Predictive lead scoring API
POST a lead, an account, or an opportunity. Get back a 0-100 conversion likelihood score and the CRM fields that drove it, from a model trained on your own conversion history. Two endpoints, Bearer key auth, no SDK.
Three different things get called a lead scoring API
Worth sorting out before you integrate anything, because the three fail in different places.
Enrichment API
Send a domain or an email, get back firmographic and contact data.
The response is data about the record, not a prediction about it. Something still has to decide what the data means.
Platform score endpoint
Read back a score the platform already computed on its own schedule.
You cannot score a record the platform has not seen, and the refresh cadence is the vendor’s, not yours.
Scoring service
A model trained on your outcomes, scoring records you send, on demand.
You can score anything you can describe, including a record that has never touched your CRM.
ax1om is the third kind. It trains a dedicated model per customer on that customer's own conversion history, converted and non-converted records alike, and scores whatever you send it. No third-party intent data, no model shared across customers.
What you send, what you get back
POST /v1/score takes up to 1000 records in one call and
returns one row per record, in order. Send the fields you have. You do
not need every field the model trained on.
curl -X POST https://api.ax1om.ai/v1/score \ -H "Authorization: Bearer ax1m_sk_your_key" \ -H "Content-Type: application/json" \ -d '{ "model_id": "mdl_abc123", "records": [ { "title": "VP Revenue Operations", "industry": "SaaS", "employees": 250 } ], "explain": true }'{ "model_id": "mdl_abc123", "scores": [ { "score": 82, "conversion_likelihood": 0.823, "enriched": false, "top_reason_1": "Title seniority", "top_reason_2": "Industry: SaaS", "top_reason_3": "Company size band" } ], "records_scored": 1, "usage": { "current": 151, "limit": 10000, "remaining": 9849 }}
The score is a 0-100 ranking of conversion likelihood and
conversion_likelihood is the same signal as a decimal. It
ranks; it is not a percentage chance of conversion, and the per-record
factors do not sum to it. Set explain and each row carries
the fields that drove it, up and down, so the number arrives with its
reasons attached rather than as an assertion.
Records the model has never seen
A lead that is not in your scored data still gets a real score from the
fields you send, flagged enriched: false. That is what
makes the endpoint usable on an inbound form or in a prospecting
workflow, not just on records you already own.
One record, flat response
POST /v1/webhooks/score is the single-record variant. The
response is deliberately flat, so tools that map fields by drag and
drop can consume it without a JSON path. model_id rides in
the query string, which keeps the body exactly the record.
curl -X POST "https://api.ax1om.ai/v1/webhooks/score?model_id=mdl_abc123" \ -H "X-API-Key: ax1m_sk_your_key" \ -H "Content-Type: application/json" \ -d '{ "Title": "VP Sales", "Industry": "SaaS", "LeadSource": "Webinar" }'This endpoint is inbound: your system posts a record and reads the score off the response. ax1om does not deliver outbound webhooks to a url you supply, so there is nothing to register and no delivery to retry.
Where the call comes from
An AI agent
Agents reach the same predictions over MCP instead of raw HTTP, so the scoring logic never has to live inside the agent.
Salesforce
An Apex callout or an External Service on lead create and on field update. Write the score and the top factors into fields reps already see. For a no-code path, CRM writeback keeps scores current without any call at all.
HubSpot
A workflow custom-code action calls the webhook endpoint and writes the flat response into contact properties. The response is deliberately shallow so property mapping stays a drag-and-drop job.
Marketo, Zapier, and friends
Anything that can POST JSON and read a flat response can use the webhook endpoint. model_id rides in the query string so the body stays exactly the record.
Agents use the MCP server, which serves the same predictions as tools. An agent that can ask for a score has no reason to guess at one.
The same model, callable by your agents
Conversion scoring is live on ax1om's MCP surface today. Point an agent at it and it calls the same trained model this API serves, gets the score back with the factors that drove it, and shows its math instead of writing a plausible number.
ax1om for AI agentsKeys, speed, and limits
Keys and test keys
Mint a key in the app. Secrets start with ax1m_sk_
and are shown once. A test key scores against the real model but
meters into a separate free bucket of 1000 records a month, so
you can wire an integration up and verify the shape before
anything counts.
Speed
Roughly 5 to 20 ms of inference once the model is warm, and up to 1000 records per call. Training is separate and asynchronous, so a scoring call never waits behind a model run.
Allowance
API scoring is on Pro and above. Pro includes 10K live API scores a month, Enterprise 30K. There is no overage SKU: past the limit scoring continues in a grace band rather than failing mid-integration, then stops with a message rather than a bill.
Errors use one stable envelope with a frozen code enum, so clients can
branch on error.code instead of parsing prose. The API is
versioned by URL prefix. Full contract, field tables, error codes, and
versioning policy live in the API reference.
Common questions
What is a lead scoring API?
An interface that returns a score for a lead, account, or opportunity so your own systems can act on it. The useful distinction is what produces the score: an enrichment API returns data and leaves the judgment to you, a platform endpoint returns a score the platform already computed on its own schedule, and a scoring service runs a model trained on your outcomes against records you send on demand. ax1om is the third kind.
How is a lead scoring API different from an enrichment API?
Enrichment answers "who is this". Scoring answers "how likely is this to convert, and why". They are complementary: enrichment often supplies fields that a scoring model then uses. ax1om does not sell enrichment and does not use third-party intent data, so the fields it scores on are the ones already in your CRM.
Can I score a lead that is not in my CRM yet?
Yes. Send the fields you have and the model scores from those, flagged as a cold record. You do not need every field the model trained on, and the record does not need to exist anywhere first. That is what makes the API usable on inbound forms and in prospecting workflows rather than only on records you already own.
What does the score actually mean?
A 0-100 ranking of conversion likelihood, alongside the underlying likelihood as a decimal. It is a ranking, not a percentage chance of conversion, and the per-record factors do not sum to it. Turn explanations on and each score comes back with the CRM fields that drove it, up and down.
How fast is it?
Inference runs in roughly 5 to 20 ms once the model is warm. Batch up to 1000 records in a single call. Training is separate and asynchronous, so a scoring call never waits on a model run.
Do I need a data science team to use it?
No. Training is a guided flow in the app, not a modelling project: pick what you predict, train, and mint a key. A first model trains on as few as 50 conversions. The API serves whatever that flow produced.
Can an AI agent call it?
Yes, and over MCP rather than raw HTTP if you would rather not embed scoring logic in the agent. The ax1om MCP server exposes the same predictions as tools, so the agent asks for a score and reasons about the answer instead of inventing one.
Train a model, then call it
The API serves whatever your training flow produced. Start free, train on your own data, and see the model before you wire anything up. The same endpoint scores organizations as well as people, so see account scoring if the record you want ranked is a company.