AI for business processes

Fix the job that eats the week.

The weekly report. The meeting follow-up. The replies waiting in the inbox. The handover that depends on someone remembering. We take one repeating job, learn how your team does it, and build AI around it, so the work comes back prepared and a person decides what goes out.

For SMEs, charities and not-for-profits. A free 40-minute call by phone or Teams. No obligation.

The approach

Not another AI tool to manage.

Most AI projects start with a tool and go looking for a use. We start with the work.

  • One job, not a platform. We pick the repeating task that costs you most and make it reliable before touching the next one.
  • Your information, not guesses. The system works from your own files, records and examples, and only sees what it's allowed to see.
  • Your voice, ready to check. Drafts come back in your tone and to your standard, not as generic text you have to start again.
  • A person approves. Anything that sends, changes or can't be undone waits for someone at your organisation to say yes.

Typical starting points

Jobs we typically start with.

BeforeAfter
Monthly figures pulled by hand from three systemsA report that builds itself and arrives for review
Enquiries waiting until someone is freeA first reply drafted and ready to approve
Meeting notes that never become actionsFollow-ups written, owners named, sent once approved
Staff hunting through folders for the right policyAn answer from your own files, showing where it came from (see Ask Your Data)
Handovers that live in someone's headA written workflow with triggers, owners and checkpoints
Shared inbox triage every morningMessages sorted, routed and drafted, with exceptions flagged for a person

A staff AI assistant is a common first job: your knowledge, your voice, and a first draft your team can work with. We shape it around one role's repeating work — report drafts, meeting notes, policy lookup or enquiry replies — using approved sources, templates and a clear review rule. The assistant prepares the work; people still approve, adapt and decide.

How it works

Map the job, then build around it.

For the people doing the work

  1. Free call. A free 40-minute call by phone or Teams. You tell us the job that keeps slipping. We tell you honestly whether AI can help.
  2. Map the job. We look at how it really happens: who does what, which files and systems are used, where it gets stuck and what “finished” looks like.
  3. Build the first version. A working system on the real job, not a slide deck. Your team tries it with a person checking every output.
  4. Make it reliable, then add the next step. When the first job runs smoothly, we connect the next one. A new enquiry is logged, a reply is drafted, someone approves it.

For your IT team or technical lead

  • Workflow first. Triggers, decisions, owners, exceptions and review points are documented before anything is automated.
  • Least privilege. Each assistant or agent can reach only the systems and fields its job needs. Limits are enforced by permissions, not just instructions.
  • Checked, not trusted. Every automated step has a written pass-or-fail check and an approval point matched to how hard the step is to undo.
  • Model choice. Local or fixed-cost models where they fit, and stronger cloud models only where they earn their place. You decide where data may go, including UK-only processing where the job requires it.
  • Audit trail. A record of what the system was asked, what it used, what it produced and who approved it.
  • Stops when stuck. If the same problem repeats or a time or cost limit is reached, it stops and tells a person instead of retrying.
  • Integrations. Microsoft 365, shared inboxes, SharePoint and shared drives, spreadsheets, CRM and case systems, databases and forms, connected through controlled, logged interfaces.

AI Rescue

AI Rescue: get an AI project earning, or stop it

AI project rescue for teams that already paid for Copilot, ChatGPT or a custom build that never showed up in the work. Make the tools earn, or stop paying. Staff went back to the old way, results depend on who's asking, and nobody can say what the subscription is for.

We look at the original aim, the tools, the prompts, the data, the costs and how the work actually runs today. Then we give you a written decision: keep, fix, rebuild or stop.

  • Find the break. Prompts, data, the supplier or the workflow: we name which one stopped the work.
  • One job, written down. Not another training day. One repeating task, the sources it may use and the person who signs off the result.
  • Stop is an allowed answer. A licence that never reaches the work is a cost, not a project. If the right call is to stop, we'll write it down so you can defend it.

Works with the tools you already have, such as Microsoft Copilot, ChatGPT, Claude or Grok, or a custom system someone else built.

Questions

Straight answers.

Do we need to know exactly what we want before we talk?

No. Most people start with a rough sense that reporting, admin or follow-up is taking too long. The free call narrows that down to one practical first job.

What makes a good first job?

Something that repeats, already has a way of being done, and still takes too much time, checking or chasing. Weekly reports, first replies, follow-ups and handovers are typical.

Will the AI send things on its own?

Not unless you decide it should, and never for anything that can't be undone. By default it prepares the work and a person approves it.

Can it use the information we already have?

Yes: spreadsheets, documents, shared folders, CRM or case records, reporting exports and databases. It only sees what you allow.

How do you keep running costs predictable?

We use local or fixed-cost models first where they fit the task, set out in advance when paid cloud models are used and what they may process, and monitor costs before anything goes live.

Where does our data go?

That's agreed before we build. Permissions, hosting choices and model routes are visible from day one, and we can design for UK-only processing where the work needs it.

We already have Copilot or ChatGPT. Do we need something new?

Not necessarily. Often the fastest win is making the tools you already pay for work on one real job. See AI Rescue.

Bring us the job that keeps slipping.

A free 40-minute call by phone or Teams, whichever suits you. We'll tell you whether there's a sensible first move, and say so if there isn't. After the call we send a written summary of next steps. We reply within a working day.