EthosPowerAIWhat has changed
Your staff always knew more than your customers did: the standards, the sizing methods, the failure modes, which product suits which job. That gap earned your margin. A prospect's own AI now prepares the load list, sizes the cables and the battery, names a make and model and lists the suppliers, your competitors among them, before anyone picks up the phone. By the time they call, you are quoting on price.
Adding AI to the processes you already run does not fix that. It does the same work faster. A process that has just become quick and pleasing to watch is one nobody wants to change. McKinsey's 2025 survey found 88 per cent of respondents using AI somewhere in the business and 39 per cent able to point to any profit from it.
Four steps, in this order
- Write the five-year plan. Decide what your clients will need from you in five years and how you will help them succeed. Move from supplying a product to advising on its whole life, the way a bank that knows a client's supply chain can help the client run it better.
- Find the problem that never ends. A chronic cost in your clients' core work, one they will pay to have solved every year, not once.
- Use the data their AI cannot reach. Test results, SCADA logs and maintenance records sit behind private networks, and the public AI models cannot see them. A client opens them only to a supplier it trusts. Your reputation is the key to that door.
- Prototype with the client, then run it. A working version in weeks, tested and changed with the people who will use it, then run as a secure, auditable service with a person signing off each decision. Tell your own engineers first what they gain from it; nothing I build cuts headcount.
A worked example: standby batteries
A network utility runs hundreds of substations, each with a standby DC system whose charger tests its own battery on a schedule and reports to SCADA over DNP3 or Modbus TCP, so decades of those results already sit on the utility's own network. Every one asked the same question: when will this battery fail, and when should I order its replacement? Get it wrong and the switchgear loses its supply; order late and a maintenance outage waits on a delivery.
The supplier who wins here builds a platform that reads those results, forecasts which strings to replace and when, checks stock, lead times and prices, schedules the site work and recommends when to raise the purchase order, with a person approving each step. The utility gets fewer surprises; the supplier becomes the one it plans with rather than the one it rings when something fails. None of this is as easy as it sounds. The hard part is always the data, not the model.
Why it is affordable now
One experienced developer can now build and deploy a platform that once took a funded team years. ebatt.ai, my battery sizing and quotation platform, is one: IEEE 485 and EN 50272-2 calculations and the standards checks I have used since 1992, in the client's own hands, and open-source so it can be your starting point. When the software can be copied, it stops being the barrier; the private data and the trust behind it are what a competitor cannot copy.
Where this comes from
I have worked this way before, on a smaller scale than the word ‘strategy’ suggests: giving a small firm's distributors remote access to its ERP system in the late 1990s so they could serve themselves, writing courses so engineers in other countries could size standby systems without waiting for me, and since 2024 building ebatt.ai. The details are on the about page.
How I work
One person holds the strategy, the design, the build and the training, so nothing is lost between the engineer who understands your business and the one who writes the software. I work from Nelson across New Zealand and Australia with no office lease and no sales layer; the rates are on the services page. Everything I build is open-source and self-hosted, so your compliance team can read the code and you keep it if you leave. The 32 platforms are on their own page.
The limits of this argument
Afanasyev's account comes from practice, two recorded interviews rather than a study, and his employer sells cloud services. My cases are three cases. Test the idea cheaply: ask your next three customers what their AI told them before they called you.