AI assistants & agents

An assistant reads a request written in plain language, asks for what is missing and prepares a result for a person to check.

FREE TEXT FOLLOW-UPQUESTIONS PROPOSAL WITHREASONING PERSONCONFIRMS FREE TEXT FOLLOW-UP QUESTIONS PROPOSAL WITH REASONING PERSON CONFIRMS

Examples

  • Free-text reports with follow-up questions
  • Answers from your own documents
  • Draft replies for review
  • Routing incoming messages
Details and limits

How we work

Fixed rules come first. A language model is used only where rules can’t cope with free text. Every proposal shows its reasoning, and decisions that affect people stay with people.

Agents

An agent also carries out agreed actions itself, such as filing a record or sending a notification.

Limits

Language models make mistakes, and some of them sound convincing. We don’t build assistants that decide alone on matters with consequences for people. A model’s answer is never presented as legal or medical advice.

Process automation

Steps repeated by hand every week, taken over by tools that Microsoft 365 already includes, such as Power Automate and SharePoint lists.

TRIGGER STEPS RUN CHECKED RESULT TRIGGER STEPS RUN CHECKED RESULT

Examples

  • Forms that write into a list
  • Approvals with reminders and a record
  • Documents filed from templates
  • Exports no longer built by hand
Details and limits

How we work

We start with one process and map it as it really runs. We build in your own tenant, with the permissions your IT team grants, and document it so that someone else can follow it later.

Limits

What is possible depends on your organisation’s licences. We don’t administer whole Microsoft 365 environments or replace your IT support. An unclear process stays unclear when it is automated, so sometimes the first step is to simplify it.

Data & dashboards

Figures from several Excel files, exports and lists, brought into one view that updates itself instead of being rebuilt each month.

SOURCES CLEANED MODEL DASHBOARD SOURCES CLEANED MODEL DASHBOARD

Examples

  • One clean table from many exports
  • Only the figures a team uses
  • Early flags for missing entries
Details and limits

How we work

First we agree which questions the figures should answer. Then we choose the tool, usually one your organisation already uses. Every number on a dashboard can be traced back to its source.

Limits

A dashboard is only as good as the data behind it. Cleaning that data often takes longer than building the view, and we say so at the start. We don’t build large data warehouses.

Software & web apps

Small applications for a specific job, where no existing product fits and a spreadsheet has reached its limit.

NEED PROTOTYPE TOOL IN USE NEED PROTOTYPE TOOL IN USE

Examples

  • Replacing a shared spreadsheet
  • Form-based web app with admin view
  • A plain website without tracking
Details and limits

How we work

We keep dependencies few and the code readable. Before building, we check whether an existing product would do the job. If it would, we recommend it.

Limits

Our team is small. We don’t take on large platforms or systems that need an operations team around the clock.

Consulting & training

We come to your premises, advise on where AI and automation make sense for your organisation, and train your team to use the tools.

VISIT ASSESSMENT RECOMMENDATIONS TEAMTRAINING VISIT ASSESSMENT RECOMMENDATIONS TEAM TRAINING

Examples

  • Reviewing processes on site
  • Where AI helps, and where not
  • Training a team on the tools
Details and limits

How we work

We look at the work where it happens, at your premises and with the people who do it. Then we train your team on the tools that fit.

Limits

Advice stays advice: we say where we see a benefit and where we don’t, including when no new software is needed. The decisions stay with your organisation.

Next step

Not sure which field your question belongs to?

That is normal. Describe the process, and we sort it out in a first conversation.