How long automation implementation takes, walked through one working week

"How long until an agent like that is actually running here?" Anyone asking this tends to picture a project plan with phases, a steering committee and milestones. How long automation implementation takes at Sophera Consulting looks rather different. An agent is built in one to two days. Testing on real cases happens in the same week, and the pilot usually starts within that first week too. That reads like a brochure line, so this article walks through such a week day by day.

The example is made up. It shows how we would set up a week like this and does not describe a client job.

How long automation implementation takes, and why days are enough

Scope does most of the work. We build one process, not a programme for the whole company. An order entry agent has one inbox, one target system and a manageable set of rules. You can build that in a day and check it the next. Larger jobs are split into sections, and each section is up and running after days. As a general range, a build takes between one and 14 days.

There is also no separate process analysis sold as a project of its own. Our way of working has three steps: selection, offer, pilot. How the process runs today and which access we need gets settled during selection. Then you receive a fixed-price offer with a cost-benefit analysis, and the next date in the calendar is the build day.

On that day somebody from your department sits at the table with us. That matters more than it sounds. A question like "what does express delivery actually mean in your house?" gets answered the moment it comes up, instead of travelling through three inboxes as an open item. Order entry from customer emails is something we build in a few hours. With one person from the department in the room, it is done in a day.

The example: a technical wholesaler and its order inbox

Picture a technical wholesaler with 45 staff receiving around 150 orders a day by email. Some customers send a PDF from their ERP. Others type three lines of part numbers into the email body, and now and then someone attaches a photo of a pump's rating plate. Four people in inside sales key all of it into the ERP. Selection landed on exactly this process, the offer is signed, and the build day is set for a Monday.

Settled before Monday

Four things are ready before we start. IT has set up read access to the order inbox and access to the ERP's API, in the test environment for now. Inside sales has put aside 40 real order emails from the last few days, exactly as they arrived, with a note next to each on what ended up in the ERP. A clerk who has known that inbox for years keeps her Monday free. And someone has been named to decide when a house rule is disputed. In our example that is the head of inside sales.

We bring the data processing agreement with us, it is part of our standard. Hospitals will also want their data protection impact assessment and the conversation with the staff council out of the way first. We sort that out together beforehand, so the build day stays a build day. The article on prerequisites covers the details.

Monday, hour by hour

8:30. We go through the 40 emails with the clerk. What kinds are there? Orders, amendments, price enquiries, complaints. What has to be on an order so the warehouse can pick it? House rules get written down as we go. "As soon as possible" means next working day. If an email carries the customer's own part number, the customer's mapping list is checked first. If a part has been discontinued, the agent proposes the successor and does not swap it in by itself.

10:00. The agent reads its first emails. It works out whether the message is an order at all, matches the sender to a customer record and pulls out line items, quantities, delivery address and the customer's PO number, wherever they happen to sit.

13:00. Master data comes next. Part numbers are checked against the item master, delivery addresses against the addresses on file. The clerk dictates the sanity rules: a customer who normally orders 50 and writes 5,000 today probably means the pack size. The agent does not guess in cases like that. It flags the field and says why.

14:30. First drafts appear in the test ERP with the status "to be checked". Each field notes where its value came from, and the original email is attached.

16:00. All 40 sample emails go through and the clerk compares them with her notes. Say 31 drafts are fully correct, six have gaps that were rightly flagged, and three are wrong. Twice the agent took a customer part number for one of the wholesaler's own, and once it read an amendment as a new order. Both causes are fixed that afternoon, then the whole set runs again.

By the evening the clerk is looking at a list of drafts in the screen she uses every day. Nothing has been booked and no customer has received anything. What she has seen is an agent following her rules. She dictated them, after all.

Tuesday and Wednesday: real cases

From Tuesday the agent reads the live inbox alongside the team. Inside sales carries on as usual while the agent puts its drafts on a review list. In the afternoon we spend half an hour comparing what the person entered with what the agent produced.

Cases turn up that were not among the 40 samples. In our example one customer sends combined orders for three building sites in a single email. Another attaches that rating plate photo. The agent reads the plate but marks the item match as uncertain. Each of these gets a rule.

On Wednesday the changed rules run against the 40 old emails plus everything that arrived on Tuesday. That way a fix cannot quietly break something that was already right on Monday. The set of cases grows every day and later becomes the basis for maintenance.

Thursday: going live, with sign-off

In our example the pilot starts on Thursday. The head of inside sales makes that call herself, based on the comparisons from the two days before. The agent now creates its drafts in the real ERP. Inside sales stops typing and starts reviewing.

Two lists make this quick. One holds drafts where every required field is filled and every match is certain, the other holds drafts with flags. The order confirmation goes to the customer only after release. So going live here means the agent handles real orders while a person still gives the commitment to the customer.

Friday: look at the numbers, settle operations

By Friday inside sales has two days of pilot behind it. Together we look at how many drafts were released unchanged and which fields needed correcting most often. These are your own numbers from your own inbox, and you can hold them against the cost-benefit analysis in the offer.

For everything after that, each automation comes with a maintenance agent, included in the fixed price. It monitors operation, checks the input and output of every interface for changes, tests adjustments against the old cases and handles the move to new AI models. Whatever it cannot solve goes to your IT as a ticket. Running costs are purely the usage costs of the AI models, billed directly to you. The article on costs does the maths.

What the start date depends on

Building is the short part. When that Monday can happen depends on the four things above: access, real examples, one person from the department and one person who decides. Most of it comes down to an email to your IT and a free day in a colleague's calendar. That is why we talk about it first.

Sophera Consulting builds agents like this individually, for a fixed price and without a subscription, and the result belongs to you. Which of your processes makes a good start, and which access we need for it, is what the free automation check is for. To watch an agent like this work through a single order step by step, read the example on order entry at a freight forwarder.

What we recommend

Pick one process, not three. Put the build day at the start of a week, so testing and the pilot fall into that same week and you have numbers by Friday instead of impressions. Free up the person who knows the process best, even if she is the hardest one to spare. Her day at the table is the reason an idea on Monday turns into an agent working on real orders by Thursday.