
The guardrail here is sourcing. If the model can't point to where a personalised detail came from, it doesn't go in the message. That keeps the output relevant and keeps you out of the uncanny valley — and it's why a human stays in the loop on Tier 1 sends.
Aligning demand gen with sales
ABM breaks at the handoff more often than at the targeting. Marketing calls an account "engaged," sales sees a cold lead, and the feedback loop that would settle it never closes. Scaling personalisation only makes this worse if the definitions underneath are fuzzy.
The fix is unglamorous and necessary: shared definitions, explicit routing, and a feedback loop.
- Shared definitions: one agreed meaning for a qualified account, an engaged account, and a handoff-ready account — written down, not assumed.
- Routing: deterministic rules for which signal sends which account to which rep, so nothing sits in a queue nobody owns.
- Feedback loop: sales dispositions flow back to marketing so tiering and targeting improve on the next cycle instead of repeating the same misses.
None of this is AI-specific — it's ownership. Decide who owns each definition, each routing rule, and each model before you automate anything. We treat that as a governance question; see our notes on governance and ownership for how to structure it.
For B2B specifically
B2B is where this approach earns its keep. Deals involve buying committees, long cycles, and a handful of high-value accounts where genuine relevance moves the needle — exactly the conditions ABM was built for, and exactly where signal-grounded personalisation beats volume. The broader playbook lives in our B2B AI marketing guide.
One caveat before you build any of this: automation amplifies whatever foundation it sits on. If your account data is dirty, your fit definition is vague, or sales and marketing disagree on what "qualified" means, AI will scale those problems faster — not fix them. Don't automate a broken foundation.
That's the gate. Before you stand up AI-driven ABM, check whether your data, definitions, and processes are ready to be automated. Our free Readiness Score walks 13 questions across democratization, instrumentation, standardization, governance, and tooling, and returns a 0–100 read on where you actually stand.
Event Follow-Up Sprint
After a conference or webinar you have a list of accounts with fresh intent — we build the automation that researches each one, drafts grounded, account-specific follow-up, and routes it to the right rep before the signal goes cold. You own the result.
See the Event Follow-Up Sprint →