How to build a prospect universe using AI

By Dave Curran, Co-Founder, Firmbase | March 2026 | 11 min read

Building a target account list is one of those things that sounds simple in theory and becomes a nightmare in practice.

"Find all the companies that match our ICP" - easy enough. You know who your customer is. You know their industry, size, location. You know what they care about.

Then you actually try to find them.

You search a list-building tool. You configure filters. You get a list of 5,000 companies. You look at the first 20. Half of them aren't actually relevant - they technically match your filters but don't fit your ICP in practice. So you add more filters to narrow it down. Now you've got 300 companies, but you're probably missing half the market because your filters are too restrictive.

This is where most teams get stuck. The problem isn't that account lists are impossible to build. The problem is that the traditional approach - filtering on standardised data like SIC codes and company size - works right up until it doesn't.

This article walks through a better framework: define and discover, score and prioritise, research and action.

Stage 1: Define and discover using natural language

The first problem with filters is that they assume standardisation. They assume everyone categorises themselves the same way, and that official categories (like SIC codes) tell you what companies actually do.

They don't.

SIC code 6201 (software development) includes developer tool vendors, management consulting firms, and in-house tech teams at other company. Same code, completely different businesses.

A company's description says "we're a digital agency." But are they digital marketing, digital design, digital production, all of the above? You can't tell from the company description alone. You need to look at their website, their job postings, their customer list.

Natural language search flips this on its head. Instead of filtering on SIC codes and standardised fields, you describe what you're looking for in plain English: "digital marketing agencies in the South East with revenue between £2M and £8M."

The system doesn't just filter on SIC code and location. It reads company websites, understands what they actually do, understands what "digital marketing agency" means as opposed to "design agency" or "marketing consultancy," and surfaces companies that match.

The quality is immediately better. You're not sorting through a list of false positives. You're looking at accounts that actually fit.

This is discovery, not filtering. And it's much more effective for building your initial universe.

Why SIC codes fail for account selection

SIC codes are broad, outdated, and unreliable for modern businesses. A SaaS company that started as a consulting business might be categorised as "management consultancy" even though it's now pure software. A company that pivoted from services to products still carries its original classification. Companies doing hybrid models get shoehorned into the closest code.

Even worse, multiple SIC codes might fit the same company, and the primary code assigned depends on what the company reported to Companies House, not on an objective assessment of what they do.

Natural language understanding bypasses all of that. It reads the company's actual website, understands their positioning, and categorises them by what they do, not by what they registered as.

Stage 2: Score and prioritise using signals

Once you've discovered your universe - let's say you've found 2,000 UK companies that actually match your ICP - the next problem is: which 200 should you actually target first?

This is where signals matter.

A signal is a public indicator that a company might be interested in buying something. The strongest signals are structural changes:

Companies House filings: Recent revenue growth (filed accounts showing turnover increase), cash position changes (suggesting they're either cash-rich or burning through reserves), and creditor balances (if growing faster than financing, they might be taking on more debt to fund growth).

Director appointments: A new finance director usually signals upcoming fundraising, acquisition, or growth acceleration. A new commercial director or sales director signals revenue focus and potential tool evaluation. A new ops hire signals systems and tooling investment.

Job postings: Multiple new job postings in revenue roles (SDRs, AEs, sales ops) signals scaling and infrastructure investment. New ops hires signal process maturity. New technical hires signal product investment.

Funding announcements: Seed round, Series A, growth capital - all signal growth mode, budget availability, and openness to new tools.

The hierarchy for UK SMB prospects is roughly:

The key point: don't just look at one signal. Weight them. A company with one signal might be worth investigating. A company with three signals (new finance director, growing revenue, hiring SDRs) is definitely worth your time first.

Stage 3: Research each priority account in depth

This is where most teams break down. You've got a prioritised list of 100 accounts. Now you need to know why you're reaching out to each one.

The traditional approach: spend 30 - 45 minutes per account researching. Read their website, check their LinkedIn, scan their job postings, look at their Companies House filing, understand their positioning, figure out why they might need what you're selling.

Do that for 100 accounts and you've lost 50 - 75 hours. Most teams don't have that time.

That's where AI agents come in.

An AI agent can do that research in 2 - 3 minutes per account:

Input: company name, Companies House number (or website URL), description of what you sell.

Process: The agent reads the company's website and understands what they do, who their customers are, and their positioning. It cross-references their job postings to understand strategic priorities and hiring patterns. It pulls their Companies House filing to understand financial trajectory and recent structural changes. It synthesises all of that and identifies why this company might need what you're selling.

Output: A structured research brief with outreach angle. "You appointed a new Finance Director in [month], suggesting you're preparing for growth. Your company is growing revenue at [X]% annually. You've posted jobs for 3 SDRs in the last quarter, suggesting you're scaling sales. We work with companies like yours to help with [specific use case]."

You don't have to spend 45 minutes researching. The agent does the work. You get a one-page brief, and you reach out with concrete context.

Multiply that by 100 accounts. You're saving 50 hours and getting better, more contextual outreach.

Bringing it together: the workflow

Week 1: Define and discover

You know your ICP. You've worked with your team to define it clearly: "B2B SaaS companies in the UK with revenue between £1M and £10M, doing product-led growth, with a first sales hire."

Instead of filtering on SIC codes and employee ranges, you use a tool that understands natural language and scans company websites. You get back 1,500 companies that actually match.

Week 2: Score and prioritise

You take that list of 1,500. You run a signal analysis: which companies have made director appointments in the last 12 months? Which have shown revenue growth in their filed accounts? Which are hiring in sales roles?

You weight those signals. You identify your top 100 priority accounts - the ones showing clear signals of growth and openness to new tools.

Week 3: Research and action

You use an AI agent to research each of your top 100. For each one, you get a one-page research brief with specific context: what they do, why they might need you, what signals suggest they're in buying mode.

You start reaching out to the top 20 this week. By next month, you're working through the rest.

The whole process - from raw market discovery to prioritised, researched outreach list - takes three weeks. With a manual approach using filters and hand research, it would take two months.

More importantly, the quality is better. You're not reaching out blind. You're reaching out with concrete context based on real signals.

Why this framework works at scale

Most teams find that once they've done this once, the process becomes continuous.

You've got your universe of 1,500 companies. Every month, new signals emerge: new directors appointed, new growth visible in filings, new job postings. You rescore your list. Your top 100 changes as new accounts hit your criteria.

Instead of "we need to rebuild our target list," it's "we need to update our priority ranking based on this month's signals."

That's a running system, not a project.

The framework works better when you have systems that do the heavy lifting

Instead of spending time on definition and discovery, score your entire UK market once and let the system continuously resurface accounts as they hit buying signals.

Start your free trial and see how natural language discovery + signal scoring changes your targeting.

FAQ

Q: How do I get accurate data about director appointments?

A: Companies House filings are the source of truth for UK director changes. They're public record and updated regularly (within two weeks of filing). That's more current than any database.

Q: Should I weight all signals equally?

A: No. A director appointment + funding announcement + job postings is a stronger signal than just job postings. You're looking for convergence - the more signals pointing to growth or change, the higher the priority.

Q: What if a company matches my ICP but shows no signals?

A: Add them to your universe but de-prioritise them. They might be in buying mode, just not showing visible signals yet. They're your secondary target list. Your primary list is accounts showing clear signals.

Q: How often should I rescore my universe?

A: Monthly is realistic. New director appointments happen continuously, accounts file new financial reports, job postings change. A monthly rescore keeps your priority list current.

Q: Can I do this manually, or do I need software?

A: You can do stages 1 and 2 manually if you have time. Stage 3 (researching 100+ accounts) is where software makes a huge difference. 45 minutes per account doesn't scale.

Q: What if my product appeals to multiple personas within a company?

A: You might create separate signal weightings. Finance-focused buyers care about director appointments and cash flow. Operations-focused buyers care about hiring and scaling signals. Define signals per persona.

Author Bio

Dave Curran is the co-founder of Firmbase, a UK B2B sales intelligence tool that helps sales teams find, prioritise, and reach the right accounts without needing a RevOps team to make it work. Before Firmbase, Dave co-founded Love Mondays (acquired by Glassdoor, where he went on to serve as VP of Product) and Openvolt. He writes about UK B2B sales, prospecting, and go-to-market strategy.

Firmbase helps UK B2B sales teams discover their complete account universe, prioritise based on real buying signals, and reach out with genuine relevance - without the complexity of enterprise tools. Start your free trial