What is forward-deployed AI?
Forward-deployed AI is AI that a team builds into the way your company already works. The team connects it to the systems you use and looks after it, and your own people approve every action it prepares.
Biissa team · Updated 16 November 2026
In short
- A team, not a login: the provider’s people map the job, connect the systems and look after it.
- It works inside the systems you already use, such as WhatsApp, spreadsheets and accounting software.
- The AI prepares the work. A named person on your team approves it before anything is sent or recorded.
- It starts with one job and grows one job at a time, on the same connections.
Where the name comes from
The name comes from the forward-deployed engineer: an engineer who works with a customer on site, inside the customer’s real problems, instead of at headquarters. Palantir made the role known, and AI companies now hire for it too.
Forward-deployed AI applies the same idea to a service. The team comes to your work, instead of asking your team to come to a tool.
How it differs from an AI tool or a consultant
- Who sets it up
- An AI tool you set upYour staff
- A consultantYour staff, after the report
- Forward-deployed AIThe provider’s team
- Where it works
- An AI tool you set upInside the tool
- A consultantOn paper first
- Forward-deployed AIInside your own systems
- What you get
- An AI tool you set upA licence
- A consultantA report and a plan
- Forward-deployed AIOne job running, then the next
- Who keeps it working when things change
- An AI tool you set upYour staff
- A consultantYour staff
- Forward-deployed AIThe provider’s team
- Fits best when
- An AI tool you set upThe job is simple and your team has time to set it up
- A consultantYou need a strategy before choosing anything
- Forward-deployed AIThe job crosses systems and people, and you want it running, not only planned
| An AI tool you set up | A consultant | Forward-deployed AI | |
|---|---|---|---|
| Who sets it up | Your staff | Your staff, after the report | The provider’s team |
| Where it works | Inside the tool | On paper first | Inside your own systems |
| What you get | A licence | A report and a plan | One job running, then the next |
| Who keeps it working when things change | Your staff | Your staff | The provider’s team |
| Fits best when | The job is simple and your team has time to set it up | You need a strategy before choosing anything | The job crosses systems and people, and you want it running, not only planned |
How it works at Sambung
Free
A first chat
You tell us about one job: where the data comes in, who decides, and where it ends up.
Free · about 10 working days
Cek Alur Kerja (workflow check)
We follow the real work from data in to decision, then tell you honestly whether AI can help.
Your past data
A test, with nothing sent
The AI prepares drafts from old data. Nothing is sent or written into your systems.
4 weeks
A pilot of one job
Your team approves every draft, and the results are measured with your own reports.
Then
The next job
It runs on the same connections as the first.
What a job can look like
- SalesExample
A salesman’s order recap on WhatsApp becomes a draft sales order in Jurnal.
The order admin approves
- PurchasingExample
A photo of a supplier’s receipt becomes a purchase bill matched to the purchase order, with a price difference flagged.
The purchasing team approves
- FinanceExample
Overdue invoices in Jurnal become reminder messages, with the ones already paid struck off.
The finance team approves
- HRExample
Attendance records and overtime forms become an overtime summary, with incomplete records flagged.
The HR team approves
These are examples, not customer stories.
When it is not the right fit
- The job happens a few times a month. Doing it by hand may cost less than setting anything up.
- Nobody owns the decision. Forward-deployed AI keeps a person in charge of each action, so someone has to be that person.
- There is nothing to read from or write to. The work needs data that exists somewhere, even if it is only a chat or a spreadsheet.
- You only want answers about your documents. A ready-made AI assistant may be enough.
Questions to ask any provider
- Who approves each action, and can we see what was approved?
- Where do the numbers come from: our systems, or the AI’s guess?
- What personal data does the AI see, and is it replaced with codes before the model reads it?
- Which companies process our data, and in which countries?
- What happens when the AI is unsure?
- Who fixes it when one of our systems changes?
Common questions
Is forward-deployed AI the same as AI consulting?
No. A consultant usually delivers advice and a plan. A forward-deployed team builds the job inside your systems, runs it with your people approving, and keeps it working afterwards.
Does forward-deployed AI replace staff?
It changes their work. The AI prepares drafts, and the people who already do the job check and approve them and handle what the AI flags.
How long until the first job runs?
At Sambung: a free workflow check of about 10 working days, a test on your past data, then a 4-week pilot of one job.
What does it cost?
At Sambung the first chat and the workflow check are free. Pilot and monthly prices are discussed after the check, once the job is clear.
Which systems does Sambung work with?
Sambung starts with WhatsApp Business, Excel, Google Sheets and Jurnal (by Mekari). Other systems, such as Accurate or a point-of-sale system, we check one by one before promising anything.