What We Build

The technology works. What's missing
is someone who owns it.

We build AI systems for companies between 50 and 500 people — assistants, agents, knowledge bases, quoting engines, and the software that doesn't exist yet. Then we stay on the hook when the rules change underneath them.

~70%of companies your size haven't started with AI at all
42%of those who started abandoned it before production
90 secwhat we cut a client's 60-minute quoting process to
7role-built assistants delivered to one leadership team
Why most of this fails

Nobody fails at AI because the models aren't good enough.

They fail because the thing got built and then nobody owned it. The person who championed it moved on. The integration broke quietly. The model underneath was retired on sixty days' notice. Half the team went back to the spreadsheet in week three.

Broken automations don't file complaints. They just stop, and you find out in March when someone asks why they haven't seen the Monday report since January. That gap — between something being built and something still being right — is the whole of what we do.

Start here

A free consultation, and every guide

Thirty minutes with Keegan on where you are with AI, where you want to go, and the shortest route there. Book it and everything below unlocks.

One-Page Overview

PDF · 1 page

Everything below, on a single page. The version to forward.

Full Capabilities Document

PDF · 13 pages

Every offering in detail — how each one works, what it connects to, and what we've delivered.

AI 101 — The Companion Guide

PDF · 11 pages

Where to start, how to make it personal, and how to unlock what AI can actually do for your business. The companion guide to the AI 101 episode of Affiliate Nerd Out.

AI 201 — Partner Scout Build Kit

ZIP · Guides + templates

Build the partner-scouting agent from the AI 201 episode of Affiliate Nerd Out — the full build guide, brand brief and ideal partner profile templates, a completely worked example, and the ready-to-install agent.

Thirty minutes with Keegan on where you are and where to go next. Every guide unlocks too.
01

Team AI Setup

Give every person an assistant that knows their actual job

We pay for the licenses. Two people love it. Everyone else opened it twice.

What we build

A role-tailored assistant for each key person, built inside whatever AI platform you already license. Each ships with a documented set of skills for that role, guardrails on what it should and shouldn't touch, and an operations guide so the person knows what to do with it on Monday morning.

How it works

  1. Discovery per role — 60 to 90 minutes mapping the recurring work: what eats their week, what they avoid because it's tedious.
  2. We build the skills — Named commands that run a defined workflow, instead of hoping someone writes a good prompt.
  3. Connect their systems — Calendar, inbox, files, CRM.
  4. Deliver the documentation — So the setup lives in your company, not in a consultant's head.
  5. Sit with each person until they use it — The step that decides whether any of this survives past month two.

How it fits what you run

Works in Claude, ChatGPT or Microsoft Copilot — whichever you already pay for. We don't ask you to switch. Connects to Google Workspace or Microsoft 365, Slack, your CRM and your file storage through standard connectors.

In practice

A seven-person leadership team at an experiential marketing agency. Executive assistant, business development, marketing, operations — each got their own. Every build shipped the same seven-document package, so the fourth took a fraction of the time the first did.

Typical timeline — 2 to 3 weeks for a team of 5 to 8

02

Autonomous Agents

Work that happens on its own, so a lean team covers more ground

It's not hard work. It's relentless — and it's the first thing to slip when we're stretched.

What we build

Agents that run on a schedule or a trigger, do the work, and report back into the channel your team already lives in. Nobody logs into a new tool.

How it works

  1. Trigger — A schedule, an inbound message, a form, a webhook.
  2. Gather — Pull the relevant data from your systems.
  3. Interpret — An AI step reads it and works out what it means.
  4. Act — Write the result back into the system of record.
  5. Report — Post a summary where your team will actually see it.

How it fits what you run

Slack, Teams, Discord, Gmail or Outlook, Notion, Google Sheets, your CRM, QuickBooks — and several hundred other tools through our integration layer. Runs on your infrastructure or ours.

In practice

A business development agent that watches a Slack channel and reads it the way a person would. Someone types “grabbed coffee with Sarah at TestCo, looking at around 2k a month” and it creates the opportunity with the contact, value and date. It nudges three times a week on anything gone quiet and posts a full pipeline summary every Monday.

Straight answer

We keep the chains short and put people at the decision points. Independent benchmarks show agent accuracy falls from roughly 58% on a single step to about 35% once a task runs many steps deep — so we don't build twenty-step autonomous chains. This supplements coordination and back-office load; it doesn't replace judgment.

Typical timeline — 2 to 4 weeks for the first, faster after

03

Knowledge Bases

Stop the hallucinations, and give your AI what your company actually knows

It makes things up, because it doesn't know anything about us.

What we build

A single grounded knowledge layer your AI reads from — so answers come from your documents, with citations you can click, instead of from the model's imagination. Your real knowledge is currently spread across Drive, old email threads, three pages nobody's updated since February, and one person's head.

How it works

  1. Ingest — Documents, SOPs, past proposals, contracts, meeting notes, product information.
  2. Index — Processed for semantic search, so it finds things by meaning rather than exact keyword.
  3. Connect — Exposed to your AI tools through one standard connection, so every assistant queries the same source of truth.
  4. Ground and cite — Answers come back with links to the source. If it can't find support, it says so instead of inventing something.
  5. Keep it fresh — New material flows in automatically, so it doesn't go stale in a month.

How it fits what you run

We deploy it inside your existing Google or Microsoft tenant. No new vendor takes custody of your data, and the agreements you already hold cover it. Connects to Drive, SharePoint, Notion, Confluence and email.

In practice

We built a knowledge platform mirroring a complete operational vault — projects, decisions, client history, standard procedures — reachable through one shared connection from a laptop, a browser or a phone, all hitting the same source of truth.

Straight answer

We won't sell you self-hosted AI on your own hardware for this. It's expensive, and the privacy benefit is mostly illusory unless you go fully local — which at your size costs more than it protects.

Typical timeline — 3 to 4 weeks

04

Marketing & Sales

Your brand voice, your decks, your pipeline — faster and more consistent

Everything written by anyone other than me sounds slightly off-brand.

What we build

A documented brand voice the AI genuinely follows, plus a set of generators your team runs on demand — decks, proposals, content, and researched, scored lead lists.

How it works

  1. Brand voice first — We codify how your company sounds: tone rules, vocabulary you use and avoid, and side-by-side examples of on-brand versus off. This is what stops AI output from reading like AI output.
  2. Decks and pitches — Structured input, on-brand slides out, in your template.
  3. Proposals — Pulls from your service catalogue and pricing, assembles the document.
  4. Content and social — Drafted in your voice, using your real examples rather than generic filler.
  5. Lead research and scoring — We turn your ideal customer profile into an explicit scoring rubric, then the system finds matching organisations, scores each, and hands your team a ranked list with the angle to use.

How it fits what you run

Outputs into Google Slides or PowerPoint, Docs or Word, your CRM and your email platform — in your existing templates. Your salespeople keep working where they already work.

In practice

For a healthcare client we built lead research and scoring across three completely different buyer types — clinics, schools and youth sports organisations — each with its own definition of good fit, scored against a 100-point rubric so the team knew who to call first and why.

Straight answer

We don't do fully automated cold outbound. Deliverability collapses, quality craters, and it damages the brand you just paid to define. We make your team dramatically faster — we don't replace them with a spam cannon.

Typical timeline — 3 to 4 weeks

05

Quote & Document Engines

Ninety-second quotes — and still correct six months from now

Every quote takes an hour by hand. And I'm not certain the numbers are still right.

What we build

An engine that produces the finished document in about ninety seconds from a short intake — plus monitoring that tells you when the rules underneath it change. Because the numbers depend on things that move without telling you: pricing, rebates, tax credits, code requirements, rate tables.

How it works

  1. Intake — A simple form, or your rep's raw notes; the system structures it.
  2. Your rules engine — Pricing logic, eligibility, options, exclusions — the things your experienced people know and everyone else gets wrong.
  3. Live external data — Anything that changes gets pulled at generation time rather than hardcoded.
  4. Generate and deliver — The finished document in your template, emailed to the customer or dropped in your drive, ready for signature.
  5. Monitor — We watch the external rules and flag you when something material shifts.

How it fits what you run

Connects to your CRM, e-signature tool, Drive or SharePoint, and your email platform. The customer sees your document in your branding — the only difference is the speed.

In practice

A home-electrification company. Proposals went from 30–60 minutes of manual assembly to about ninety seconds, pulling live incentive data at generation time. In the twelve months that followed, five separate federal energy provisions changed on a published schedule — and because the engine reads that data live rather than hardcoding it, the proposals stayed correct without a single code change.

Typical timeline — 4 to 6 weeks

06

Connected Systems

Your tools already work. Getting them to talk is the problem.

Somebody rebuilds the same report by hand every Monday, and the numbers never quite match.

What we build

The connective layer between your systems, plus the reporting that finally comes out of it automatically. This is the most common problem we see — and among finance leaders at companies your size, 77% say they're integrating what they already own before buying anything new.

How it works

  1. Map — What lives where, what's the source of truth for each thing, where the duplicates are.
  2. Connect — Scheduled or real-time pulls from each system.
  3. Reconcile — Normalise the mismatches: different customer IDs, different date formats, the same company spelled four ways.
  4. Report — What took someone half a Monday, generated and delivered on schedule.
  5. Alert — Thresholds that matter get pushed to you, instead of waiting to be noticed.

How it fits what you run

QuickBooks, your CRM, project management tools, spreadsheets, databases, e-commerce, payroll, scheduling. If it has an API — and nearly everything does now — we can connect it. Where something genuinely can't be, we'll tell you before you're invested.

In practice

We run our own operations this way: CRM, workflow automation, AI observability and uptime monitoring on one server, all connected, all reporting into one place. We built it for ourselves first, because we weren't going to sell something we hadn't lived with.

Typical timeline — 3 to 6 weeks depending on system count

07

Custom Platforms

When the software for how you work doesn't exist

We've evaluated everything. It's all close but wrong — so the real work happens in spreadsheets.

What we build

A real application, built for your actual process, that you own. Usually the strongest move isn't replacing everything — it's building the specific piece nothing covers, and connecting it to the tools that are working fine.

How it works

  1. Specification first — Epics, user stories, acceptance criteria, data model. You read it and approve it. Changes are cheap at this stage and expensive later.
  2. Phased build — Working software at the end of each phase, not a big reveal at the end.
  3. You own it — The code, the data, all of it.
  4. We keep it running — Hosting, updates, and the maintenance owned software genuinely requires.

How it fits what you run

It integrates with what you already have rather than replacing it wholesale. Your data stays yours, and it's hosted where you want it.

In practice

A membership platform for a professional association. An operations platform for an international nonprofit. A three-application suite — public site, subscriber portal and admin console — for a financial intelligence firm, all sharing one design system. A full management platform for golf simulator facilities, including payments and scheduling.

Typical timeline — 8 to 16 weeks for phase one

Where this usually starts

Most clients don't start with a build.

AI Rescue

“We already tried this and it doesn't work.”

A half-finished project, an agent that hallucinates, an automation nobody trusts. We audit it and tell you what's salvageable. Every finding is verified by a person before it reaches you — we're not going to hand you a pile of AI-generated false positives and call it a report.

AI Readiness Assessment

“We know we should be doing something. We don't know what.”

A structured review across six layers — where your data lives, what your tools can reach, how work actually flows, and where AI genuinely helps versus where it's a distraction. You get a written report with scored findings and a prioritised roadmap, useful whether or not you hire us for the build.

After launch

This is where most AI projects die, so it's worth being explicit.

Things break silently.AI models get retired on sixty days' notice — that happened six times in the last year. Integrations change. External rules shift. A broken automation doesn't raise its hand. So we monitor, we alert, and we send you a report showing what we caught.

People drift back to the old way.The hard part isn't the technology, it's adoption. The companies that succeed aren't the ones with better tools — they're the ones where somebody kept driving usage after launch week. We stay in that role: check-ins, refreshers when people join or change roles, and new skills as the work changes.

Start here

Tell us what's breaking. We'll tell you whether we can help.

No deck, no discovery theatre. A conversation about the specific thing that isn't working, and a straight answer about whether it's worth building.

Start a conversation

Client names are withheld throughout — all referenced engagements are active. References available under NDA.