Almost everyone who reads your pricing page leaves without typing anything. They read the docs, they compare you to two competitors, they send a link to a colleague, and none of that reaches your CRM. Then a quarter later someone books a demo and the first question in the room is how they found you.
What this piece gives you
- Treat it as a pipeline, not a campaign. A campaign fails quietly. A pipeline fails at a stage you can name.
- Score fit and intent separately. Blended scores collapse into a middle where nothing gets acted on.
- Exclusions are most of the build. Skip them and your top account will be a university.
- Routing is where automation stops. The list has to be short enough for one person to read on a Monday.
- A prompt at the bottom runs the first pass of this on your own business, in your own agent.
The usual fix is to buy an account list and run a campaign at it. That is the wrong shape. A campaign is a thing you launch and then report on. What this actually needs is a set of stages that run continuously, each one with an input, an output, and a way to tell whether it is working.
Run as a campaign
Launches, runs, gets reported on. When results are poor you learn that the whole thing underperformed, and the next move is to guess which part was at fault.
Run as a pipeline
Always on, stage by stage. When results are poor you can see that identification is fine, enrichment is thin, and scoring is the broken part. You fix that one.
This is how I have built that motion for a B2B developer-tool company. The tooling is deliberately generic below, because the choice of vendor is the least interesting decision in it.
The five stages
| Stage | What goes in | What comes out | How you know it broke |
|---|---|---|---|
| Identify | Anonymous sessions | Company names, probabilistic | Resolution rate falls, or resolves to nonsense |
| Enrich | Company names | Size, stage, stack, region, funding | Fields come back empty on your best accounts |
| Score | Enriched records | A fit score and an intent signal, kept apart | Everything lands in the middle of the range |
| Route | Scored accounts | A ranked list with evidence attached | The list is longer than anyone reads |
| Measure | Touched accounts | Which ones opened pipeline | You cannot do the join at all |
Fit and intent are not the same number
The first version of the scoring I built combined firmographic fit with behavioural signals into a single score. An account that matched the ideal profile perfectly but had read one blog post scored about the same as a poor-fit account that had hammered the pricing page four times in a week. Both came out somewhere in the middle, and the middle is where nothing gets acted on.
Keep them separate. Fit is a property of the company and changes slowly, if ever. Intent is a property of this week and decays fast. When they are separate you can say useful things:
| Low intent | High intent | |
|---|---|---|
| High fit | Nurture segment. Right company, wrong week. | This is the list. Route it today. |
| Low fit | Ignore. Costs nothing to skip. | One human look, then usually ignore. |
The trap: behavioural signals feel more actionable than firmographic ones, so they quietly dominate any blended score. You end up prioritising whoever browsed the most, which on a developer-tool site is frequently a student, a competitor, or your own team on a VPN.
Exclusions are most of the work
Nobody writes about this part because it is boring, and it is where the majority of the build time goes. Before any of the scoring means anything, remove:
- Your own team, on every network they use, including home connections.
- Existing customers, unless you are deliberately running expansion.
- Competitors, who visit constantly and look exactly like high intent.
- Anyone who has asked not to be contacted. This has to survive every later rebuild of the list.
- ISPs, universities, hosting providers, and the long tail of resolutions that are accurate and commercially meaningless.
The last one is the surprise. A large share of what reverse-IP hands back is real, correct, and useless. If you do not filter it, your top-scoring accounts will be a telecoms company and a university, and the first person you show the dashboard to will stop trusting it. That is very difficult to undo.
Not sure whether your traffic is worth this? That is usually a twenty minute question, and the answer is often no.
Book fifteen minutes →Routing is where the automation stops
Identification, enrichment and scoring can run unattended. I have not found a version of the next step that can. Deciding that a specific account is worth a specific person's next hour is a judgement call, and the moment it is fully automated you are sending generic messages to companies who did nothing more incriminating than read your documentation.
So the pipeline is built to stop. It produces a short ranked list with the evidence attached, and a person decides. The useful constraint is on length: if the list is longer than a person will actually read on a Monday morning, the threshold is too low, and the honest fix is to raise it rather than to hire someone to work through the overflow.
This is the same thing that happens everywhere agents get pointed at marketing work. The machine moves the bottleneck to the human step. It does not remove it. Every account sitting in a ranked list unread is not progress, it is work owed, and it should be counted that way.
What I would do differently
I built the dashboard before the exclusions. That is backwards. The first view anyone saw was full of noise, and the credibility cost of that first impression was higher than the time it would have taken to filter properly beforehand.
I also scored on too many attributes at first, because each one seemed defensible in isolation. A model with a dozen weighted inputs is impossible to argue with in a room, which sounds like a strength and is not. When a salesperson asks why an account scored well, the answer has to fit in a sentence. If it does not, they will use their own judgement instead, and then you have bought software to produce a number nobody uses.
Run the first pass yourself
You do not need a vendor to find out whether this is worth building. You need an hour and an honest answer about who actually buys from you. Paste the prompt below into Claude, or any agent you already talk to, and work through it.
Copy this into your agent
Build your fit model before you buy anything
It will interview you, draft a defensible fit score, keep intent separate, and hand you the exclusion list you will need first. Nothing here needs access to your data.
I want to turn anonymous traffic on my website into an account motion, and I want to design it before I buy any tooling. Work through this with me one step at a time, and ask me for what you need rather than assuming. 1. Ask me what my company sells, and what my best-fit customer actually looks like: size, stage, technical stack, region, and who signs the contract. 2. From my answers, draft a FIT score using at most five attributes. Give each a weight, and for each one write the single sentence a salesperson would accept as the reason it matters. If I cannot defend an attribute in one sentence, tell me to drop it. 3. Separately, draft an INTENT signal list based on pages a serious evaluator would read on my site. Keep this completely apart from the fit score. Do not blend them into one number, and tell me why if I ask you to. 4. Write the exclusion list I need before any of this is trustworthy: our own team, existing customers, competitors, anyone opted out, and ISP, university and hosting resolutions. Ask me for the specific names where you need them. 5. Tell me which single page on my site only a serious evaluator would read, and why you picked it. 6. Give me a threshold rule that produces a list short enough for one person to actually read on a Monday morning, and tell me what to do with everything below it. Finish with the three things most likely to make this fail for my business specifically. Where you are guessing about me, say so instead of guessing.
If you would rather start smaller
Pick the page that only a serious evaluator reads, usually pricing or a specific integration doc. Resolve the companies that read it, filter out the five categories above by hand, and look at what is left. If that list is interesting, the pipeline is worth building. If it is not, no amount of tooling will make it so, and you have found that out in an afternoon instead of a quarter.
Written from work done for a B2B developer-tool company. Specific tools, figures and account names are deliberately left out. If you want to talk about whether this shape fits your traffic, fifteen minutes is usually enough to tell.