mash. / guides

Guides.

Background, setup, and the edge cases — the section most guides skip. Every one of these ends with what actually goes wrong and the business-continuity option, not just the happy path.

Central email signatures in Microsoft 365: there's no API for it, so here are the routes that work: Graph can't set an Outlook signature. The three routes that can, what each one can't do, and why I designed for an add-in and a mail-flow connector.

AI Max, Performance Max and Advantage+ on a small budget: what to check before you let the machine drive: Google and Meta now switch their AI features on by default. On a small budget, check tracking first, then the settings that decide where the money goes.

Using AI for SEO without writing for a robot: Where AI earns its keep in SEO (research, briefs, titles, links) and where it gets you in trouble. Plus what Google says AI search needs: nothing special.

Building your own social media scheduler: the code is the easy part: A scheduler is a queue and a cron job. The hard part is each platform's app review, paid API and token rules. What to plan for before writing any code.

The DSPT for a small organisation: where the evidence already is, and where people get stuck: The Data Security and Protection Toolkit for a small clinic, practice or care provider: find the evidence you already have, then fix the gaps.

IT due diligence on a small acquisition: what to check before completion, and the first 90 days after: Buying a clinic or small business? What to check in its IT and Microsoft 365 before completion, and how to bring people, mail and data across afterwards.

ChatGPT, Claude and Copilot: a taste of each, from basic to advanced: A taste of ChatGPT, Claude and Copilot at three levels, with prompts to try, what each one can see, and what to keep out of any of them.

AI prompts for real work: a map of the tips and guides: Copilot, Claude and ChatGPT prompts for HR, finance, admin and IT, grouped by job, each with the check to run before you trust the answer.

APIs and webhooks: why most integrations need both: A webhook tells you something happened. An API lets you check. Why most integrations need both, and how to wire them so a missed event doesn't hurt.

Asking for help in a new job without wasting anyone's time: A simple way to ask good questions at work: what to try first, how to phrase it, and when to stop struggling alone and ask.

A first-week plan for a new starter, from both sides: What a new starter and their manager should each do in week one, day by day, so nobody spends Wednesday waiting for a laptop login.

Running a small project without the paperwork: A lightweight way to run a small project: a one-page brief, a weekly update and a clear finish line, without a PMO or a Gantt chart.

Meetings that earn their slot: A short test for whether a meeting should exist, and a simple structure that keeps it to the point and leaves a record of what was decided.

Write a one-page 'how we work' for your team: A short page that tells new and existing people how your team works: where things live, how to ask for things, and what 'done' means.

Documentation your team will read and trust: Why most internal documents go unread, and a few habits that fix it: one home, one owner, a review date, and writing for the person in a hurry.

Prompt templates that turn a rambling process description into a first-draft Power Automate flow: Building a flow from a blank canvas means guessing triggers and error branches. Give the model a plain-English process first and build against its design.

Prompt templates for a content brief that targets what someone's actually searching for: 'Write a brief for a page about X' gives you a brief about X, not one that answers what the searcher wants. Ask for intent first.

Prompt templates for a Conditional Access review that finds the policy nobody remembers writing: Twenty Conditional Access policies look more secure than three. Often it's just harder to tell which still do anything, or which one cancels another.

Prompt templates that turn scattered notes into a 1:1 you're actually ready for: Skimming last week's Teams messages isn't preparation. A prompt that turns your own scattered notes into an agenda takes five minutes.

Prompt templates that get a structured business document, not a wall of waffle: Ask an AI for 'a policy' or 'a proposal' and you get prose that drifts. A fixed structure per document type keeps it on track.

Prompt templates that turn a budget export and an actuals export into a P&L someone can sign off: Two exports and 'build me a P&L' gets numbers that look neat and aren't reconciled. Templates that get the subtotals right with no invented figures.

Prompt templates for triaging a pile of invoices without missing the one that matters: Spotting the odd invoice in a batch suits an AI tool, but a confident wrong answer can cost real money. Here's how to use it and still check.

Prompt templates that turn a variance spreadsheet into a narrative someone will actually read: 'Explain this variance' gets you the numbers in sentences. A useful narrative says why the number moved and what happens next, and that comes from you.

Getting Copilot or Claude to draft a performance review that isn't a wall of adjectives: 'Write a positive review for a project manager' gets flattery with no evidence. Feed the model your own notes and it can only write what you can back up.

Prompt templates that write a job description a real candidate would recognise: 'A job description for a project manager' gets a list of adjectives. Feed the model the real shape of the role and candidates can self-select.

Evaluating a new vendor's sales deck without just believing it: A vendor pitch answers the questions that flatter the vendor. Have a model read the deck for what it leaves out, then turn that into questions to ask.

Turning 'ask Dave' into a runbook someone else can actually follow: Every team has a process only one person fully remembers. Getting it out of their head is an interview problem, and the model is good at the interview.

Turning a workshop's sticky notes into a RACI that actually resolves arguments: A RACI built in the workshop ends up with three people 'Accountable' because nobody says no to a face. Build it afterwards, from the raw notes.

Writing change communications that survive being read by the people affected: The comms draft that reads well in the steering group and lands badly on the floor usually answers 'what's changing' before 'what happens to me'.

Minutes that read the same whoever took them - a template, not a writing style: Minutes read like whoever wrote them, not like the meeting. A fixed structure fixes that better than a style guide nobody reads.

Turning a shared inbox into an action list Copilot or Claude can actually maintain: A shared inbox doesn't need a cleverer filing system. It needs something that reads each thread and says what it needs, who owns it and by when.

Structured data won't rank you higher, but it decides how you look when you do: Schema markup doesn't move you up the page. It decides whether you get a plain blue link or one with stars, a price or an FAQ. A click problem.

MFA 'enabled' isn't the same as MFA that actually stops anything: MFA looks the same on a checklist whether it's phishing-resistant or a text code an attacker can talk your helpdesk into resetting.

A mailing list without consent tracking isn't an asset - it's a liability with a spreadsheet: 'We've got 4,000 people on our list' means nothing alone. What matters is proving why each person is on it and that they said yes.

Getting Claude or Copilot to write a postmortem that doesn't read like a blame report: Ask an AI to 'write up what happened' and you get a play-by-play with a name on the mistake. Ask for the right structure and people can learn from it.

Why I still run Laragon, not Docker, for Symfony development: Docker solves a team-parity problem I mostly don't have. For solo Symfony work on Windows, Laragon gets me to a working vhost faster.

Wiring Copilot into a real data source to build a finance report that isn't a lie: Asking Copilot to 'summarise our finances' gets you something plausible you can't check. Ground it in a connected dataset and you can.

Prompt templates that get Claude or Copilot to draw the diagram you actually need: Ask an AI for 'a diagram' and you get arrows in prose. A reusable prompt template per diagram type gives you something you can paste into a doc.

Accessibility isn't a specialist's job - five checks anyone building a page can run today: The cheapest accessibility fixes are the ones caught while you build. Five checks that need no special tool and about ten minutes.

M365 JML automation: two linked flows, not one, and why that's the whole design: Joiners, movers and leavers in Power Automate, Azure SQL and Microsoft Graph: two flows linked only by a database, so one can fail without the other.

Giving a Copilot Studio agent real actions: human-in-the-loop is not a checkbox: The difference between a Copilot agent that answers questions and one that does things, and why the approval step is the whole security model.