Giacomo Balli profile picture
Giacomo Balli
Complex decisions. Clear thinking.

Independent advice for business owners navigating technology decisions, challenges and opportunities.
Fluent in business. Grounded in technology. Paid only by you.

Let’s Talk It Through LinkedIn

Building technology products since 2010 · 100+ apps

Blog

Writing on technology decisions, AI and buying software: 683 posts since 2012.

71 posts in Reading

2026

316 posts
  1. Where does a customer complaint go in your business?Coinbase's CEO told a story about their answer. An engineer gave the AI tools that use Coinbase a way to send feedback directly.

    2 min readAI

  2. If you cancel your AI agent in two years, what do you keep?"Give the agent access to your email, the shared drive and the customer system, and it will learn how your business works." That is the pitch now, and there...

    1 min readAI

  3. A 50-person contractor will never hire a full-time analystA 50-person contractor will never hire a full-time analyst. There is not $100k a year of analysis to do.

    1 min readBuying Software

  4. Almost every AI pitch an owner forwards me is priced the same way: hours savedAlmost every AI pitch an owner forwards me is priced the same way: hours saved. Twelve hours a week off invoicing, times a wage, times fifty weeks.

    1 min readAI

  5. PHP's emergency security fix: ask whoever runs your old website for a dateThe PHP Foundation and National Security Agency have issued a warning. People are taking down their websites as precaution until a fix is available.

    2 min readEngineering

  6. The weekly report takes 20 minutes now instead of two hours. Who reads it?Your operations manager now builds the weekly report in 20 minutes instead of two hours. Ask who reads it. That is a real gain on a task.

    1 min readAI

  7. Ask "what should we do with AI" and the answers come from people who sell AIAn owner who asks "what should we do with AI" gets answers from people who sell AI. Start from a different list.

    1 min readAI

  8. The one sentence in an AI agent proposal that is the whole contractThe proposal says the AI agent "connects to your accounting system and handles refunds under $500." Read that sentence twice.

    1 min readBuying Software

  9. Check what your booking system can already do before you replace itThe proposal to replace your booking system usually arrives before anyone has checked whether the current one can already do what you need.

    1 min readProduct

  10. Before you price a new inventory system, price the spreadsheetBefore you price a new inventory system, price the spreadsheet. Count what the drift cost you last quarter.

    1 min readEngineering

  11. Go-live is a guess until your busiest staff run a full day in the new systemIf your two busiest employees have not run a full working day in the new system before go-live, the go-live date is a guess.

    1 min readEngineering

  12. Before you take another AI meeting, write down the one task your staff hate mostBefore you take another AI meeting, write down the one task your staff hate most. Every pitch starts from the vendor's product.

    1 min readAI

  13. If users leave your app at the login screen, ask why the login is thereIf most people who download your app leave at the login screen, find out why the login is there before you pay anyone to redesign it.

    1 min readProduct

  14. Before a new tool demo, walk through the last time the problem happenedWhen your operations manager asks for a new tool, ask them to walk you through the last time the problem happened, step by step, before anyone watches a demo.

    1 min readPricing

  15. Ask whoever recommends technology whether they earn anything if you say yesBefore you act on a technology recommendation, ask the person making it, in writing, whether they earn anything if you say yes.

    1 min readPricing

  16. Your competitor's app is just where one better customer task livesWhat changed for your competitor's customers is probably one task: booking at 11pm, reordering in two taps, seeing when the technician will arrive.

    1 min readMobile

  17. A project stuck at 80% for months is no longer measured by that numberWhen a project has been 80% done for months, the number has stopped measuring the project, and you are paying for it every month it holds.

    1 min readEngineering

  18. Three questions decide whether an AI agent can handle a jobIn plain terms, an agent is software that carries out a multi-step job across your tools. Chasing unpaid invoices is a good example.

    1 min readAI

  19. If the agency registered your domain in its name, it holds the keysIf the agency opened your domain, your app store account and your code repository in its own name, you paid for the work and someone else holds the keys.

    1 min readBuying Software

  20. A failing technology project shows up at the accountant firstThe first sign of a technology project going wrong usually reaches the accountant before it reaches anyone else.

    1 min readReading

  21. No security consultant can sign off on your SOC 2No security consultant can sign off on your SOC 2. Only a licensed CPA firm can issue the report.

    1 min readEngineering

  22. AI-built product security worries usually land on the wrong layerWhen an owner worries about the security of an AI-built product, the worry usually lands on the wrong layer.

    1 min readAI

  23. The same question comes round the table every few monthsThe same question comes round the table every few months wearing different clothes. Should we build it. Should we replace it.

    1 min readAI

  24. App vs firmware is the most expensive decision in a connected productEvery feature you put in the app is one your device cannot do on its own. Every feature you put in firmware is one you cannot fix quickly.

    1 min readEngineering

  25. For a hardware company, the most expensive screen is the pairing screenFor a hardware company, the most expensive screen in the product is the pairing screen. The device is designed, tooled, certified and boxed.

    1 min readEngineering

  26. Before hiring an AI consultancy, find the part that actually needs AIBefore hiring an AI consultancy, work out which part of the problem actually requires AI. It is usually smaller than the proposal.

    1 min readAI

  27. The size of the invoice is the wrong measure of a decisionOwners tend to judge whether a decision deserves scrutiny by the size of the invoice. The invoice is the wrong instrument.

    1 min readReading

  28. Your IT person is the wrong one to ask about a $200,000 buildThe person who keeps your network running is usually the first one an owner asks about a $200,000 build. They will answer.

    1 min readPricing

  29. When a roof needs replacing, nobody thinks it is strange to call a second rooferWhen a roof needs replacing, nobody thinks it is strange to call a second roofer. With software, most owners ask the vendor whether the vendor's plan is a...

    1 min readEngineering

  30. Before your next technology investment, ask someone who gains nothingBefore your next major technology investment, ask one person who does not make more money if you say yes. Count the people who have been involved so far.

    2 min readProduct

  31. Technology due diligence should not happen after you have decided what you wantTechnology due diligence should not happen after you have decided what you want to do. By then it is not diligence.

    2 min readProduct

  32. The best recommendation I can give is sometimes "don't build anything"The best technology recommendation I can give a client is sometimes "don't build anything." It is also the one clients find hardest to accept, because they...

    1 min readEngineering

  33. "Our competitor is doing it" is the worst justification for a tech investment"Our competitor is doing it" may be the worst justification for a technology investment that a sensible executive regularly accepts.

    2 min readEngineering

  34. A $5,000 technology decision can deserve more scrutiny than a $500,000 oneA $5,000 technology decision can deserve more scrutiny than a $500,000 one. Owners scrutinise by price, and price is a poor guide to consequence.

    1 min readAI

  35. A CEO sent me a technology proposalA CEO sent me a technology proposal. These are the first three things I ignored. The feature list.

    1 min readBuying Software

  36. A popular website with no app is usually not an opportunitySomeone will eventually bring you a popular website with no app and call it an opportunity. Usually it is not one.

    1 min readMobile

  37. You do not have a technology problemYou do not have a technology problem. You have nobody representing the buyer. Look at who is in the room when a $400k decision gets made.

    2 min readEngineering

  38. The contract clauses that create the most lock-in get the least scrutinyThe technology contract CEOs scrutinise least is often the one that creates the most lock-in. Price and term get negotiated.

    2 min readBuying Software

  39. The most worrying sentence in a vendor demo: "Yes, we can customise that"The sentence in a vendor demo that should worry you most is "Yes, we can customise that." It sounds like flexibility.

    2 min readBuying Software

  40. In a SaaS agreement, the data exit clause matters more than liabilityWhen a software or SaaS agreement crosses your desk, the liability language usually gets read hardest. The clause that decides what actually happens is quieter.

    1 min readPricing

  41. A CEO I advise canceled the company's hosting accountA CEO I advise canceled the company's hosting account, expecting the servers to keep running until the end of the month.

    1 min readEngineering

  42. Every CEO eventually sits in the sunk-cost meetingEvery CEO eventually sits in the sunk-cost meeting. It opens with a sentence that sounds like an argument and is not one: "We've already spent $300k." That...

    2 min readSecurity

  43. Your software project is six months lateYour software project is six months late. The options are fix it, replace the team, cut scope, or kill it, and every person advising you has a stake in one...

    1 min readEngineering

  44. Your business probably does not need an appYour business probably does not need an app. I have shipped apps since 2010, so this is not a position I hold lightly.

    2 min readPricing

  45. The $180,000 dispatch system and the question about month five"The $180,000 dispatch system will take four months. Before AI it would have been nine." Fine. Ask what happens in month five.

    1 min readAI

  46. I run paid ads across a portfolio of more than 40 appsI run paid ads across a portfolio of more than 40 apps. Exactly one campaign makes money. It sells a conduit bending tool to electricians.

    1 min readPricing

  47. The last moment a replacement decision is still cheap to changeA client mentions, almost in passing, that they are getting a quote to replace the system that runs the business.

    1 min readEngineering

  48. Before spending $250k building software, spend a week trying to kill the ideaBefore spending $250k building software, spend a week trying to kill the idea. Not a week of planning.

    2 min readEngineering

  49. The most useful question before approving a software projectAsk it in the meeting where the budget gets requested, and watch the room. If the answer is a number, keep listening.

    1 min readEngineering

  50. Beautiful software architecture historically mattered because humans neededBeautiful software architecture historically mattered because humans needed to understand and modify systems cheaply.

    1 min readAI

  51. Universities are becoming very expensive places to buy time, friendshipsUniversities are becoming very expensive places to buy time, friendships, and adulthood. My son is in kindergarten.

    1 min readEngineering

  52. A custom domain can hide the builder from your customersA custom domain can hide the builder from your customers.It cannot hide it from the internet.

    1 min readEngineering

  53. Nobody hacked Hugging FaceNobody hacked Hugging Face. At Black Hat this month, OpenAI presented what happened when they turned unconstrained agents loose on a cybersecurity evaluation.

    1 min readAI

  54. Everyone's talking about agentsEveryone's talking about agents. Almost nobody has written down what theirs is allowed to do. An agent isn't a smarter chatbot.

    1 min readAI

  55. "We don't have anyone who knows how to do this." Neither did the companies"We don't have anyone who knows how to do this." Neither did the companies that figured it out.

    1 min readAI

  56. Most AI pilots produce nothing measurableMost AI pilots produce nothing measurable. That's not a technology problem. It's a scoping problem, and it looks like this: the pilot starts with "let's see...

    1 min readAI

  57. "I don't trust it to be right." You shouldn't"I don't trust it to be right." You shouldn't. That's the correct instinct and it's the wrong conclusion. The mistake is framing this as trust vs no trust.

    1 min readAI

  58. Everyone seems focused on when we’ll reach AGIEveryone seems focused on when we’ll reach AGI.But I think we’re missing the more immediate point: the AI we already have is capable of doing a huge portion...

    1 min readAI

  59. Four months after I shipped SnackCheckApp.comFour months after I shipped SnackCheckApp.com, a stranger filed my app's name as an EU trademark. Same name, same three classes my app operates in.

    2 min readMobile

  60. How to make AI "know everything" about your businessI often get asked by companies how they can make AI "know everything" about their business.

    2 min readAI

  61. I spend a lot of time helping business owners think through AI strategyI spend a lot of time helping business owners think through AI strategy. One thing that keeps coming up is that we've been assuming AI agents will use...

    2 min readAI

  62. Everyone seems to be reading this chart from Financial TimesEveryone seems to be reading this chart from Financial Times as: "AI is creating lots of apps nobody wants." That's not what the research says.

    1 min readAI

  63. At #WWDC26, Apple finally delivered much of what it had promised two yearsAt #WWDC26, Apple finally delivered much of what it had promised two years earlier. Many people will look at that and conclude Apple was behind on #AI.

    2 min readAI

  64. I grew up in ItalyI grew up in Italy. I live in the US now. Same human problem, two opposite answers, and I've watched both up close.

    2 min readAI

  65. A 2014 study shows walking significantly boosts creative thinkingJust came across a fascinating 2014 study from the American Psychological Association that scientifically validates something I've experienced for years...

    1 min readProduct

  66. Another INNOVIT - Italian Innovation and Culture Hub startup mentorship eventAnother INNOVIT - Italian Innovation and Culture Hub startup mentorship event in the books. Always refreshing to work with non-tech founders.

    1 min readStartups

  67. Most people assume digital has already wonMost people assume digital has already won. Recent data from Pew Research Center shows a different reality: - Print is still the dominant format - ~75% of U.S.

    1 min readPricing

  68. The AI job everyone is talking about does not existThe AI job everyone is talking about does not exist. “Prompt engineer” was a temporary phase.

    2 min readAI

  69. Everyone is talking about “AI agents replacing ecommerce.”Everyone is talking about “AI agents replacing ecommerce.” Reality check: the first real experiment didn’t go well.

    1 min readAI

2012

111 posts
  1. Donations in iPhone apps - one of the few things to be done outside the appLast week I finished the app for Florence's mayor Matteo Renzi. I personally like him and he will be running to be prime minister.

    2 min readMobile

  2. 31 Major Causes of FailureBefore leaving for San Francisco back in August, my older sister Monica (yes, the celebrity wedding planner) gave me a book she really liked: "Think and...

    5 min readReading