Read the input
An email, a PDF, a form with free text, or a support ticket, turned into fields your systems can store.
SiteFusion is an AI automation agency for founders and operations leads in the US, UK, EU, Canada, Australia, the UAE, Saudi Arabia and Qatar. We build AI automation services for business teams: software that reads an email, a PDF, a form or a ticket, applies your rules, and writes the result into the systems you already use.
Most useful AI automation is ordinary software with a language model in one or two steps. The model reads or classifies something that arrives unstructured. The system around it creates a record, routes a request, drafts a reply for someone to approve, or flags an exception.
An email, a PDF, a form with free text, or a support ticket, turned into fields your systems can store.
A CRM record, a routed request, a drafted reply, or a flag on the exception a person should see.
The first version sends output to someone for approval, and logs every correction.
A record of what the automation did, what a person changed, and where it is failing.
Budget for the integrations, the rules around the model, the screen where a person checks the output, and the logging. The model call is one step in that system. If a proposal is mostly about which model will be used, ask what happens around it. Our workflow automation guide covers the processes teams usually automate first, and AI explained covers the terms you will see in a proposal.
Farita, our AI phone agent, is one product in this family. It answers business calls, collects the job details a team needs, books the job, and calls customers back about quotes waiting on a decision.
If the process is simple and both systems are common products, a feature inside software you already pay for, or an off-the-shelf integration, may be enough. An agency makes sense when the inputs are messy, when the rules are specific to your business, when the data is sensitive and you need control over where it goes, or when the automation has to sit inside an internal tool or a customer portal. The build vs buy guide goes through that decision. A good agency asks this too, and will sometimes tell you a tool you can buy is enough.
An AI workflow automation company should be able to name the process, the systems it touches, how you will measure it, and what the running cost looks like. Our post on measuring AI workflow automation shows how to set a baseline before the build starts. For document-heavy work, the document workflow playbook goes further.
We start with one high-volume task, written down step by step, including the exceptions. A proposal to automate the whole back office in one go is a reason to slow down.
Models make mistakes, and the mistakes are often confident. The first version sends output to a person for approval. The review step comes off a category only once the error rate is known and acceptable.
You see a working version on real documents or requests early, while a change is still small. A fixed price before anyone has looked at those samples is a guess.
Model providers change and retire models, inputs drift, and the bill moves with volume. Someone reviews corrections and updates the prompts and rules. That can be us, or your own team.
Every model call costs money, and the bill grows with usage. Ask for an estimate of cost per request, the worst case, and the limits that stop one account or one bad loop from running up the bill. We have written about how to keep LLM feature spend under control.
Ask where documents and prompts are sent, whether the model provider can train on them, how long anything is stored, and who can see logs. For buyers in the EU and UK this connects to GDPR. Buyers in the Gulf may have their own data residency rules. You want specific answers in writing, and you still need your own legal advice.
Model provider accounts, cloud hosting, integration keys and the code repository should be in your company's name. If the agency leaves, the automation should keep running. The dev partner contract checklist lists the clauses to look for.
Ask for automations that run in production, and speak to a client if you can. You can see shipped work on the portfolio.
You should get a clear answer on the stack. At SiteFusion we mostly build with Next.js, React, Node.js and PostgreSQL, as described on the full-stack development page.
AI automation needs attention after launch. Plan for someone to review error logs and corrections, and to adjust prompts and rules. We describe how we handle the period after launch on the post-launch support page.
If the automation has to live inside a product you sell, see SaaS MVP development. If it is software your own staff use every day, see internal tools development.
More of these are in dev partner red flags and the guide to hiring a software development partner.
It designs and builds software that uses an AI model for one or two steps in a business process, such as reading a document, classifying a request, or drafting a reply, and connects that software to the systems you already use. A good agency also sets up review, logging and monitoring, and can support the automation after launch.
There are two costs: the build, and the running cost. The build depends on how many systems are involved and how messy the inputs are. The running cost includes model usage per request, hosting and paid integrations, and it grows with volume. Ask for both before you sign. Any number given before someone has seen real samples of your documents or requests is a guess.
It depends on the integrations and how much sample data is available. A first version for one process is usually a matter of weeks, with time added if access to systems or data takes a while to arrange. Ask for a plan that puts a working version with real data in front of your team early.
It can be, if it is set up with care. Check where documents and prompts are sent, whether the provider can use them for training, how long anything is stored, and who can see logs. Some providers offer business terms or regional hosting, and some workflows can strip personal data before it reaches the model. If you are in the UK or EU, ask how the setup supports your GDPR obligations. In the Gulf, ask about local data protection rules and whether data has to stay in the region. You still need your own legal advice.
For a first automation, an agency brings a team that has done similar work before. Once you have several automations running and a steady stream of changes, hiring someone in-house can make sense. Many companies do both: an agency builds the first version and hands it over with documentation.
Yes. That is often where this work fits, because the review screen and the business rules live in the same place as the rest of the tool. We build those tools as part of our internal tools and full-stack work.
If you have a process in mind and want a second opinion on whether it is a good candidate, book a call. If an off-the-shelf tool would do the job, we will say so.