Concept buildA worked example of how we would build this for a typical business. Not client work: no client, no client data, no achieved results.
AI Agents
Concept buildA concept build for a research agent that prepares a cited company brief before a sales call, so reps start from checked facts, not a pile of browser tabs. It plans its own research, works only through allow-listed tools, links every line to a source, and drafts a CRM note that a rep approves before anything is saved.
The brief
A typical UK sales team at a business services firm researches every target account by hand before the first call. Reps read the company website, the latest filings on the public register, recent news and the account’s history in the CRM, then paste notes into a document nobody else can find. The work is slow, uneven between reps and easy to skip when the calendar is full, so first calls often open with questions the prospect has already answered in public. Notes rarely make it back into the CRM in a consistent shape, so the next rep repeats the research. A data subscription does not fix it on its own: the facts still have to be read, checked and turned into a point of view for this particular call.
Explainer
Explainer · 7 slides · about 46 s
Concept buildConcept build: Pre-call Account Briefs for a UK B2B Sales Team
Concept build, explained in 7 slides: a concept build for a research agent that prepares a cited company brief before a sales call, so reps start from checked facts, not a pile of browser tabs.
Slide 1 of 7
01 · AI agents
Pre-call Account Briefs for a UK B2B Sales Team
Concept build: Pre-call Account Briefs for a UK B2B Sales Team.
02 · The problem
What a typical sales team at a business services firm is up against
A typical UK sales team at a business services firm researches every target account by hand before the first call.
03 · The approach
How we would build it, in 5 steps
We would build it in 5 steps, from brief template and source allow-list to evaluation set and a side-by-side pilot.
04 · Architecture
How the pieces connect
How the parts connect, from the first input to the person who reviews the result.
05 · Demo · sample run
A cited brief before a first call
A sample run, step by step, including where a person steps in.
06 · Typical targets
What we would measure it against
Typical targets, each with the basis it would be measured against.
07 · Next step
Scope a build like this for your business
Book a strategy call to scope a build like this.
Architecture
Architecture of the Pre-call Account Briefs for a UK B2B Sales Team concept: Rep request → Planner model (task); Planner model → Research tools (tool calls); Allow-listed sites → Research tools (fetch); Public register → Research tools (lookup); Research tools → Citation check (facts + links); Citation check → Rep approval (cited brief); Rep approval → CRM via its API (approved note).
Diagram with 8 parts: Rep request (source), Allow-listed sites (source), Public register (source), Research tools (processing), Citation check (processing), Planner model (model), Rep approval (human), CRM via its API (system of record).
Connections:
Sample run
Rep asks for a briefRoutine
Sample account: a mid-size UK logistics software firm, first call on Thursday.
Planner splits the taskRoutine
Five questions: what they sell, recent changes, filings, news and CRM history.
Allow-listed tools runRoutine
Company site, public register and news search: 14 tool calls, inside the cap.
Citation checkNeeds care
Two lines with no source removed; one conflict flagged with both dates.
Brief shown to the repRoutine
Eight cited facts, three likely priorities and four questions for the call.
Rep approves the CRM noteA person takes over
One line edited, then saved as agent-drafted with its source list.
Build plan
We sit with two or three reps and the sales lead to define what a good brief contains: what the company does, what has changed recently, likely priorities, open questions and what the CRM already knows. In parallel we agree the source allow-list: the company’s own site, the public companies register, a news search API and the CRM. Sites whose terms forbid automated access stay off the list.
Each source becomes a narrow, read-only tool with its own rate limit and timeout. The agent cannot browse freely, send email or write anywhere except a draft note. A per-run cap on steps, tool calls and model spend stops a run that loops, and every call is logged with its inputs and outputs so a run can be replayed when something looks wrong.
A planner model breaks the request into questions, picks tools and records each fact with the URL or register entry it came from. Text fetched from the web is treated as data, never as instructions, so a page that tries to steer the agent is ignored. Before the brief is shown, a citation check removes any line that has no retrieved source behind it.
The rep sees the brief beside its sources, edits it and approves the CRM note on one screen. Only approved notes are written to the CRM through its API, tagged as agent-drafted with the date and the source list. Rejected drafts are kept with the rep’s reason, and those reasons become test material for the next round of tuning.
We build an evaluation set from accounts the team has already researched by hand and score the agent’s briefs on accuracy, source coverage and usefulness for the call. A pilot with a few reps then runs alongside their normal process for two to three weeks, and both paths are timed on the same accounts before use widens.
Typical targets
Typical target
A reviewable, cited account brief in under 5 minutes of agent run time
Basis: Timed on the pilot’s test accounts; step and tool-call caps bound how long a run can take
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
Every factual line in a brief links to its source
Basis: Design property: the citation check removes lines without a retrieved source before display
Typical target
No CRM change without a rep’s approval
Basis: Design property: the CRM write sits behind the approval screen; the agent’s own access is read-only
Your project
Every action the agent may take, with its read or write scope and which ones need a person’s approval first.
Each run records the plan, every tool call with its inputs and outputs, and the approval decisions.
Real examples from your backlog scored before every prompt, tool or model change goes live.
Step and cost ceilings per run, so an agent that loops or meets bad data stops and asks instead of carrying on.
Where this fits
FAQ
No. Nobody runs this agent today; it is a concept build that sets out how we would build it for a typical UK business services firm. The account, tool calls and brief in the demo are invented to show the flow. A real engagement would start by timing the team’s current research on its own accounts, which is where the typical targets on this page get confirmed or revised.
The design assumes some pages will try. Fetched text reaches the model as quoted data rather than instructions, every research tool is read-only, and the only thing the agent can write is a draft that a rep has to approve. A page that says “ignore your instructions” can at worst spoil one draft, and both the citation check and the rep see that draft before it goes anywhere.
It researches the organisation, not individuals. Named contacts come only from your CRM, personal social profiles are not on the source list, and briefs keep business-role facts only. Your data protection lead still owns the legitimate-interests assessment under UK GDPR; keeping the data this narrow is what makes that assessment short.
Conflicting facts are shown side by side with both sources and their publication dates, and the brief marks the point as unresolved instead of picking one. When the allow-listed sources return too little, the agent says so and lists what it checked, rather than filling the brief with general statements about the industry.
Each run has a ceiling on steps, tool calls and model tokens, agreed at scoping and logged per brief. Summarising a fetched page goes to a lighter model, while the planner’s larger model is used only to decide what to look up next. During the pilot the cost per brief is reported every week, so the ceiling is set from real runs.
Worked examples
Concept builds are worked examples, not client work.
Concept buildAI Agents
A concept build showing how we would build an AI agent that qualifies property leads on WhatsApp and website chat at any hour. It would ask about budget, area and timing, score each lead against the agency’s own rubric and book viewings for qualified buyers, so agents spend their time with people who are ready to view.
Concept buildAI Agents
A concept build for a US online marketplace where sellers upload products with thin titles and half-empty attributes. An agent reads each new listing and the supplier’s spec sheet, fills category attributes and a description from that data only, validates the result against the category schema and sends it to a merchandiser for approval. Nothing publishes without a person.
Concept buildAI Agents
A concept build for the purchasing team of a UAE trading and distribution company that chases overseas suppliers by hand for order confirmations, ship dates and shipping documents. Four agents split the work: a planner picks the orders that need chasing, a researcher gathers the facts, a writer drafts one email per supplier and a reviewer checks every figure against the sources. A buyer approves each email before it goes, so nothing reaches a supplier on its own.
Next step
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.
Free 30-minute call · no commitment
