Find the outcome you need.
Start with a symptom, a platform or the thing you want working.
With JavaScript, search filters the whole collection in your browser. Without it, use the linked pages and section navigation below; query filters need JavaScript.
Showing entries 1081–1120 of 1179 in this first collection.
Troubleshooting guide
A forecast must beat a simple baseline: how to test one on your own history
Rolling-origin testing against mean, naive, seasonal-naive and drift forecasts, and why a language model is not a reason to trust a forecast.
Troubleshooting guide
Asking your data questions: define the metrics before you let a tool write queries
What two 2026 and 2024 studies found about query-writing tools on raw tables, and why definitions and a question set come first.
Troubleshooting guide
Barcodes and product identifiers in a feed: supply them only when you are sure
What Google Merchant Center says about GTINs and categories, and a simple rule for any tool that fills or matches product data.
Collection
Adding AI to an engineering team: the order to put checks in, and what each one proves
A path from containing the workflow, through tuning the AI reviewer and checking agent pull requests, to judging bot pull requests, sorting CI failures and the backlog, with a bounded job at each step.
Buyer collection
For an engineering lead whose team has stopped reading the AI reviewer: rules, a replay and a monthly audit
Why default AI review gets ignored, what a written rule set can and cannot do, and how to test it on pull requests the tool has not seen.
Buyer collection
For a small team whose AI-written pull requests outrun review: what to check first and what to leave to people
How to take the mechanical checks off reviewers' plates and keep the judgement with them, without letting any bot approve.
Buyer collection
For a lead who inherited a backlog and a codebase: sort the issues, ask the code, close nothing blind
How to get a trustworthy picture of a long issue list and unfamiliar code before planning, with evidence you can check and no automatic closing.
Troubleshooting guide
How to rate your AI reviewer's comments and test new rules without fooling yourself
A way to sample the reviewer's comments, rate each one, hold back pull requests before writing rules, and read a pass mark with its sample size.
Troubleshooting guide
Issue and pull request text can steer an AI agent in GitHub Actions: how to limit what it can do
What researchers have shown, what GitHub does for its own agent, and four controls you can apply: a read-only job, minimum token, checked outputs and a bot account with the least role.
Troubleshooting guide
Coverage says the lines ran; a mutation score says whether your tests would notice a change
Why AI-written tests can reach high coverage and still catch little, what a mutation tool counts, and how to pick three behaviour changes yourself to check.
Inspectable example
Example verdict on one dependency-bot pull request: what changed, what uses it, what to hold
An invented bot pull request with a one-page verdict showing the evidence fields, the cautious wording and the not-examined list.
Inspectable example
Example backlog sheet: one state per issue, with the evidence and a reopen list
Five invented issues sorted into keep, possible duplicate, needs information and proposed stale, with the evidence for each and a closure batch a person approves.
Buyer collection
For a bookkeeper: check supplier statements and payment runs before the money moves
Check statements against the ledger by closing balance and reference, and put a second pair of eyes on every payment run for duplicates and changed bank details.
Buyer collection
For an office manager: get the accounts mailbox sorted and the receipts matched to the card statement
Decide where each kind of mail should go, measure the misroutes that matter, and pair receipts to card lines so missing ones show up as two lists.
Buyer collection
For a finance manager: draft the month-end variance commentary with every figure traced, and know where AI tools send your data
Trace each variance against budget or last month to the ledger lines, ask for the reason from a named person, and write down what each AI tool does with the data you put in.
Buyer collection
For sales admin: turn customers' parts-list requests into quote drafts priced only from your list
Match each requested line to a catalogue code, take the price only from the approved list, flag look-alike codes and units, and let a person send the quote.
Collection
Finance admin checks with AI, in order: start with a file and no access, end with a monthly service
An ordered path through the finance-admin jobs: files only first, then document reading with tie-outs, checks before money moves, drafts for a person, workflows in your own account, then monthly care.
Troubleshooting guide
AI document reading: why a high field accuracy still leaves many documents with an error
A field rate is not a document rate. Count documents, state the sample size and pass mark, and see what a pass actually shows before you rely on a reading tool.
Troubleshooting guide
Checking a supplier statement against your ledger by hand: six kinds of difference
Start with the closing balance, match each line by reference, and classify every difference as missing, extra, amount, date, reference or an unapplied credit.
Troubleshooting guide
Sorting a shared accounts mailbox: keep an Unsure bucket and count the misroutes that matter
Label each email with one of up to eight labels, send doubtful ones to Unsorted with a best guess, and count the invoice, statement, legal or tax-authority mail that lands in the wrong place.
Troubleshooting guide
Hostile text in invoices, emails and receipts: what prompt injection means for finance automation
Any document an automated reader opens can carry instructions. Limit what the reader may do, check its output with code, and test with planted hostile documents, run more than once.
Troubleshooting guide
Before you pay a batch: duplicates, changed bank details and why the builder should not be the only checker
Duplicates survive an exact-match check when the number is reformatted, and changed bank details look like any other row. Add a second checker, a call-back and a list of things noticed.
Troubleshooting guide
Overdue invoice reminders to business customers: interest, fixed costs and who must not be chased
GOV.UK sets optional interest and fixed recovery costs for late business payments. Build a do-not-chase list before drafting, and have a person read every reminder.
Buyer collection
For the office manager who also does the IT: know who has left, who has access and what renews
Start from the staff list, ask for read-only views rather than admin logins, and put every renewal on one calendar with an owner and a backup.
Buyer collection
For a people lead whose inbox holds the same handbook questions: answer from the text, send the rest to a person
Sort questions into answered by the handbook, not covered and for a person, and test the refusals before the answers.
Buyer collection
For an IT lead before a customer security review: be able to show who has access and why
Start from three lists (staff, accounts, admins), know what your licence can and cannot do, and show what changed since last month.
Collection
People and IT housekeeping in order: six checks from the handbook to the last renewal
An ordered path through handbook consistency, staff questions, new-starter lists, leavers, accounts and renewals, each check done from files or read-only exports.
Inspectable example
Synthetic leavers check: eight staff, a directory and three apps, five leaver accounts and one unlisted account
Invented staff list and exports, a short script anyone can run, and the result: which leavers still have an active account, and one account not on the staff list.
Troubleshooting guide
Microsoft 365 access reviews and leaver workflows need specific licences: what the licensing page says
What Microsoft's licensing page says about access reviews and Lifecycle Workflows, who needs a licence, and the manual route from exports.
Troubleshooting guide
Read-only access for an account review: Global Reader, Reports Reader and Google's report scopes
What the read-only roles and scopes are, why read-only is not the same as narrow, and why exports you run yourself are the tighter route.
Troubleshooting guide
A handbook question assistant needs a quote for every answer and a list of what goes to a person
Why an assistant that answers from memory goes wrong, how to design the test around refusals, and what never to put in front of it.
Troubleshooting guide
Tools that sort job applications: what the data protection regulator found in 2024
What the ICO found when it audited AI recruitment tools, which uses the EU AI Act lists as high risk, and what a fenced alternative looks like.
Troubleshooting guide
Before you merge CRM duplicates: what HubSpot says a merge cannot undo
Why a merge needs a sample and an export first, what HubSpot documents about limits and workflows, and how to band duplicate groups.
Troubleshooting guide
Unsubscribes across email tools: what the UK and US rules say, in plain terms
Who counts as an individual, how fast an opt-out must be honoured and what the penalties are, so you know what to test. Not legal advice.
Troubleshooting guide
Google's bulk-sender rules and how to check every tool that sends as your domain
The thresholds Google publishes for spam rate and the authentication needed from large senders, and a monthly checklist per sending source.
Troubleshooting guide
Asking customers for reviews: what the FTC, Google and the UK allow, and what they do not
Asking everyone is allowed. Screening by satisfaction and hiding bad reviews are not, and Google bars rewards for reviews. A short, sourced summary, not legal advice.
Troubleshooting guide
Putting AI call notes in your CRM: the notice and consent you need first
What UK rules ask of a business recording calls, the US lawsuit about notetaker bots, and what HubSpot already offers, before you build anything.
Troubleshooting guide
AI-assisted SEO work and Google's spam policy: what stays on the safe side
Google allows AI as an aid and expects human fact-checking. It treats many pages made to manipulate rankings as spam however they are made.
Troubleshooting guide
Ad spend alerts: what Google already gives you free, and what a set-up job adds
Google publishes a free anomaly script for Ads accounts. Know its limits before you pay for alerts, and what a test on days the bands did not use shows.
Collection
Revenue operations: eight checks in order, before anything is merged or sent
An ordered path through lead flow, junk leads, CRM data, stale deals, email sending, reports, enquiries and quotes, with the test for each and the job that fits.