Service
FDE outsourcing
We put engineers on your site to wire AI into the systems, permissions and workflows you already run — and hand over a system you can accept.
What we actually do
We translate "what AI can do" into "what your business should do first" — and then build it, integrate it and run it in production. That middle stretch is where most enterprises get stuck.
01
Decide what to build
Business teams feel the pain but cannot specify the requirement; engineering understands models but not the shop floor. We go and watch the process, talk to the people who will use it, and turn a vague pain into a task someone can execute.
02
Build it
Large models, agents, retrieval — combined with your own systems and data to produce something that runs, rather than a demonstration that impresses.
03
Integrate it
Interfaces, permission models, data synchronisation, failure handling. The new system has to live inside your IT environment, not become another island where somebody copies data by hand.
04
Take it to production
Launch, evaluation, monitoring, documentation, and a review three months later. A system is not delivered until people are genuinely using it.
Four engagement models
Choose by how well-defined your requirement is and how much you want to commit. Pricing differs across the four; none of them is a day rate.
FDE assessment
- Duration
- 1–2 weeks
- Best for
- You are not yet sure whether to do this, or where to start
A full assessment on site: the use case, the data, the organisation. Output is a written judgement with a priority order. The conclusion may well be that now is not the time — and that is a useful result.
DeliverableWritten assessment, prioritised use-case list, draft scope and acceptance criteria
Fixed-scope pilot
- Duration
- 3–8 weeks
- Best for
- Proving capability at small cost
One clearly bounded use case, running on real data with real users. Scope is frozen at the start, acceptance criteria agreed in advance, and scale-up decided only after thresholds are met.
DeliverableRunning system, evaluation report on real data, launch plan
On-site delivery (Beijing)
- Duration
- Project duration
- Best for
- Sensitive data, or work that needs face-to-face collaboration
Engineers work from your premises alongside your business and IT teams. Suited to data that cannot leave your network, or requirements that need frequent direct conversation. Available in Beijing.
DeliverableAs the pilot, plus an on-site collaboration process and knowledge transfer
Ongoing partnership
- Duration
- Monthly or quarterly
- Best for
- Several use cases to land in sequence
A steady cadence of work moving use cases into production one after another. Fits teams that have validated the approach once and now want to do this systematically.
DeliverableQuarterly roadmap, continuous delivery, internal capability transfer
Note: outside Beijing, delivery is remote — remote access, screen sharing and scheduled calls. Remote delivery asks more of your side, so we set out clearly what we need from you before starting.
Planned: we are building a network of forward deployed engineers so that one near you can work on site. This is not in place yet. If you have a requirement in your city, tell us — it shapes where we recruit.
Six common scenarios
These recur across manufacturing, financial services, healthcare, the public sector and retail. A fuller list is on the scenarios page.
Customer service and tickets
Agents handling repeat questions, with full context passed to a human on escalation. The hard part is rarely the model — it is the quality of historical ticket data and the rules for escalating.
Enterprise knowledge search
Turning knowledge scattered across documents, wikis, email and chat into something queryable. The hard part is permission isolation: different roles may see different content.
Document and invoice processing
Automatic recognition and structured capture of contracts, orders, invoices and inspection reports. Results are measurable here, which usually makes the return easy to calculate.
Data questions and analysis
Letting business staff query data in natural language instead of queuing for a SQL request. The hard part is consistent definitions and explainable results.
Process automation
Chaining repetitive cross-system operations into an automated flow, removing manual copying and waiting. Fits processes with clear rules but no integration between systems.
Quality and equipment monitoring
Anomaly detection and early warning from production data. These scenarios depend most on data completeness, so the data foundation has to be assessed first.
When to talk to us
We would rather talk you out of a project up front than discover halfway that the direction was wrong. The criteria are explicit.
A good fit
- The requirement is not standard and no existing product covers it — somebody has to define it first
- It involves integrating several existing systems, untangling permissions or consolidating data
- You have already tried once and it did not land; you need someone to find where it sticks and push it through
- There is a gap between the business team and the technical team, and somebody has to speak both languages
- There is a specific business metric to move, not just a wish to "try AI"
Not a good fit
- The requirement is standard and a mature product or SaaS already covers it — buying it is cheaper
- What you actually want is inexpensive headcount for well-defined development — a staffing firm costs less
- No business stakeholder is involved yet; this is a technical team wanting an AI project
- You expect to buy engineers by the day and direct their work yourself
- Budget and timeline cannot accommodate even a minimal assessment and pilot
How to tell whether an FDE provider is credible
This section is a tool for buyers, not a pitch. Use it on any provider you are considering, including us. A provider willing to answer these directly is usually a better partner anyway.
- 01
How much work does your first client take, and how much does your tenth in the same industry take?
This is the sharpest test separating FDE from outsourcing. If the answer is "about the same", the firm is not compounding — you are paying for hours, not method.
- 02
When a project ends, what remains besides the system?
A good answer names specific reusable assets: components, connectors, templates, evaluation sets. A vague answer is usually "full documentation" — documentation is part of delivery, not compounding.
- 03
How do you decide a use case is not worth doing? Give an example of a client you advised against.
A provider that has never talked anyone out of a project either screens clients extremely hard, or takes everything.
- 04
Who sets acceptance criteria, and when are they fixed?
If acceptance is discussed at handover, the project will most likely end in a dispute. Criteria should be agreed in writing before work starts.
- 05
If a key assumption fails halfway, how do we settle?
This exposes the business model. Phased settlement with the option to stop shows confidence in their own judgement; large upfront payment or day-rate settlement puts the risk on you.
- 06
Who owns it after launch, and how fast do you respond when it breaks?
Many AI projects die in the three months after launch — model drift, changing data, nobody maintaining it. Delivery without a named owner is incomplete delivery.
- 07
Do our data and business rules end up in the work you do for other clients?
This question is about the boundary of compounding. General methods and components can be reused; your specific business rules, data and system detail cannot. The answer must be enforceable in the contract.
- 08
Who exactly will do this work, and what have they done like it before?
FDE work depends heavily on the individual. Ask about the people who will actually be assigned and their experience in similar settings, not the firm's overall credentials.
Common questions
How is this different from an ordinary outsourcing company?
Three layers. First, timing: a traditional software outsourcing company usually arrives once the requirement is clear and builds to it; we arrive while it is still vague and define it first. Second, the unit of pricing: an outsourcing company sells time by the day or by headcount; we sell outcomes by scope and phase. Third and most important, what remains at the end: a system that merely runs is outsourcing, whereas turning what was learned — stripped of your information — into reusable components that make the next similar project faster is forward deployed engineering. In one line: not whether anyone sits on site, but what is left behind.
Can you work on site? What about elsewhere?
On-site in Beijing. Elsewhere in China and internationally we work remotely, through remote access and scheduled calls. We are building a network of forward deployed engineers so that one near you can attend in person — that is still at planning stage.
What does a project cost?
We do not quote a day rate. We price by scope and phase, depending on how many use cases are involved, how complex your existing systems are and how usable the data is. An assessment is the smallest commitment and the sensible place to start. See the engagements page.
Which industries have you worked in?
The industry experience we can discuss publicly sits in manufacturing, financial services, healthcare, the public sector and retail. Confidentiality means we have no named client cases we can publish, and we will not dress that up. During assessment you can ask us to walk through comparable situations and the problems we hit.
How do you handle data security?
It has to be settled before work starts rather than promised. We agree the scope of access, the identity used, how it is audited and which data must not leave your environment. Where data is sensitive, working on site inside your network is an option.
Not sure whether your use case is worth doing?
Start with an assessment. The output is a written judgement with a priority order — including, where it applies, a clear recommendation not to proceed.