The starting point. A short, paid engagement that produces a practical audit, a risk map, a policy starter pack, and a prioritised roadmap. Two weeks, fixed scope, fixed price.

We walk through how the business actually runs, where the time goes, where the rules are clear enough for AI to help, and where they are not. You leave with a document you can act on, with or without us.

What we look at

• The workflows and tools the business already runs.
• Where AI could realistically help, and where it would create risk.
• Data, access, and the boundaries that need to be set.
• A prioritised, plain-English roadmap for the next six to twelve months.

What you keep

A written readiness review, a prioritised list of workflows worth automating, a one-page AI policy starter, and our honest take on which jobs to leave alone for now.

One workflow at a time, built with approved tools and human approval points. We start with whichever job from the readiness review will pay back fastest: typically quoting, inbox triage, customer follow-ups, or a monthly report.

Each build follows the same shape: scope, design, build, pilot with a human in the loop, document, then hand over. You see the work each step of the way.

Typical first build

• Quoting assistant that drafts from your existing templates.
• Email triage that sorts, summarises, and surfaces what needs a human reply.
• Document assistant that answers staff questions from your own knowledge.
• Monthly client report drafted in your tone, ready for a partner to sign off.

What you keep

The accounts, the prompts, the integration credentials, the operating documentation, and the right to walk away.

The pantry of the operation. A searchable, AI-assisted knowledge base built from the documents, procedures, and policies the business already runs on. Staff ask it questions in plain English; it answers from your own material, with sources.

What goes in

Procedure manuals, contract templates, SOPs, briefing notes, frequently asked questions, supplier handbooks, training material. Whatever the business reaches for when it needs to remember how it does things.

What it does

Answers staff questions from approved material, surfaces the source on every answer, refuses to guess when the answer is not in the larder, and tells you what is being asked so you can fill the gaps.

The boring, load-bearing part. A short, plain-English AI policy your team will actually read, an approved-tools list, an AI register that records what is running and who owns it, and human-review steps for the jobs that need them.

What we produce

• An AI policy in plain English, sized for a small business.
• A list of approved tools, with a clear path for adding new ones.
• An AI register: each workflow, its owner, its boundaries, and its review cadence.
• Templates for human-in-the-loop sign-off where it matters.

Why it matters

Most AI failures in small business are governance failures, not model failures. A workflow without an owner goes stale. A tool without rules ends up holding data it should not. We write the rules so the rest of the system can be trusted.

When a single-step automation is not enough, an agent can plan and execute a multi-step job: read the enquiry, look up the customer, draft the quote, file the record, raise the follow-up. Useful, but only when the operating rules are firm and the audit trail is real.

Where agents fit

• Multi-step workflows that span more than one system.
• Customer-facing assistants with clear escalation rules.
• Voice agents for after-hours triage with strict handoff to a person.
• Internal copilots that draft, never send.

What we insist on

Permissions are scoped. Sensitive actions need human sign-off. Every step is logged. There is an off switch the owner can reach. Nothing runs without an owner reviewing the logs monthly.

The part nobody costs in. Once a workflow is live, someone has to watch it, tune it, fix what breaks, retire what is not earning its place, and write up what happened. Larder does that, monthly, on a fixed monthly fee.

What we do every month

• Monitor every active workflow, every failure, every exception.
• Tune prompts, rules, and data where the workflow has drifted.
• Ship one small improvement, agreed at the previous review.
• Write a plain-English report covering volume, value, errors, and what is next.

What you do

Read the report. Approve the next improvement. Tell us if priorities have shifted. That is it.