Conversation

The first call is short. We want to know what the pile of text is, who lives with it, and what a Tuesday would look like if a tool took the dull middle of a task. Example files and a plain description of the week are enough.

We need a person who can say yes or no to a sample of answers, and a sense of where the documents sit. A named contact for access comes later. If the job is a public chatbot, a content mill, or a request to score people, we will say so.

The stage ends with a written note: whether a discovery week is worth the five days, what we would want to read, and the dates we can start. If the six slots are taken, we offer a place on the waiting list.

Stage 02

Discovery week

We spend five days inside the process, at a desk in your office or on a shared screen. We read a real slice of the archive and sit with the people who touch it. It is a reading week with enough conversation to know which questions actually get asked when a customer, a patient, or a supplier is waiting.

From you we need a sample under the confidentiality agreement, a few hours with the people who do the work, and one person who can tell us when an answer would be wrong in a way that matters. We keep the access small and write down what we touched.

The week ends with a paper: three options, a view on effort, and a go or stop. Stopping is a complete outcome. We have ended weeks by recommending that nothing be built.

Stage 03

Prototype

If the paper says go, we build a slice on real material, used by the same people who would live with the finished tool. We would rather watch a clerk search for a clause they remember than run a polished demonstration on a file we chose. Failures are collected in a list.

From you we need time with those users, a stable sample of documents, and a willingness to say when a reply is politely wrong. We keep the prototype inside the access we already have.

The stage ends when the slice does the core job on a typical morning, with a written list of gaps that hardening must close: access, logs, evaluation questions, the queue for doubtful cases. If the slice cannot do the core job, we stop.

Stage 04

Hardening and evaluation

This is the unglamorous middle. We put access rules in place so that a person who cannot open a file cannot ask about it. We turn on a log of queries. We write the hundred to three hundred questions with you, agree the answers, and wire a score that we can re-run when a prompt or a model moves.

From you we need decisions: who may see which folders, what a passing score looks like, who sits on the exception queue. Those decisions belong to the organisation.

The stage ends when the evaluation set runs without us in the room, the logs can be read by someone on your side, and the failure path is something a person can follow on a Friday afternoon.

Stage 05

Hand-over

We sit for half a day with the people who will use the tool and the person who will keep the files in order. We walk the ordinary path and the failure path on your material. We leave a written guide, repository access, and a short list of knobs that are safe to turn. The rest we ask you to leave until you call us.

From you we need that sitting to be protected time. A hand-over squeezed between other meetings tends to become a recording that nobody watches. We would rather move the date than pretend a rushed hour was enough.

After the sitting, the studio steps back. An optional monthly visit can follow if you want the scores, the spend, and the small repairs watched. If you do not, you still have the harness and the guide.

Rhythm

Pace and how to reach us.

The studio keeps ordinary office hours: Monday to Friday, 09:30 to 18:30 Singapore time. During a live piece of work we schedule two short calls a week unless the week itself is a discovery week, in which case we are effectively in the same room. Between those calls we write in mail. We do not run a chat that expects an answer in ten minutes.

We reply to mail within one working day, and usually the same morning if it arrives before noon. Urgent access problems during a build are the exception; we give a direct number for those days. Visits to Bussorah Street are by appointment. The rooms are on the second floor and we would rather set a time than leave someone waiting on the stair.

When a project is in the monthly retainer, the visit is a booked half-day. Outside that, we are not on call. That boundary is how six projects a year stay finished rather than half-open.

Notes from a discovery week

Money

How we count the work.

After discovery, each stage has a fixed sum agreed in writing before we start it. We do not bill by the hour and then surprise you with a week that ran long. If a stage is going to miss the shape we agreed, we stop and rewrite the stage rather than quietly padding it.

Discovery itself is also a fixed sum for the five days. You pay for the paper even if the paper says stop. That is the point of the week: a cheap, honest no is better than a build that should not have been started.

Spend on hosted models and programming interfaces is billed separately, at cost, with the invoices or usage records attached. We do not mark it up, and we do not hide it inside the stage sum, because it moves when a vendor changes a price or when a week is busier than the last. You see the number. You can cap it.

Care

Data and access.

We take the smallest access that still lets us do the reading. Shared folders with expiry dates are better than a standing account. We copy only what we must copy in order to work, and those working copies are deleted after hand-over. We write down what we deleted.

A confidentiality agreement is signed before discovery, not after we have seen the files. We prefer storage in Singapore or the nearest region that your own rules already allow. If a document cannot leave the building, the build sits next to it. That decision is made in discovery, not as an afterthought in week six.

We do not train a public model on your material. We do not use the files as examples in a later pitch. The archive stays yours; the code we write for the tool is handed over as agreed in the stage papers.

Stop

When the right ending is to build nothing.

Discovery can finish with a recommendation to leave the tools you already have, tidy the archive, and wait. That has happened, and it is a normal outcome. The week still earns its fee because you get a written view instead of a polite build that nobody asked to run.

We would rather have that conversation in week one than in month three. If you already know you want a product regardless of what the files say, we are the wrong studio. If you want someone to look, and to say so if the look is disappointing, write to us.

Start a conversation See the practice