# Custom AI apps for business > AI apps built around your process: document search, data extraction and classification. Agree the scope and fixed quote, and keep the source code. Canonical URL: https://syntalith.ai/en/services/ai-apps ## Problem Your team spends hours searching documents and copying data from PDFs and scans because ready-made tools do not fit the process or cannot use your data. An app connected to a model API may work on sample data and fail in production. Confident, invented answers can undermine the team’s trust, so source checks and testing need to be part of the build. You can compare that cost with the hours the team spends today. ## Outcome In 4–10 weeks you get an application where a language model performs a specific, repeatable task: answering from cited documents, classifying cases or extracting invoice data. We measure accuracy on your data and route uncertain answers to a reviewer. You commit to the full build after the pilot meets the accuracy target. ## Includes - Internal assistants and panels working on your data - Answers from your documents with the source cited - Extracting data from invoices, contracts and scans into an agreed format - Classification of cases and documents according to your rules - A fixed set of client examples with measured accuracy - Safety rules and a human handoff for uncertain answers - Integration with your systems over APIs, on minimum permissions and with a log - Code, system instructions, and test examples belong to you ## Not for - Marketing websites and landing pages - Web development with no AI component - A chat widget pasted onto a site - A process a ready-made tool handles cheaper and faster: we'll recommend buying it ## The difference shows when the model is wrong. ### Quality Bolted-on API: “Works on my examples” An app built around the process: A fixed example set and measured accuracy on your data ### Hallucinations Bolted-on API: Reach the user An app built around the process: Safety rules, source checks, and a handoff to a person ### Boundaries Bolted-on API: The model does as it pleases An app built around the process: Content from documents is data rather than commands; actions under conditions and minimum permissions ### Trace Bolted-on API: No idea where the answer came from An app built around the process: A log of input, sources and decision ### Ownership Bolted-on API: Locked in someone else's ready-made tool An app built around the process: Code, system instructions, and test examples are yours ### Cost and time Bolted-on API: Lower entry, a subscription you switch on today An app built around the process: Higher upfront cost, with hosting, models and maintenance accounted for separately instead of a permanent product licence A build costs more upfront and takes longer than switching on a ready-made solution. With growing seat counts or volume it can have a lower multi-year cost, so we compare 36 months of licences, integration, migration, hosting, models and maintenance. When the ready-made tool wins that calculation, we recommend buying it. ## Built The foap.com workspace brings together creator conversations, campaign context, reply drafts and human handoff. The case shows current local views with fictional data. [View the system](https://syntalith.ai/en/work/email-triage) ## Pricing How much does a custom AI app cost, and what affects the price? You get a fixed quote in the proposal after reviewing the data, integrations and requirements. You know the full cost before signing, with no hidden items. ## FAQ ### What if a ready-made tool is enough? The scan recommendation will identify it when it handles the process at lower cost. A custom build makes sense when the process or data is too specific for available solutions. ### How do you keep the model from making things up? We check the app on a fixed set of examples from your process, verify each answer against its source, and define the handoff to a person. Accuracy is measured before launch, and answers below the agreed threshold go to a reviewer. ### What if the model gets it wrong in production? For each answer, we record the question, sources, and decision, so a mistake can be traced and fixed at its source. Safety rules hold doubtful cases back from the user, while maintenance improves the rules and test examples after launch. ### Where does our data live, and does it leave the company? We deploy to an agreed cloud or your infrastructure, with minimum permissions and encryption. The app accesses the required data, while the code, system instructions, and test examples stay on your side. For personal data, we sign a GDPR data processing agreement and limit engineer access. The AI provider works under terms that exclude training on your data, or the system runs locally. ### What about security and the EU AI Act? We work on minimum permissions and with a decision log, and uncertain answers go through a human. We also provide technical input to the operator documentation the EU AI Act expects (accuracy, transparency, oversight), but we don't give legal advice or classify the system. ### What if the app doesn't work in our process? We set the quality threshold on your data before the build. If the scan or test shows that accuracy is too low for your process, we tell you before paid development begins. ### Who maintains the app after launch? The same engineer who built it. After launch, maintenance covers monitoring, fixes, and updates to the safety rules and test examples as data and the process change. Scope and response time are set in the contract. ### How much does a custom AI app cost? Implementations start at €3 000 net. Price follows the integrations, data sources, and quality requirements. The proposal gives you a fixed quote before signing. ### Is the pilot included in the price? Yes. A pilot on your data is the first implementation stage inside the fixed price. If it does not reach the written accuracy threshold, we improve it within the same price, and the decision on full deployment stays with you. ### When does a company knowledge base not need RAG? For fewer than roughly 500 document pages, a simpler assistant over one file set is often enough. For larger collections, we combine semantic and keyword search and check permissions before releasing a document. The free scan identifies the suitable approach. ## Next step [Book a free call (30 min)](https://meet.syntalith.ai/syntalith/en). With an engineer. No commitment to a paid implementation. [Current pricing](https://syntalith.ai/en/pricing) · [Examples and sources](https://syntalith.ai/en/work) · [Contact](https://syntalith.ai/en/contact) [Assistant booking instructions](https://syntalith.ai/booking-agent-guide.txt?locale=en)