Skip to content
Outbound Intelligence

Scout.

Manual sourcing is slow, inconsistent, and impossible to run at volume. Scout is a five-phase AI pipeline that monitors multiple sources, scores each opportunity against a target profile, personalises outreach, and tracks every response, twice a day, untouched.

Status
Running in production
Type
Autonomous pipeline
Runs / day
2 (automated)
Human input
One-tap approvals
565+
Leads per week
<£0.10
Cost per scored lead
5
Pipeline phases
0h
Manual input per run

The problem

Outbound sourcing at volume is broken.

Sourcing across channels by hand takes hours, and most of that time is spent on leads that are obviously wrong. The few genuinely relevant targets get the same unfocused attention as everything else. Follow-up is inconsistent. Nothing is tracked.

The real problem isn’t finding opportunities, it’s signal-to-noise. Hundreds of signals exist; a handful are worth pursuing. Finding them manually at volume is not a sustainable operation.

Scout solves it with an AI scoring layer that reads the same signals a human analyst would, against a defined profile, without the time cost.

Before Scout

2 to 3 hours daily sourcing across channels manually
Generic outreach with no personalisation
No tracking of who was contacted or when
Follow-up inconsistent or forgotten entirely

After Scout

565 opportunities sourced and scored before 7 AM, twice daily
Outreach drafted for each target and context
Every contact logged in a GDPR-compliant schema
Telegram review takes under 5 minutes per batch

How it works

Five phases. One pipeline.

  1. 1

    Data Ingestion

    Scrapes multiple sources on a schedule, normalises every listing into one schema, and deduplicates against previous runs. Nothing missed, no duplicates through.

  2. 2

    AI Scoring Against ICP

    Claude Haiku scores each opportunity against a defined Ideal Client Profile and returns a fit score with reasoning. Chosen for cost and latency, hundreds scored for pennies.

  3. 3

    Profile-Matched Personalisation

    For anything above threshold, the system writes outreach tailored to that specific target and context. Not a template, it reads the full brief and writes to it.

  4. 4

    Outreach Automation

    A Telegram flow presents scored cards with pre-drafted outreach for one-tap approval. Daily volume caps protect sender reputation; GDPR cleanup is built into the schema.

  5. 5

    Analytics & Follow-Up

    Every contact is tracked, opens, responses, follow-up timing, so scoring thresholds and outreach copy keep improving over time.

The only human touchpoint

A Telegram approval flow. Each morning, scored cards arrive with pre-drafted outreach. One tap approves and sends; one tap archives. A deliberate UX call, make the daily review take under five minutes, not replicate a full CRM.

Model choice

Claude Haiku over GPT-4

Lower cost and faster latency for a task that runs at volume. Scoring needs consistency and speed, not frontier reasoning.

Architecture

Volume caps built in

Daily send limits set at the schema level, not bolted on later. Protecting sender reputation was a day-one requirement.

Compliance

GDPR from day one

Contact cleanup is baked into the data schema from the first version, not retrofitted at the end.

Memory

Living spec file

A continuously updated implementation record prevented feature regression across all five phases.

Technology

Built with the right tools.

Every tool was chosen for a reason. Claude Haiku scores at volume without burning budget. FastAPI serves the internal endpoints. Telegram delivers approvals without a separate dashboard. SQL keeps the data model simple and queryable.

Each pipeline stage is discrete. If one fails, the others continue. Failures are logged and surfaced, never silently dropped.

Claude API (Haiku)PythonFastAPISQLTelegram Bot APIREST APIsScheduled jobs (cron)GDPR-compliant schemaJSON structured outputs

Why no vector database?

Scout uses structured AI scoring, not semantic search. A prompted score with reasoning is more interpretable and auditable than a similarity vector. You can read why a lead scored 8/10. You can’t read a cosine distance.

What we learned

The things that only show up in production.

Model latency matters more than capability at volume. GPT-4 scored marginally better; Haiku scored good-enough at a fraction of the cost and 4× the speed. For a task running hundreds of times a day, the trade-off is obvious.

The approval UX is as important as the pipeline. An early version sent raw scored data to review. Nobody read it. One-tap Telegram job cards cut review time from 20 minutes to under 5.

GDPR compliance is not a feature you add at the end. Building contact cleanup into the schema from the start cost nothing. Retrofitting it would have taken days and added risk.

A living spec prevents drift across long builds. A five-phase project built over months regresses without a continuously updated record of what was decided and why.

Need a pipeline built for your process?

Scout was built for outbound. The same architecture applies to any repetitive sourcing, scoring, and outreach workflow. Tell us the problem.