Skip the big strategic project. Start with the task someone on your team does every week, the same way, that nobody enjoys. We design and build that, put it to work in your environment, and keep measuring whether it is still getting it right.
A boring task is one where you already know what a good answer looks like. That makes it easy to tell whether the agent got it right, cheap to correct when it did not, and obvious whether it was worth doing. The ambitious project gives you none of those things, which is why so many of them end as a pilot nobody talks about any more.
Sorting the same customer questions. Pulling the same numbers into the same report. Chasing the same status updates. Checking the same documents against the same rules. The work is unglamorous and it is exactly where the time goes.
Four things hold on everything we deliver. They are the reasons our clients let this software near their operation at all.
An agent we build drafts, sorts, drafts again and suggests. It does not quietly commit things on your behalf. Where an action matters, somebody on your team sees it before it happens, and can see the reasoning behind it.
The agents we build work from the documents, records and answers your own people have approved. That is what keeps the answers consistent with how your business actually operates, rather than plausible and wrong.
We do not use your data to train models, and we do not take it anywhere it does not need to go. Access is scoped to what the job requires, and you keep what we build.
An agent that was right in week one and drifts by week twelve is worse than no agent, because people have stopped checking by then. We set up the measures that show whether it is still getting it right, and the alerts for when it is not.
The hardest part is usually picking the task. This is the sheet we hand out at conferences, and it takes about ten minutes. You name up to three tasks you repeat, work out what they cost you in hours a year, then score each one twice: how well it suits an agent, and how clear the process behind it already is.
Those two scores land the task in one of six places. Some are a strong first agent. Some are worth automating but need the process made visible first, which is the other half of what we do. Some should stay with people for now, and the sheet says so plainly rather than pointing everything back at us.
Print it, fill it in with your own team, and keep the answer. There is no form and nothing to sign up for.
Plenty of this work arrives on its own, and that is fine. It also arrives the other way round, out of the improvement work or out of PeelForward, which watches an operation continuously and names the few processes where automation would actually pay off. Either way you are not guessing which task to start with.
That is usually the right first conversation, and it is a short one.
