A company memory the AI works from, a brief on my desk every weekday morning, a Friday review, a research radar and an outreach system that find the next client, a portal that turns client emails into briefs, a brand system that checks every draft and a pipeline that produces the finished work. Together they let a small senior team work at the pace and accuracy of a much larger one.
The operating brain · 90 seconds · Figures from Pivitt's own records
Median response to a brief, June to July 2026.
Of tracked briefs approved first time, with no amends. 8 of 11, same period.
Client conversations captured as briefs, up from 14 of 36 before the capture rebuild.
Mistakes caught before they reached a client, in the system's first five days.
Morning briefs composed from the memory, emailed and filed, 20 July to 16 October 2026.
Standing rules, each written from a real correction and applied to every output after it.
Commercial
Invoiced fees per month in the first two months after the systems went live, against the three months before.
Committed revenue for the coming quarter, October 2026 against the position on 21 July.
Revenue in view, committed revenue and open pipeline together, October 2026 against 21 July. Nearly three times.
From the outreach system's first sends to the conversation that became a shortlisted brand tender. August 2026.
Figures from Pivitt's own records, June to October 2026. The memory sits at the centre: every task starts from it and every correction is written back to it.
The memory, and the rhythm that keeps it current without anyone having to remember to.
3 systemsFinds the right companies at the right moment and writes to them with something true.
4 systemsTakes a request from the inbox to finished, checked, approved work.
5 systemsHow the business gets found, and how nothing slips through.
2 systemsOne · Runs the business · 1 of 3
Every task starts from what the business already knows, and every correction is kept.
Every decision, price, approved claim, piece of client history and correction in one written record. The AI reads it before every task and writes back what changed, and every change is logged with what and why.
96 standing rules · 2,100+ logged updates since July
Swipe to see the whole model
One · Runs the business · 2 of 3
The business reviews itself on a schedule, whether or not anyone remembers to.
A morning read of the calendar, inboxes and call transcripts, a one-page brief at 07:20, a research radar twice a week, a Friday review and a monthly distil, all writing back to the memory.
62 automated morning briefs sent
Swipe to see the whole model
One · Runs the business · 3 of 3
Problems show up the same day, and anyone can pick the business up.
An activity log of everything that happens in the portal, a private metrics dashboard, a filing check that raises the alarm, and a runbook any operator or AI tool can start from.
1,110 portal events logged, backfilled to May
Swipe to see the whole model
Two · Wins the work · 1 of 4
The next opportunity is found the week it is published.
Public tender feeds, ten named accounts and sector sweeps read twice a week and ranked. Tenders are marked official record; press is marked unverified.
2,821 tender notices read in a single run
Swipe to see the whole model
Two · Wins the work · 2 of 4
Every target gets a sequence written to them, and nothing is forgotten.
Each account gets a dossier: the trigger, the problem it points to, the proof that answers it. Then a five-touch sequence drafted in advance. I approve and send every message myself.
First week of sending: a conversation that became a shortlisted brand tender
Swipe to see the whole model
Two · Wins the work · 3 of 4
Every pitch is personal, and I know the moment it is read.
Immersive pages from one shared builder, carrying each prospect's name, note, sector and closest proof, with an alert the minute a named reader opens one.
156 builds, versions 7 to 40
Swipe to see the whole model
Two · Wins the work · 4 of 4
Proposals stay private, measured and under control.
Every proposal is a private page with a consent gate, an access window, read tracking and a switch to withdraw it or reopen it on request.
Swipe to see the whole model
Three · Delivers the work · 1 of 5
Nothing a client sends gets lost, and every approval is on record.
Client emails become brief cards on their own, with the full thread, their own words and every file. Replies land on the right card, amends are recognised, and approvals are kept permanently.
36 of 36 conversations captured · 1 hour median response, June to July · 28 of 28 tables secured
Swipe to see the whole model
Three · Delivers the work · 2 of 5
Mistakes are stopped before a client ever sees them.
Every draft passes four checks, voice, approved claims, banned words, names and figures, and comes out ready, ready after fixes, or held.
15 mistakes stopped in the first five days · 73% approved first time
Swipe to see the whole model
Three · Delivers the work · 3 of 5
The brand travels into every AI tool the team uses.
The voice and the check packaged as two skills with one rules file, so any AI tool drafts on brand and checks its own work.
22 banned words · 29 approved figures
Swipe to see the whole model
Three · Delivers the work · 4 of 5
Every figure and name can be traced to its source.
Each claim held with its exact wording and source. If it is not on the register, it does not ship.
Swipe to see the whole model
Three · Delivers the work · 5 of 5
Finished work, produced to standard and checked before it travels.
Decks, reports and web pages from one source, checked for fonts, figures, page breaks and claims.
A 34-page tender bid built in four days
Swipe to see the whole model
Four · Brand, visibility and control · 1 of 2
Found by search engines, by AI assistants and by the right readers.
Site-wide search fixes and structured data, 20 guides with drawn models, and seven interactive tools that email the reader their result.
4,716 internal links rewritten · 26 five-star Google reviews
Swipe to see the whole model
Four · Brand, visibility and control · 2 of 2
Money left on the table gets found.
A spend log matched against every invoice raised, so work that was never billed shows up.
£484.81 of unbilled print found
Swipe to see the whole model
The workspace shown is a fictional demo client. Real client workspaces are private.
Briefs scattered across inboxes, half the files missing.
Every brief captured with the full thread and every file, in the client's own words.
House rules live in people's heads and leave when they do.
96 written rules, applied to every output by default.
Figures and client names drift between documents.
One claims register, exact wording, checked before anything is sent.
Every document built by hand, every check done by eye.
Produced from one source and checked automatically before it travels.
Context lost between conversations and handovers.
A written record the work starts from and writes back to.
Mondays start with an inbox and a memory of last week.
Every weekday starts with a one-page brief and one priority.
Tenders and buying signals found by chance, if at all.
Read twice a week, ranked, and filed with their source.
New business written one message at a time, follow-ups forgotten.
Every target researched, every touch drafted in advance, every reply tracked.
Your voice, claims, rules and checks written as instructions, so every AI-drafted proposal, deck and email comes out on brand.
Decisions, prices, client history and standing rules in one record your team and its AI tools work from.
Requests that arrive by email become tracked briefs with every file, through to approval and delivery.
Reports, proposals and decks generated to your standard from one source, checked before they leave.
Your inboxes, calendar and records read every morning into one page and one priority, and a Friday review of what actually moved.
Tender and trigger research, written sequences and reply tracking, so new business runs every week rather than when there is time.
Find where your team loses time and where the same mistake keeps happening.
Write your rules down and build the system around the work you already do.
Use it on live work, measure it from day one, and keep it current as the business changes.
A set of tools and written rules that let AI do part of a team's work to the business's own standard: a record of what the business knows, the rules its work has to follow, checks on every draft, and the routines that keep all of it current.
Three stages. Map where the team loses time and where the same mistake keeps happening. Build the rules and the system around the work the team already does. Run it on live work, measure it from day one and keep it current as the business changes.
A one-hour median response to a brief and 73% of tracked briefs approved first time, June to July 2026, 36 of 36 client conversations captured as briefs, and 15 mistakes caught before they reached a client in the system's first five days.
Both. The work starts as consultancy, mapping where the team loses time and what to build first, and carries on into building the systems, running them on live work and measuring what they change.
Yes. Each system is rebuilt around your rules, your tools and the way your team works, and scoped and priced before work starts. The first step is a conversation about where your team loses time.