When I tell an operations lead that our pilot takes ten working days, the next question is almost always the same: what do we actually get at the end? It's the right question. A lot of pilots end with a slide deck and a nice video, and the team is no closer to using anything.
So here is the honest answer, day by day. Every task is different and the days shift a little, but the shape stays the same.
Before day one
We agree on one task. Not a department, not "AI for finance", just one piece of work that eats your team's week, like supplier invoice intake or sorting customer emails.
You name one owner on your side. This is the person who knows how the task really works, can approve access and can say what "good" looks like. We sign an NDA and a data agreement before any data moves, and we settle where the data is allowed to go for this task.
From you, the pilot needs that owner, a set of real examples, access to the systems involved and about two hours a week of your team's time.
Day 1: measure how it runs today
We sit with the people who do the task and watch them do it. Then we write down the starting numbers: how long one item takes, how many come in each week, how often something goes wrong and what that costs.
From those we work out the money number, which is what the task costs you today and what it would be worth to fix. We also write the success line together: the result that would make you say yes to going live. If we can't find a number worth chasing, we tell you on day one and stop.
Days 2 and 3: build the test set
We collect 100 to 200 of your real cases with the right answers attached. That means the messy ones too: crooked scans, missing fields, emails that switch between Arabic and English, the supplier who always sends two PDFs.
This test set is what we check every change against for the rest of the pilot. It stays with you afterwards, so you can use it to check any AI tool you try later, ours or anyone else's.
Days 4 and 5: test the models
We run Claude, OpenAI and open models through the same test set. For each one we record how often it gets the answer right, what one finished task costs and where your data goes. That becomes the scorecard.
The "best" model is often different for different steps of the same task. Sometimes a plain rule beats all of them for one step, and we say so.
At the end of day 5 we sit down with your owner and go through the scorecard. If the results are poor, this is the point to stop, and you keep the numbers and the test set.
Days 6 to 8: build the first working piece
This is where most pilots fall short, so we spend the most time here. We build the first working piece inside the tools your team already uses: reading from the inbox, the ERP or the shared drive, and writing back in a way a person can check.
A person approves anything important before it goes anywhere. The AI only handles the kinds of cases it has been tested on, and anything else goes to a person as it does today.
Day 9: run it on real work
We run it on the current week's work, next to your team. Then we measure again, the same way we did on day one, so you can see the before and after side by side on time, cost and accuracy.
Day 10: the readout
On the last day you get:
- The starting numbers and the new numbers, measured the same way.
- The test set built from your real work.
- The model scorecard, with accuracy, cost per task and where the data goes.
- The working first piece, running on your data inside your tools. It carries on into the full rollout. Nothing gets thrown away.
- A runbook and handover doc, so your team can check it and knows what to do when something looks wrong.
- A fixed price and a 4 to 6 week plan for the full rollout, with the result we must hit agreed in writing.
Or, if that's what the numbers say, a plain recommendation to stop.
What ten days can't do
Ten days gets one task working on real data with honest numbers. It does not get that task live for your whole team, connected to every system, with every rare case covered. That is what the 4 to 6 week rollout is for.
It also won't fix a process that is broken for reasons that have nothing to do with AI. If the real problem is a missing approval step or a form nobody fills in properly, we will tell you that instead of building around it.
Why we price it this way
The 10-day pilot has one fixed, published price: USD 7,500 (about AED 27,500), agreed before we start. If you go ahead with the full rollout within 30 days, the fee counts toward it.
We don't sell a model, so we have no reason to push one. The pilot is there to give you a decision you can trust, backed by your own numbers.
If one task is eating your team's week, start a chat with us on the site, or book a call through the Applied Labs page. Bring the task and a few real examples, and we'll tell you whether ten days is enough to show it working.
