Business strategy
Every business needs its own software in the AI era
My belief is simple: in this era, every company needs to build some of its own software. Not because every company should become a tech startup. Because your people should not spend their day doing work a system could already handle.
The expensive part is often the work between the tools
Picture a service business with 40 employees. A customer sends a request by email. Someone copies the details into the CRM, looks up the previous quote, asks a colleague about availability, and writes a reply. None of those steps looks dramatic. Together, they eat the morning.
Buying another subscription does not necessarily fix that. The CRM stores the customer. The inbox stores the conversation. A spreadsheet stores the schedule. A person is still acting as the connection between all three. That is where I would look first for an AI project.
Own your workflow, not every piece of technology
When I say ‘build your own software,’ I do not mean rebuild email, accounting, or a language model. I mean own the part that reflects how your business actually operates: your quoting rules, your handoffs, your approval process, and your definition of a good outcome.
A small internal application can connect existing tools and let an AI agent prepare the next step. For the service business, it could assemble a request summary, retrieve the right customer record, flag missing information, and draft a quote for a person to approve. The value is in the complete workflow, not the chat window.
Why I set a 50% target
I think businesses should aim to remove at least half of the repetitive work in suitable office workflows. If a team still copies, summarizes, categorizes, and follows up on everything manually, I would question whether it is using the tools available to it well enough.
That is my target, not a claim that AI has already cut every office job in half. The 2023 NBER working paper Generative AI at Work reported a 14% average productivity increase in one customer-support setting. Results varied by worker. That is a good reason to measure your own process rather than promise a universal number.
Half the repetitive work also does not mean half the people. I want people spending more time solving customer problems, making decisions, and building the business. Removing busywork gives them room to do that.
Start smaller than you think
Choose one task that repeats every week. Watch somebody do it from beginning to end. Write down the inputs, the systems they open, the decisions they make, and the mistakes that would matter. Then build the smallest version that takes a useful chunk of that work off their plate.
Run it alongside the current process first. Have the team review the output. If it saves time after review and correction, expand it. If it only produces impressive demos, change the workflow. A mid-sized business does not need an enormous AI transformation plan to make its first meaningful improvement.
The question I would ask on Monday
Where are capable people spending hours moving information instead of using their judgment? Pick one answer. Assign an owner. Measure the current workload. That is a more useful starting point than asking which AI tool everybody else is buying.