https://voxalone.com/insights Practical writing on software engineering, AI delivery and technology decisions. en-gb https://voxalone.com/insights/golden-dataset-300-labelled-examples https://voxalone.com/insights/golden-dataset-300-labelled-examples At 50 examples, a 90% accuracy could really be 79% or 96%. At 300 it is 86–93%. How to build the labelled set that tells you whether an AI system works. Anitha Prince AI & Machine Learning Wed, 23 Sep 2026 00:00:00 GMT https://voxalone.com/insights/cost-of-automating-document-processing https://voxalone.com/insights/cost-of-automating-document-processing Reading a page costs about a cent. The real cost of document automation is exceptions, matching and upkeep, and below a certain volume it never pays back. Anitha Prince AI & Machine Learning Tue, 22 Sep 2026 00:00:00 GMT https://voxalone.com/insights/making-tax-digital-what-a-practice-can-automate https://voxalone.com/insights/making-tax-digital-what-a-practice-can-automate Automate the MTD quarterly update, which is unadjusted category totals. The records chase and the year-end judgement are where a practice’s time actually goes. Anitha Prince Strategy Mon, 21 Sep 2026 00:00:00 GMT https://voxalone.com/insights/offline-first-field-apps https://voxalone.com/insights/offline-first-field-apps The local database is the easy half. Offline-first field apps break at the queue: partial signal, idempotency windows shorter than the outage, and conflicts. Anitha Prince Engineering Sun, 20 Sep 2026 00:00:00 GMT https://voxalone.com/insights/integrating-with-systems-that-have-no-api https://voxalone.com/insights/integrating-with-systems-that-have-no-api A system with no API usually has four ways in besides the screen. Use the one furthest from the UI, and treat reading and writing as different problems. Anitha Prince Engineering Sat, 19 Sep 2026 00:00:00 GMT https://voxalone.com/insights/uk-gdpr-language-models-where-your-data-goes https://voxalone.com/insights/uk-gdpr-language-models-where-your-data-goes One LLM API call copies personal data into provider logs, stored state, your logs, tracing tools and embeddings. Under UK GDPR each copy needs a deletion date. Anitha Prince AI & Machine Learning Fri, 18 Sep 2026 00:00:00 GMT https://voxalone.com/insights/replacing-the-spreadsheet-only-one-person-understands https://voxalone.com/insights/replacing-the-spreadsheet-only-one-person-understands A critical spreadsheet is a production system with an undocumented spec and one maintainer. Replacements fail when they copy features, not decisions. Anitha Prince Engineering Thu, 17 Sep 2026 00:00:00 GMT https://voxalone.com/insights/why-ai-pilots-fail-to-reach-production https://voxalone.com/insights/why-ai-pilots-fail-to-reach-production The gap between an impressive AI demo and a production system is evaluation, guardrails and operational ownership. Here is what actually closes it. Anitha Prince AI & Machine Learning Thu, 17 Sep 2026 00:00:00 GMT https://voxalone.com/insights/buy-versus-build-software-decision https://voxalone.com/insights/buy-versus-build-software-decision A framework for deciding whether to buy off-the-shelf software or build custom, based on process differentiation, total cost of ownership and switching risk. Anitha Prince Strategy Thu, 17 Sep 2026 00:00:00 GMT https://voxalone.com/insights/what-good-technical-due-diligence-looks-like https://voxalone.com/insights/what-good-technical-due-diligence-looks-like Technical due diligence beyond code quality: the operational, delivery and key-person risks that decide whether a software asset is worth what it appears. Anitha Prince Engineering Thu, 17 Sep 2026 00:00:00 GMT