# How Trailspark Works: The AI Lead Scoring Process | Trailspark > How Trailspark works in four steps: ingest signals, resolve one account story, evaluate it against your definition of good, and write the result to your CRM. _URL: https://www.trailspark.ai/how-it-works_ --- How it works # How Trailspark turns signals into account evaluations Connect the tools you already run and describe what a good account looks like. Trailspark resolves the evidence into one account story, evaluates it against your definition of good, explains why, and sends the result back to your stack. [Sign up free](https://app.trailspark.ai/register)Walk through the four steps Ingest → Resolve → Evaluate → Activate ## From signal to action. Product activity, marketing engagement, CRM history, warehouse data, and web behavior resolve into one account story. Trailspark evaluates that story under the lens that applies and writes the result where your team already works. 1Ingest See what is happening Your evidence, wherever it lives - Product analytics - Marketing tools, via webhook - Website & forms - CRM - Data warehouse Product activity arrives at full fidelity - app.created · 09:41 - log\_source.connected · 11:06 - alerting.configured · Day 3 Named events, configuration milestones, and usage breadth over time Webhooks, Segment, or scheduled SQL from BigQuery 2Resolve Know which account the activity belongs to Brightwave, Incone account story - MKTGPriya Nair · webinars, pricing pages - PRODUCTSam Cole · SSO, integrations, daily usage - JOINEDdana@brightwave.com · matched by domain, no form ever filled Lead-to-account matching by email domain, or one shared record across CRM and product 3Evaluate Know what the evidence means here AcquisitionExpansionRenewal\+ your ICPs HIGHre-evaluated A champion on pricing while two engineers evaluate. That is the account you described as worth pursuing. The whole account story in one pass, under your definition of good, re-evaluated when the evidence meaningfully changes 4Activate Put the result to work Where your team already works - HubSpot / Salesforcepropensity, reason, evaluated lens & roles on the account · workflows trigger on the fields - Slackalerts routed per ICP to the channel you choose - Segmentlead evaluations as track events or identify traits Step by step ## What happens at each step What Trailspark does with your data, and the part your team owns, from the first event to the field your reps read. 1. Step 1 · Ingest ### Signals in, from the stack you already run Product events from your backend, Segment, or Heap, marketing activity by webhook, accounts and deals from your CRM, hourly or daily BigQuery syncs, and behavior on your website all arrive as signals. What you do: Connect your CRM and send the events you already track, as any JSON. You map them to signals in Trailspark. 2. Step 2 · Resolve ### One account story, in order People from your product, CRM, and marketing tools are matched to one account by CRM account, product workspace, or company email domain, even when no other system joins them. Their activity lands on one account timeline, and anonymous website activity attaches once the visitor is identified. What you do: Nothing to build. Open any account in Trailspark to read the story it resolved. 3. Step 3 · Evaluate ### The lens that applies, and the reason Each account is assigned the lens that applies to it and evaluated against that definition of good. Conflicting evidence lowers confidence, disqualifying criteria you wrote outweigh positive signals, and every propensity ships with a written reason. What you do: Describe what good looks like for each motion you sell, or start from a SparkSense draft built from your won deals. 4. Step 4 · Activate ### The result, where your team already works Propensity, the reason, the evaluated lens, and buying-group coverage are written to your CRM, alerts go to Slack, and lead evaluations go to Segment. When the evidence changes, the account is evaluated again and the fields update. What you do: Map the fields and choose the channels. The workflows you already run trigger on them. The definition of good ## Describe what good looks like. An evaluation lens in Trailspark is a written description of the account you want and the behavior that shows it. You write it in the language you already use in your planning docs. Trailspark applies it to the account story every time it evaluates. ### SparkSense · ICP setup Drafting Connect CRM 2 Read closed-won 3 Map patterns 4 Draft ICP Connecting to your CRM... ### What a lens actually says Four definitions a team might write. Nothing here is a weight or a threshold. Acquisition “A trial worth pursuing: someone technical wiring the product into their stack in the first two weeks, and at least one more person from the same company showing up. Mid-market B2B SaaS, roughly 100 to 1,000 employees.” Expansion “Room to grow: usage spreading past the team that bought, new workspaces or environments appearing, and someone running into the limits of their current plan.” Renewal “A renewal I do not have to worry about: steady weekly usage across more than a couple of people, and no drop-off in the two months before the date.” Enterprise vs SMB “Enterprise is a different bar. A hundred active users at a 15,000-person company is a foothold, so I want a second business unit involved before we call it adoption.” #### No points to tune The business judgment stays in business language instead of being translated into a point system somebody has to maintain. Each lens still carries an eligibility rule that decides which accounts it applies to, and priority rules settle which definition an account is evaluated under when more than one fits. [How to set eligibility rules on docs.trailspark.ai (opens in a new tab)](https://docs.trailspark.ai/docs/icp-eligibility-qualification) #### As many as you sell Add a definition for every motion and segment you treat differently: expansion and renewal alongside acquisition, trial users separately from free-plan users, enterprise separately from SMB. [How to manage multiple ICPs on docs.trailspark.ai (opens in a new tab)](https://docs.trailspark.ai/docs/managing-multiple-icps) #### Start from what you already won SparkSense can draft a first lens from your closed-won deals, mapping the company traits and buying roles behind them. You review and edit it before anything is evaluated against it. Or write your own from scratch. [How to draft with SparkSense on docs.trailspark.ai (opens in a new tab)](https://docs.trailspark.ai/docs/icp-sparksense-mode) Keep the business judgment in business language. Trailspark uses the definition you wrote every time it evaluates the account. Who maintains what ## You own the judgment. Trailspark delivers the story. The model sees a prepared account story. Everything around the model call is the part a team would otherwise build and maintain itself. ### You decide - Which sources to connect, and which events matter - What good looks like for each motion, in your own words - Which accounts each lens covers, and which lens wins when more than one fits - Which CRM fields, Slack channels, and Segment traits get the result - Where a lead evaluation got it wrong, with a written correction ### Trailspark handles - Matching people across product, marketing, and CRM to one account - Ordering the evidence into a timeline the model can read consistently - Assigning each account the lens that applies - Deciding when an account's evidence has changed enough to evaluate again - Writing the result back, with the reason attached Go deeper ## Each step, in more detail [AI lead scoring ### How the evaluation reads the evidence Why a point model flattens meaning, and what a reasoned evaluation returns instead. Read more](https://www.trailspark.ai/ai-lead-scoring)[PQL scoring ### Product activity, read as part of the account Every product user evaluated with the account they belong to, not against a threshold. Read more](https://www.trailspark.ai/product-qualified-leads)[Buying Group Intelligence ### The people behind the account Which buying roles are engaged, which are latent, and which are missing. Read more](https://www.trailspark.ai/buying-group-intelligence)[Integrations ### What connects, and what gets written back CRM, product data, marketing, CDP, enrichment, and notifications, with a setup guide for each. Read more](https://www.trailspark.ai/integrations) ## How Trailspark works: common questions ### In four steps. Ingest: product events, marketing activity, CRM accounts and deals, warehouse data, and website behavior arrive through native integrations and webhooks. Resolve: the people behind that activity are matched to one account and ordered into one account story. Evaluate: the account is judged against the plain-language definition of good that applies to it, with a written reason. Activate: the propensity, the reason, the evaluated lens, and [buying-group coverage](https://www.trailspark.ai/buying-group-intelligence) are written back to your CRM, with alerts in Slack and lead evaluations in Segment. ### First-party evidence from [the tools you already run](https://www.trailspark.ai/integrations): named, timestamped product events sent from your backend, Segment, or Heap; marketing activity from any tool that can send a webhook; accounts from Salesforce, HubSpot, or Airtable, and deals from Salesforce or HubSpot; hourly or daily BigQuery syncs; and behavior on your own website. For extra coverage, you can connect your own Clay account for company details, or Reo.dev or PredictLeads for buying signals on accounts you already track. ### People are matched to one account by CRM account, product workspace, or company email domain, even when no other system joins them. [A product user with a work email](https://www.trailspark.ai/product-qualified-leads) lands on the right account without ever filling out a form, and a visitor's earlier anonymous website activity attaches once they are identified. Personal email domains such as Gmail are never used to join people to an account. ### A lens is a written description of the account you want for one motion and the behavior that shows it, such as acquisition, expansion, renewal, or cross-sell. Each lens also has eligibility rules that decide which accounts it covers. You write it in plain language [instead of assigning points or weights](https://www.trailspark.ai/ai-lead-scoring), and you can add one for every motion and segment you treat differently. In the Trailspark app, lenses are set up as ICPs. ### Lenses are kept in a [priority order](https://docs.trailspark.ai/docs/managing-multiple-icps). Each account is evaluated under the highest-priority lens whose eligibility rules it matches, unless an admin pins a lens to that account, and any other lenses it also matches are shown alongside. An optional default lens covers accounts that match no other rules. Without a default, those accounts are not evaluated, which is a deliberate way to evaluate only the accounts you have scoped. ### When the evidence behind the evaluation has actually changed, not on every event. A daily check catches accounts that have gone quiet, but the model only runs again if something it reads has changed, and a cooldown holds an account steady through a busy week. Each re-evaluation updates the current call and reason, and [an account counts once per billing month](https://www.trailspark.ai/pricing) no matter how many times it is re-evaluated. ### Where your team already works. Propensity, the written reason, the evaluated lens, the evaluation date, and buying-group coverage are written to Salesforce, HubSpot, or Airtable fields you map, and per-person buying roles to Salesforce or HubSpot, so your existing workflows can trigger on them. [Slack alerts are routed per lens](https://www.trailspark.ai/integrations/slack) to the channel you choose, and Segment can receive each lead evaluation as a track event or identify traits. MCP clients such as Claude can read your workspace, and configure it if you allow, through scoped agent keys. ### At each step. Ingest: it reads the first-party product, CRM, and website activity you already collect, not third-party intent data and IP-to-company identification, as in 6sense, or anonymous visitor identification, as in Warmly. Resolve: it evaluates the whole account and the buying group behind it, not one lead at a time. Evaluate: it judges the account against a definition of good you write in plain language, not point rules you maintain, as in HubSpot's native scoring, or a predictive model, as in MadKudu. Activate: the written reason and the evaluated lens land in your CRM next to the propensity. [See each tool compared](https://www.trailspark.ai/alternatives), including Pocus and Common Room. ### No. There is no scoring model to train and no propensity weights to derive. You describe what good looks like in plain language, the way you would explain it to a new rep. If your CRM has closed-won deals, [SparkSense can draft a first lens from them](https://docs.trailspark.ai/docs/icp-sparksense-mode), and you edit it before anything is evaluated against it. ### The evaluation prompt leaves out contact names and email addresses. People are referenced by internal ID and job title, and accounts by ID, company name, and firmographics such as industry, region, and employee count, alongside the signals and when they happened. [The Security page](https://www.trailspark.ai/security) covers AI data handling in detail. ## Evaluate every account for what comes next. The Free plan is the full platform on a small set of accounts. Connect the signals you already collect, describe what good looks like, and read the reasoning behind every result in your CRM. [Sign up free](https://app.trailspark.ai/register)