The tag view groups articles by a concrete operational problem or implementation type, without exposing hundreds of raw frontmatter labels.
ChatGPT helps join exports from your systems, find mismatches and prepare a report in a spreadsheet. Start with one report, for example weekly sales by branch. Check the totals before and after joining the tables, and that the formulas recalculate after the data changes. At 8 reports a month cut from 3 hours to 1, €491 a month is left after licences for 3 people.
A ChatGPT rollout for a team of up to 10 people costs from €1,150 net at Syntalith, paid once, and takes 2 weeks. You buy ChatGPT Business licences directly from OpenAI: $20 per person a month with annual billing. The first year for 10 people comes to about €3,292, which is €27 per person a month.
A ChatGPT pilot measures finished work: accepted quotes, reports or replies. The number of messages in the tool shows only use. Set a unit of work and a sign-off criterion, then time the current method and the work with ChatGPT, corrections included. In a scenario of 200 tasks a month at 30 and 18 minutes, €1,008 a month is left after process care and licences.
Choose a ChatGPT project for a case that develops: a quote for one client with further agreements. Choose a custom GPT to share the company's way of doing a task with a whole department. Keeping these needs apart makes documents easier to maintain and limits mixing of client data. Neither one needs an app built on the API.
In a small business, a ChatGPT rollout starts with one task someone can check quickly, for example a draft quote from a brief. Each employee gets their own company account, and the owner decides which data may be entered. For 5 people licences cost $125 a month, and the rollout from €1,150 net. If each person saves 1.5 hours a week, the whole thing pays back in about 2 months.
ChatGPT helps a salesperson get from a customer's enquiry to a draft quote: it sorts the requirements, picks service descriptions and points out the information needed for pricing. The condition is one source of prices and one person who approves discounts. At 120 quotes a month, cutting the work from 45 to 30 minutes gives 30 hours. A rollout for a team of up to 10 people costs from €1,150 net and pays back in 1.3 months.
Claude Cowork goes through several steps of work on files and prepares a result for checking, for example a weekly report on project statuses. It works on paid plans, and access to local files needs the Claude Desktop app. For 5 people Claude Team costs $125 a month, and the rollout from €1,150 net. At an hour saved per person a week, the payback takes 1.8 months.
Codex analyses code, prepares changes, runs commands and checks for errors. The rollout starts in one repository, with tasks that have a clear finish criterion, and ends with timing from the start of a task to the change being accepted. Five ChatGPT Business Standard seats with Codex cost $125 a month. A training day from €1,150 net pays back in 0.8 months at 2 hours saved a week.
The OpenAI API brings a model into a process run by the company's software. When an enquiry arrives, the system reads the requirements, checks the catalogue and saves a draft quote for approval in the CRM. At Syntalith such an integration costs from €3,000 net. At 400 enquiries a month the GPT-6.1 Sol model costs about €11, and the implementation pays back in 1 month.
At a wholesaler, an AI agent completes a request for quote before pricing: it finds the customer's earlier order, recognises missing parameters such as the unit or the variant, and writes questions for the salesperson. It adds no technical parameters of its own. The agent costs from €3,000 net at Syntalith and, with 300 requests for quotes a month at 12 minutes each, pays back in 2.1 months.
An AI knowledge base in maintenance finds the procedure for a specific machine and situation, shows its version and source, and leaves decisions about work on the machine to approved procedures and authorised people. An AI application on company documents costs from €3,000 net at Syntalith. At 500 technicians' questions a month and 12 minutes of searching each, it pays back in 1.0 months.
You consider your own supplier portal when access fees grow faster than the value of the features you use, and the company needs a simple flow: the supplier uploads a document, sees its status and responds to a gap. The comparison covers the full cost of running it, upkeep, migration and user support.
Fine-tuning helps classify tickets when a general model regularly confuses the company's specific categories. First check whether people understand the categories the same way and whether past assignments show the correct decision. If tickets went to whichever team had time, the history teaches the model who was available.
Low use of AI licences comes from missing skills, the wrong tool, no access to data or the way work is organised. First check what stops people on a specific task. With 40 seats at $25 and 4 in 10 unused, the company pays €357 a month for empty accounts. A rollout for a team of up to 40 people costs €2,300 net at Syntalith and pays back in 1 month when 16 people save an hour a week.
A local AI model suits confidential proposals when data rules rule out your current tool. The decision covers quality on your documents, user access, performance and responsibility for maintenance, and buying a server is only one of those elements. At Syntalith an app on a local model costs from €3,000 net plus GPU hardware. With 40 proposals a month at 120 minutes each, it pays back in 1.7 months.
Supplier statement reconciliation automation shows matching items and differences with an explanation: a payment sent but not yet settled, a credit note with a different reference, or one amount covering several invoices. Rules compare the identifiers and amounts, and the model helps only with unusual descriptions. At Syntalith the automation costs from €3,000 net, and at 150 supplier reconciliations a month of 25 minutes each it pays back in 1.8 months.
Codex connects to a custom model provider through the Responses API, so it also works with a local Qwen3.8-27B on vLLM. In our experiment on an RTX 3090 the connection worked, but the code fix from Codex scored 87/100, while the one from Qwen Code on the same server scored 100/100. A working connection says nothing about the quality of the changes.
You test a local coding agent on a task written like a ticket, in a clean copy of the repository, with public and hidden tests. In our CSV export test all four runs passed 3 of 3 public tests, but two passed only 1 of 3 hidden ones. The fastest rejected run was 5.55 seconds quicker than the accepted one.
After several failed attempts, we kept Qwen Code with medium reasoning effort set in the client and in the request, preserved thinking, a 150,000-token context and no fixed limit on response length. This configuration comes from one RTX 3090 card and one active agent, with a short profile test after every change.
Qwen3.8-27B found the three facts asked for in both a 50,059-token and a 230,085-token input, and the shorter run took 59.3 s while the longer took 563.4 s. You size the context window to the 95th percentile of real input length plus a margin, because a bigger window costs time without a better answer.
Next step
Walk an engineer through it in 30 minutes. Within 2 business days you get, in writing, what it costs, what to do first and the starting price.
Times are shown in your own time zone. We work with clients across time zones.
See where to startOpenAI makes ChatGPT and Anthropic makes Claude. We belong to the partnership programs of both companies, so we work with their tools every day and learn from the creators themselves.
Claude Partner NetworkParticipation in programs does not mean that OpenAI or Anthropic endorses our services.