Skip to content

Use case · Sales research

Sourced account briefs and outreach drafts a rep approves

Before a call or a first email, reps need to know what changed at the account and why it matters now. Three agents prepare it: a researcher gathers facts from sources you allow, a writer turns them into a brief and a draft, and a reviewer removes anything without a source. The rep edits and sends. The system never sends on its own.

Book a Strategy CallSee the step trace

At a glance

Connects to
  • Your CRM
  • Allow-listed web sources
  • Public company registers
  • A news search API
  • Your email tool
  • OpenAI, Anthropic Claude or Google Gemini
Typical rollout
Week 1 → Weeks 5–6See the plan

The problem

Where the work gets stuck today

Good outreach starts with a reason to write now: a new product, a new market, a hiring push, a change of leadership. Finding that reason takes a stack of browser tabs per account, so busy reps skip it, and the message that goes out could have been sent to anyone.

Handing the job to a general AI writer fails differently. It writes fluent paragraphs about facts it cannot point to, mixes up companies with similar names and, once connected to an email tool, can send all of it at a scale nobody reviewed.

Where the time goes in account research
StepBy hand todayWith the assistant
Find what changedCompany site, news, filings and job posts, one tab at a timeThe researcher checks allow-listed sources and dates each fact
Check the CRMSearch past deals, notes and open ticketsRead through the CRM API and summarised with links
Pick the angleThe rep’s judgement, often skipped when busyThe writer proposes an angle tied to cited facts; the rep chooses
Write the messageA blank page, or a template with the name swappedA draft in your voice, every claim traceable
Log itOften forgottenThe approved brief saved to the CRM as an agent-drafted note

Task catalog

What the system takes on

  • Task 01

    Watch accounts for triggers

    For the accounts you choose, the researcher checks allow-listed sources on a schedule: the company site, press and careers pages, public filings and a news search. A dated change worth a conversation becomes a short alert to the account owner.

  • Task 02

    Build a sourced brief

    What the company does, what changed, likely priorities and open questions, with every factual line linked to its source and date. Conflicting facts are shown side by side, never merged.

  • Task 03

    Map people from your CRM

    Contacts and their roles come from your CRM and past conversations, not from scraped profiles. The brief notes who the team already knows and when they last spoke.

  • Task 04

    Draft the outreach

    The writer drafts an email or a message for the rep’s channel in your team’s voice, built on one or two cited facts and a clear reason for writing now. The rep sees which line rests on which source.

  • Task 05

    Review before the rep sees it

    A reviewer agent compares every claim with the retrieved sources, removes lines without one, flags guesses and checks the draft against your do-not-say list.

  • Task 06

    Save what the rep approves

    Approved briefs are written to the CRM, tagged as agent-drafted with their sources. Rejected drafts are kept with the rep’s reason, which becomes test material.

Step trace

One run, step by step

Sample multi-agent run: a planner splits the request, a researcher fetches allow-listed sources and records dated facts, a writer drafts a brief and an email, a reviewer removes an unsourced sentence, and the run stops at the rep’s approval gate. The approved email is saved as a draft; nothing is sent.
Sample data
account-research.runSample data
  1. NOTERepBrief and first-email draft for a sample account: a regional logistics software firm, call on Thursday.
  2. PLANPlannerFive questions: what they sell, what changed recently, hiring signals, CRM history, likely priority.
  3. TOOLResearcherAllow-listed fetches: company site, press page, public register, careers page, news search. Each fact stored with its URL and date.
  4. OKResearcherEleven facts recorded. Two sources disagree on a launch month; both kept with their dates.
  5. TOOLCRM toolRead-only lookup: one closed-lost deal last year, reason “timing”; two known contacts.
  6. PLANWriterAngle: the new warehouse product plus the earlier timing objection. Brief and a short email drafted.
  7. RETRYReviewerA sentence about revenue growth has no source: removed. A phrase on the do-not-say list: rewritten.
  8. OKReviewerEvery remaining factual line links to a source. The launch-month conflict is flagged for the rep.
  9. APPROVALAccount executiveApproval gate: the rep reads the brief beside its sources and edits the email.
  10. APPROVEDAccount executiveBrief approved and saved to the CRM as agent-drafted. The email lands in the rep’s drafts folder.
  11. DONESystemNothing sent. Sending stays a manual step in the rep’s own email tool.

Sourced brief

A brief where every line has a source

A shortened sample brief for an invented account. Each fact carries its source and when it was published; the reviewer’s removals stay visible, so the rep knows what was dropped and why.
brief · sample-accountSample data

Sample account: a regional logistics software firm

What changed

  • Launched a warehouse scheduling product for mid-size depots.

    CitedSource: Company press pageLast month

  • Hiring implementation staff in two new regions.

    CitedSource: Company careers pageThis week

  • The launch month differs between the press page and a news article.

    ConflictSource: Press page and news searchBoth dates shown

What we already know

  • Closed-lost last year, reason given as timing.

    CitedSource: Your CRM, deal recordLast year

  • Two known contacts: an operations lead and a finance lead.

    CitedSource: Your CRM, contactsCurrent

Removed by the reviewer

  • “Revenue grew strongly this year.”

    RemovedSource: No source foundNot shown to the account

Rules every brief follows

  • No factual line without a retrieved source. Suggested angles are labelled as suggestions.
  • Sources come from an allow-list you approve; sites whose terms forbid automated access stay off it.
  • Facts describe the organisation. People and roles come from your CRM, not from personal profiles.
  • Fetched text is treated as data, so a web page cannot instruct the agents.
  • Drafts never leave on their own. Sending is a person’s action in their own tool.
  • Each run has a cap on steps, tool calls and model spend, logged with the brief.

Systems involved

What it connects to

Your existing tools stay the system of record; the AI reads and writes through their APIs.
Your CRM
Accounts, contacts, deals and notes, read through its API. Writes are limited to notes a rep approved.
Allow-listed web sources
The account’s own site, press and careers pages, fetched with rate limits and timeouts.
Public company registers
Filings and officer records, where your markets publish them.
A news search API
Recent coverage with publication dates, under a licence that covers automated use.
Your email tool
Receives approved drafts in the rep’s drafts folder where its API supports drafts. It is never given a send permission.
OpenAI, Anthropic Claude or Google Gemini
The planner, researcher, writer and reviewer models, on accounts in your name.

Guardrails

  • No send permission. The agents can create drafts but never send email or messages, and the API scopes they hold enforce it.
  • Every factual line needs a retrieved source and date, or the reviewer removes it.
  • Research tools are read-only and limited to the allow-list. The agents cannot browse freely.
  • Text fetched from the web is handled as quoted data, never as instructions.
  • No bought contact lists and no scraping of personal profiles. People come from your CRM.
  • A per-run ceiling on steps, tool calls and model spend stops a run that loops.
  • A do-not-say list covering claims, competitors and pricing is checked on every draft.

Human checkpoint

The account owner is the only one who can approve a brief, save a note to the CRM or send a message. The approval screen shows the brief beside its sources, the lines the reviewer removed and any conflicts, with the draft underneath. Edits and rejections are kept with the rep’s reason, and a sales lead reviews a sample each week to tune the brief template and the do-not-say list.

What we plan for

Typical targets

Typical targets are planning ranges we use when scoping, each with its basis. They are not results from a client project; your own baseline decides the real number.

  • Typical target

    40–70% less

    Rep research time per account

    Basis: The same reps researching the same accounts by hand, timed before the pilot, against the time they spend reviewing the brief.

  • Typical target

    All of them, by design

    Factual brief lines with a source

    Basis: The reviewer removes any factual line without a retrieved source before the rep sees the brief

  • Typical target

    None, by design

    Messages sent without a person

    Basis: The agents hold draft-only permissions; sending happens in the rep’s own tool

Rollout

Week by week

Typical ranges for a first release; the scoping call fixes the plan for your systems.

  1. Week 1

    Brief template agreed with a few reps and the sales lead; source allow-list and do-not-say list written; read-only CRM access.

  2. Weeks 2–3

    Researcher, writer and reviewer built with run caps and logs. Evaluation set built from accounts the team already researched by hand.

  3. Week 4

    Side-by-side pilot with a few reps on live accounts, drafts reviewed daily, template and tone tuned.

  4. Weeks 5–6

    Trigger alerts for a watch list, CRM notes after approval, drafts into reps’ email tools where supported, and a weekly quality review handed over.

FAQ

Frequently asked questions

Why use three agents instead of one?

Splitting the work makes each step checkable. The researcher can only fetch and record facts, the writer can only use what was recorded, and the reviewer has one job: compare every claim with its source. One model doing all three tends to fill gaps with plausible guesses, and you cannot tell where.

Can the drafts sound like our team rather than like AI?

Largely, yes. The writer’s instructions are built from messages your best reps have actually sent, plus your do-not-say list. Reps still edit most drafts: the point is to start from cited facts and a sensible angle, not to replace the rep’s voice.

Which research sources are switched on by default?

None until you approve them. A typical list is the account’s own website, its press and careers pages, public company registers in your markets and a licensed news search. We check each source’s terms for automated access and leave out anything that forbids it.

Can it keep an eye on a large list of target accounts?

Yes, as a scheduled watch list rather than on demand. Trigger checks run in the background within source rate limits and run caps, and only an account with a dated change worth a conversation produces an alert, so reps are not buried in briefs nobody asked for.

Does it plug into our sales engagement tool?

It can hand approved drafts to any tool whose API accepts drafts or tasks, so reps send from the tool they already use. We do not connect the agents to a sequence’s send step, because that would turn one approved draft into an automated campaign.

What if a brief gets a fact wrong?

The rep sees the source beside every line, so a wrong fact is usually caught at a glance. A rejected line is logged with the reason, added to the evaluation set and traced to the step that failed, whether that is a source, an extraction rule or the reviewer’s check.

Who owns the briefs and the research data?

You do. Briefs, notes and run logs live in your CRM and on accounts registered to your business. The proposal lists each model provider’s data-use terms, so your data protection lead can sign them off before anything runs.

Next step

Have a workflow in mind? Let’s scope it together.

Book a free 30-minute strategy call. Bring one process you want to automate and we will talk through scope, integrations and a sensible first step.

Book a Strategy CallGet an AI Plan

Free 30-minute call · no commitment

Send an RFP or security questionnaire