Inside the PrescientIQ Prospecting Agent: how it finds, scores, and hands off accounts

Most target lists are built once and worked until they run out. In between refreshes, accounts that just showed real buying intent sit alongside accounts that went cold months ago, indistinguishable on the list a rep opens Monday morning. A commonly cited industry range puts CRM record decay at up to roughly 30% within a year — enough stale data that reps lose real selling time to manual list hygiene before they ever open a sequence tool. Separately, a human SDR working that list by hand tops out at somewhere around 40 to 60 genuinely personalized touches a day, regardless of how good the list is.

The Prospecting Agent is built against both constraints at once: it removes the staleness problem by scoring continuously instead of periodically, and it removes the research bottleneck by handing a rep a ranked, qualified account instead of a name and a company domain.

What it actually does

On our own pricing page, this is described in one sentence: it “identifies ICP-fit accounts from intent and firmographic signals with deterministic, sandboxed scoring — no LLM arithmetic on anything with a numeric consequence.” Unpacked, that is three distinct steps running continuously rather than on a schedule.

First, it ingests signals — CRM activity, firmographic data, and whatever intent sources you already license — rather than waiting for someone to pull a report. Second, it scores each account against your ideal customer profile using sandboxed, deterministic code, not a language model’s best guess at a number. Third, it hands qualified accounts to the Outbound Agent’s approval queue with no manual research step in between.

Why the scoring is deterministic, not model-generated

It is tempting to let a language model produce a priority score directly — it is fast, and the output looks plausible. The problem is that “plausible” and “reproducible” are different properties, and a score a rep is going to act on needs the second one. Two identical accounts scored a week apart should get the identical number, and if a score looks wrong, someone needs to be able to trace exactly why it came out that way.

That is why any calculation with a numeric consequence — a rank, a priority, a threshold that decides whether an account gets worked today — runs through sandboxed deterministic code instead of being asked of the model. The model still does the language-heavy work upstream (interpreting signals, adding context); the arithmetic that decides what a rep sees first does not touch it.

What a rep actually sees, day to day

In practice this collapses the first hour of a rep’s morning into nothing. Instead of opening five tabs — the CRM, an intent platform, a firmographic lookup, a LinkedIn search, and a spreadsheet someone half-maintains — a rep opens one list, already ranked, with the signal that earned each account its position attached to the row. An account that just showed a real trigger — a leadership change, a technology-stack update, a spike in product-page visits — shows up at the top with that trigger named, not buried in a separate tool a rep has to remember to check.

What does not change is who decides an account is worth working. The agent surfaces the ranking and the reason behind it; a rep or their manager still sets the bar for what “qualified enough” means for their own territory, and can override the agent’s ranking at any time. Scoring removes the research tax, not the judgment call at the end of it.

From scored account to working pipeline

A qualified account does not sit in a report waiting to be noticed — it moves. The hand-off to the Outbound Agent happens without a manual step, and from there the account progresses through a normal pipeline, with each stage change and every AI-assisted read on sentiment logged as it happens rather than summarized after the fact.

Nothing external leaves on the strength of the Prospecting Agent’s score alone. Every draft the Outbound Agent produces from a qualified account still waits for a named human to approve it:

What else the platform surfaces once you are talking

Once a qualified account turns into an active conversation, the same governed approach extends past prospecting. A call or demo recording gets parsed into structured decisions and action items instead of relying on someone’s notes — a downstream convenience, not something the Prospecting Agent itself performs.

A named human still owns every commitment that view surfaces — the agent drafts the record, not the decision. That distinction holds everywhere in the platform, which is the point of the next section.

Every action, logged before it is trusted

A score, a stage change, an approval, a rejection — each is written to an append-only ledger with the actor, the confidence, and the rationale at the moment it happens, not reconstructed from application logs afterward if someone asks.

That is also what a security review is actually checking for when it asks about an AI tool’s audit trail — not whether the product can produce a summary on request, but whether the record existed before anyone asked. We cover what a review checks in full in a dedicated checklist.

Where it fits, and what it does not do on its own

The Prospecting Agent is one stage of a longer loop, not the whole motion. It does not draft or send outbound — that is the Outbound Agent’s job, gated by approval. It does not manage a trial once an account converts, and it does not surface expansion signals on an existing customer — those belong to later stages of the same coordinated loop, covered in a companion post on full revenue loop coverage. And it inherits whatever your CRM data already is — a Prospecting Agent scoring against inconsistent or duplicate records will surface that inconsistency faster, not fix it silently. We cover that specific failure mode in a deeper look at CRM data debt.

A typical deployment timeline, subject to your own CRM’s data quality and integration scope:

Where the data runs

The architectural answer, stated plainly:

PrescientIQ is hosted and operated by MatrixLabX on Google Cloud, which maintains SOC 2, ISO 27001, and PCI DSS-attested infrastructure. Per-agent least-privilege identities, prompt-injection defense on every inbound surface, and an immutable audit ledger record every action, its rationale, and the approving human.

What it costs

The Prospecting Agent ships as one line of the full Revenue Accelerator platform fee — on our pricing page it is listed as the “Prospecting Digital Worker,” alongside Outbound, Trial Conversion, and Expansion, under one published rate:

How that compares to what you are paying now

A platform fee only means something next to whatever is currently doing this work. For most mid-market teams that is one of two things: an agency retainer, or an in-house SDR pod. Both comparisons below are cost-structure comparisons only. Neither claims that one agent produces the output of a given number of people — that is a measurement we do not have and will not assert. What is comparable is what each model charges for, what it includes, and how the cost behaves when volume changes.

1. Against a mid-market agency retainer

An outbound agency sells the same function this agent performs: somebody else finds and prioritises the accounts. Retainers for that coverage commonly run somewhere between $60,000 and $300,000 a year in the mid-market, scoped by team size and channel mix — a general range worth using to sanity-check a quote rather than a number for any specific firm.

What a retainer buys that this does not: human judgment on genuinely ambiguous accounts, a named strategist a prospect can be escalated to, and the ability to redirect effort mid-quarter with a phone call rather than a configuration change. Those are real, and a platform priced on execution does not replace them by default. The full version of this math is in our agency-retainer cost comparison.

2. Against a seven-person SDR pod

The in-house comparison is harder, because the number most teams carry in their head is the base-salary line rather than the run rate. For a seven-rep pod — the standard shape, since seven is about one manager’s span of control — the published model looks like this:

The structural difference is not the headline gap between that run rate and a platform fee — it is what happens next year. A pod scales by hiring, so covering more accounts means another seat, another ramp, and eventually another manager. A platform fee does not move with volume inside a typical deployment. That is the comparison worth running; the rest is arithmetic on your own numbers.

The line-by-line decomposition, including the turnover and ramp drag that produce the 76% effective-capacity figure, is in the seven-person SDR team cost model, with the per-seat multiplier broken out in base salary vs. fully loaded cost.

Which comparison is actually yours?

The honest answer depends on what is doing this work for you today. This routes you to the version of the audit that starts from your actual baseline.

Can you run the audit on your data right now?

The Autonomous Audit Report is free, covers six dimensions — governance, non-human identity, shadow AI, data readiness, workflow suitability, and evidence — and comes back as a written assessment within 48 hours of the intake session. The most common reason teams delay it is a belief they need to clean something up first. Usually they do not.