A few years ago, researching one target company properly took most of an afternoon. You read the website, the press page, the last few job postings, the owner's LinkedIn activity, and maybe a local business journal article, and you came away with two or three facts worth building a conversation on. For a list of thirty companies, that was a month of evenings, which is why most small B2B firms never really did account-based marketing. They had a list. They did not have the hours.
That constraint is mostly gone. AI tools will now suggest which companies belong on the list, summarize each one in a few minutes, flag which ones are paying attention to your ads, find the people inside them, and draft the first message. Every step that used to eat the afternoon now takes a coffee break.
What did not get faster is knowing whether any of it is true, current, or meaningful. That part is still a person's job, and it is the part most firms are quietly skipping. The result is a new kind of mistake: a confident, well-written email about a company's "recent expansion" that happened two years ago, sent because a sudden burst of interest turned out to be one intern reading your guide for a training assignment.
This article walks through the five steps of account-based marketing in order. For each one: what AI now does well, and the specific check you still have to make yourself.
AI Will Build Your Target List in Seconds. Start It From Your Best Customers Anyway.
The first job in account-based marketing is choosing the companies, and this is where AI tools are most eager to help. LinkedIn Sales Navigator will recommend accounts based on your existing customers: on the Core plan ($119.99 a month at LinkedIn's current list price) you upload them as a spreadsheet, and on Advanced ($159.99 a month) you can also pull them straight from your CRM. On the advertising side, LinkedIn has an AI audience it calls buyer groups, which it describes as reaching "the group of professionals that collectively make product purchasing decisions for a company," and which you can combine with your own uploaded company list.
Both are genuinely useful. Both also inherit a problem: the AI is matching on what it can see, which is industry, size, location and job titles. It cannot see the things that actually made your best customers good customers. Maybe your three most profitable accounts all run a second shift, or all came to you after failing an inspection, or all have an owner who still signs every check over $5,000. None of that is in a firmographic database, so none of it shapes the suggestions.
So use the suggestions as a draft, not a decision. The list still starts from the question we laid out in You Don't Need More Leads. You Need the Right 30 Companies.: which customers would you clone if you could? Write down what those customers have in common, including the messy, non-database reasons. Then run each AI-suggested company through that filter. Keep the ones that pass. You will usually cut more than half, and that is the system working.
Two settings deserve a look before any ads run against the list:
- Audience expansion. LinkedIn turns it on automatically wherever an ad set offers it, and its job is to show your ads to people "similar" to your audience. On a campaign built around thirty named companies, similar is exactly what you do not want. If the option appears on your ad set, switch it off.
- The size floor. LinkedIn will not run an ad set below 300 members. Thirty companies narrowed to two job titles can drop under that line fast, especially if the companies are small. If it does, widen the roles before you widen the list. The point of the list is the companies.
AI Can Summarize a Company in Three Minutes. It Can't Tell You the Facts Are a Year Old.
The research step is where the time savings are biggest, and where the risk hides best. Paste a company's website, recent news and reviews into ChatGPT, Claude, Gemini or Perplexity and ask what changed in the last year and what seems to be going wrong. You get a tidy summary in minutes. Inside Sales Navigator, a feature called Account IQ does something similar automatically, building an account summary from public sources.
Here is what the summary does not tell you: how old each fact is, and whether it is a fact at all.
The research on this is not encouraging. When the Columbia Journalism Review's Tow Center tested eight AI search tools on 1,600 queries, asking each to identify the source of a news excerpt, the tools collectively gave incorrect answers to more than 60 percent of them. The paid versions were not safer. The researchers found they were more likely to give "definitive, but wrong" answers instead of admitting they did not know. A separate study coordinated by the European Broadcasting Union had journalists from 22 public broadcasters review more than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity. Forty-five percent had at least one significant issue, and 20 percent had major accuracy problems, including outdated information.
The vendors say the same thing in their own fine print. LinkedIn's user agreement warns that AI-generated content "might be inaccurate, incomplete, delayed, misleading or not suitable for your purposes." Anthropic's help center tells Claude users not to treat it as "a singular source of truth."
Small companies get hit harder. When LinkedIn expanded Account IQ in late 2024, it extended coverage to "the 10,000 most viewed accounts with 50-200 employees." A 35-person manufacturer two counties over is unlikely to be one of them. For the companies most small B2B firms actually sell to, the AI often has less to work with, and a tool with less to work with fills the gap with whatever it can find, including a press release from 2022.
The fix takes about five minutes a company, and it is a habit more than a tool:
- Every fact that might end up in an email gets a source. If the summary says they opened a second location, find the page that says so.
- Every source gets a date. A new location from last spring is a reason to reach out. A new location from three years ago is a sign you did not look.
- Anything about a person gets checked on their profile today. People change jobs. The VP the summary names may have left in June.
- If you cannot find the source, the fact does not get used. It does not matter how confident the summary sounded.
A Burst of Ad Clicks From One Company Is a Question, Not a Buyer.
The third step is noticing which companies on the list are paying attention. This is where AI and the ad platforms produce the most exciting-looking data, and where owners most often act on a misreading.
It helps to know what the signals actually count. Sales Navigator's Buyer Intent feature looks at activity like people from the company following your LinkedIn Page, visiting it, viewing your team's profiles, engaging with your ads, completing a lead form, accepting an InMail, and visiting your website. Two catches. Website visits only count if your site has the LinkedIn Insight Tag installed. And for most of these activities you see a general profile of the person, a role and a seniority, not a name. On the advertising side, Campaign Manager's Companies Hub ranks the companies on your list by an engagement level that LinkedIn calculates by adding the paid engagement rate to the number of organic engagements. That is a measure of attention, not intent.
Third-party intent data works differently. Providers such as Bombora watch which topics a company's employees are reading about across a network of business sites and flag a "surge" when the last three weeks of reading run above that company's own 12-week baseline. Website visitor identification tools try to match your anonymous visitors to companies by their internet connection, which got noticeably less reliable once so many people started working from home. One provider admitted on its own blog that "the effectiveness of IP lookup algorithms took a dip" when lockdowns began.
None of these signals is useless. All of them answer a narrower question than it looks like. They can tell you that someone at a company did something. They cannot tell you who, why, or whether it has anything to do with buying.
Picture a regional logistics company on your list that suddenly shows ten ad engagements and three downloads of your maintenance checklist in one week. The dashboard turns it orange. The tempting move is to shift budget toward that account and have someone send a "noticed you've been looking" email. The real story could be a buying committee forming. It could just as easily be a new operations hire told to read up on vendors, an employee working on a school project, or one of your competitors doing homework from a home office the tool matched to the wrong company.
Forrester's Brett Kahnke has made the point bluntly: intent data "should not replace the qualification process." He also warns that without letting old signals expire, "it will soon appear that every company in your database is demonstrating intent." That is the other half of the trap. A spike from August still sitting on the dashboard in November is not interest. It is history.
The check is one human conversation before you change anything. If your salesperson already knows someone there, they ask a plain question: "We've seen a lot of interest from your team lately. Is something going on we could help with?" If you do not know anyone yet, treat the spike as a reason to start the research step on that company, not as permission to skip it. Then decide on budget. A single phone call costs less than a month of ads aimed at a company that was never shopping.
AI Can Find All 13 People in the Deal. It Can't Tell You Which One Is Stuck.
Most B2B purchases are not one person's decision anymore. Forrester's 2026 State of Business Buying research puts the typical buying decision at 13 internal stakeholders, plus nine outside influencers such as consultants and peers. For a small firm used to selling to one owner or one manager, that number explains a lot of deals that died quietly.
AI tools are good at the mapping part. Sales Navigator has a Lead Finder feature built to surface the decision-makers inside an account, and LinkedIn has been rolling out Message Assist, which drafts InMails using what Account IQ knows about the company. In an afternoon you can have names, titles and a draft note for most of the people who might touch your deal.
What none of these tools can see is how those people relate to each other and to you. The org chart does not say that the operations director brought in your competitor four years ago and will defend that choice, or that finance is the real decision-maker because the last project went over budget. Those are the things that decide deals, and you only learn them by asking the person who already likes you. "Who else needs to be comfortable with this, and what will they worry about?" is still the most valuable question in B2B sales, and no tool will ask it for you.
On the outreach itself, let AI draft and make yourself edit. The research only pays off if the message proves it, and a draft that opens with a compliment and a job title throws that work away. We covered what a researched first message looks like, sentence by sentence, in Your Buyer's Inbox Is Full of AI-Written Sales Emails. Only the Researched One Gets an Answer. The same standard holds for every one of those 13 people: one real reason this person, at this company, should care this month.
If your ads and your salesperson are working the same list, keep them on literally the same list. The weekly routine for that is in Your Sales Team Messages the Same Companies Your LinkedIn Ads Reach. Nobody Lines Them Up.
Your Target List Will Never Produce Many Leads. Count the Companies That Moved Forward.
This is the step where most small account-based programs get killed, and AI makes it worse, not better, because it makes the activity numbers look so busy.
A list of thirty companies is not supposed to produce a pile of form fills. If you judge it by cost per lead, it will look like the most expensive campaign you run, and somebody will turn it off in month three, right before it would have started paying. Forrester has been arguing for years that the lead was the wrong unit to begin with. Its research found that "fewer than 1% of leads convert to closed deals," and clients that delivered three or more members of a buying group to sales reported a 50 percent increase in conversion from meeting to closed deal. The thing worth counting is companies and the people in them, not form fills.
So measure the list the way you built it: company by company. Give every account on it a stage and update it once a month:
- Not reached. Your ads have not shown there yet, or nobody has seen them.
- Engaged. People there are interacting with your ads or content. This is where most of the AI signals live.
- In conversation. A real person from the company has talked to a real person from yours.
- Proposal. You have scoped work and put a price on it.
- Customer. They signed.
The monthly question becomes "how many companies moved up a stage, and which ones are stuck?" That is a question a thirty-company program can actually answer, and it tells you something useful whichever way it comes out. Notice that the AI signals only ever move a company from "not reached" to "engaged." Every stage after that requires a conversation, which is the whole argument of this article in one table.
The platforms can help with the counting if you connect them. LinkedIn's Companies Hub can show won revenue against the companies you targeted once your CRM is synced. Google Ads can import what happened after the click, but only within 90 days of it, and many B2B deals take longer than that, so send Google the stage that happens early, such as a qualified first meeting, rather than the signed contract. We walked through feeding real outcomes back to LinkedIn in Teach LinkedIn Which Leads Are Actually Worth Having, and why its default tracking window cuts off long deals in Your LinkedIn Website Tag Stops Counting at 90 Days. Your CRM Can Count for a Year. Once companies reach the proposal stage, the question shifts to how long they sit there, which is its own discipline: Your B2B Deals Don't Die. They Sit in One Stage for Months.
AI Can Read Every Company on Your List Tonight. It Still Can't Tell You Which One to Call.
Put the five steps together and the split is consistent. AI is excellent at reading: lists, websites, news, activity logs and org charts, all faster than any person could. It is unreliable at knowing what any of it means for your business this month, and at knowing whether it is still true.
A workable week for a small firm running a target list looks like this:
- Picking the list. Let AI suggest companies like your best customers. You cut the ones that fail your own customer test.
- Research. Let AI summarize each company in minutes. You source and date every fact you plan to use.
- Spotting interest. Let AI flag spikes in engagement. You ask a person before moving budget.
- Reaching out. Let AI map the people and draft the note. You edit until it holds one real reason to care.
- Measuring. Let AI pull the activity numbers. You move each company's stage by hand, once a month.
None of this needs an enterprise budget. A Sales Navigator seat, a LinkedIn campaign aimed at a well-built list, an AI research tool and an hour on Monday will run it for most small B2B firms. The time AI saves is real. Spend some of it on the checking, and the rest on the conversations no tool can have for you.
If you would rather have help building the list, running the LinkedIn ads that warm it up and setting up the stage tracking, that is the kind of B2B marketing work we do with small firms every week. And if you would rather run it yourself, everything above is yours to use.
