The question
We built an AI assistant for our own website, and it is running. Should we sell it to others now? If so, to whom, and how should we charge?
Three options
- Sell now. Sites we build and sites clients already have, both.
- Build a small part first. Put it on one more of our own sites to prove a second one can follow the same steps; then take one or two pilots.
- Not yet. Use it on our own sites only, and decide again when the conditions are in place.
Scope and architecture
The assistant answers only from what the site itself says, names its sources, and hands over to a person when it cannot answer. Sites come in two kinds, and their costs are far apart: on a site we build, the knowledge comes straight from the site's build; on a site a client already has, pages must be fetched and cleaned, and the client's domain and security settings dealt with.
The charging structure follows the one we already use: a one-off setup plus a monthly subscription. Not per question: questions cost little, and charging by them only makes clients afraid to use it.
Three budget levels (amounts removed from this sample)
In the plan you receive, this section gives three amounts and what each one includes. Here we only show how we worked them out:
- Cloud costs at the worst case, a single paying site, and then what they come to when shared across more sites.
- People's time in three parts: the one-off work to set up a site; the platform work that must be finished before selling, shared by every site; and the monthly care of each site.
- The internal hourly rate is an assumption, written down on its own for the owner to accept.
What we did not know, and how to find out
- The hours are estimates, not timesheets. Time the next site, then come back and correct the table.
- Model costs move. A different model can cost many times more. Price on the default model only; when the model changes, read the official price page again rather than writing a figure in advance.
- How many sites it takes to break even depends on the two points above. Recalculate with the first few sites' real figures.
Risks and priorities
- Selling to sites clients already have needs a large piece of platform work that is not done: cross-site embedding and a general page fetcher. It comes later.
- The free plan has limits that would bite together: the number of scheduled tasks, the daily question allowance, and the assistant pausing while an evaluation runs. Upgrade when the first paying client signs.
- Regulated industries — mortgages, insurance, health, law — stay out of the standard price list: quoted separately, and checked line by line by a licensed person before going live.
Conclusion: not yet
Do not sell now. When we do, sell only for sites we build at first. Next step: put it on one more of our own sites to prove it can be repeated; then take one or two pilots; only then put prices on a page.
What happened next
On 10 October 2026 the assistant also went onto the PWE Studio product pages and the PWE Tools site — the second and third sites, both ours. Offering it to others is still in preparation, and no page carries a price.
This is what the workshop hands you: the question, the options, what is unknown and how to find out, the risks, and a next step you can act on. Sometimes that step is "not yet".