Forward-Deployed Engineering04
The pilot that stalled,connected, secured, scored,and in front of real users.
We take a pilot that worked in the demonstration and do the work between it and production: connectors, a security review, evaluation, monitoring, a rehearsed go-live. The criteria are written in week one and met on record.
- Duration
- Eight to twelve weeks
- Starts with
- A pilot that exists
- Week one
- A gap report against the criteria
- Ends with
- Go-live criteria met, on record
FDE / 01Fit
Whether it is a rescue
Signs of a stalled pilot
- It runs on a sample export, not on the system of record
- It impressed the steering group and no one has used it since
- Nobody can say how often it is right, because nothing scores it
- Security has not reviewed it, so it cannot be given real access
- The person who built it has moved on and the code has no reviewer
When it is not a rescue
- There is no pilot yet; the Production Sprint starts from the workflow instead
- The pilot showed the workflow is not worth automating; that is a finding, not a fault
- The problem is a research problem no model yet solves
- No one on your side owns the outcome
- The pilot must stay a demonstration until a decision that has not been made yet
FDE / 02The hardening checklist
Four things a pilot lacks.
Almost every stalled pilot is missing the same four. The engagement is the work of adding them, in this order.
Integration
The pilot reads from the system of record and writes back to it, under a person's permissions, with every action attributed. The sample export is retired.
Security
A threat model, a review by your security team, secrets in your vault, egress controlled. The pilot is given real access only after the review is signed.
Evaluation
A golden set of real cases with known answers, agreed with the people who do the work, and a harness that scores every release against it. The score is the argument for going live.
Monitoring
Dashboards, alert thresholds and a person they page. A rollback that has been rehearsed before it is needed.
FDE / 03Week by week
W1 – W10
Ten weeks, drawn as a plan.
The common shape. A pilot with more systems to connect or a longer security review uses weeks eleven and twelve.
The go-live criteria are signed in week two. Nothing is added to them afterwards, and nothing is waived.
FDE / 04Go-live
6 ITEMS
The go-live criteria.
Written in week two, signed by the owner, and ticked one by one before real users see the system.
- Connected to the record
Reads from and writes back to the system of record; the sample export no longer exists.
- Scored on the golden set
The release passes the agreed threshold on real cases with known answers.
- Security review signed
Threat model reviewed, findings closed, sign-off from your security team on file.
- Monitoring live
Dashboards and alert thresholds in place, with a named person they page.
- Rollback rehearsed
A rollback performed on the production path by your engineers before go-live, not after.
- An owner trained
One named person on your side who has released and rolled back the system.
FDE / 05Specified
10 FIELDS
The engagement, specified.
The terms as they stand before the first call. The pilot moves the detail; it does not move the shape.
- Duration
- Eight to twelve weeks, set at scoping by the systems to connect and the review to pass
- Start
- A scoping call, a read of the pilot's code, then the go-live criteria agreed with the owner
- Engineer
- One named senior engineer; a second for the integration weeks where the systems need it
- Input
- The pilot as it stands: its code, its prompts, its sample data, and whoever built it if they are still there
- Cadence
- Your standups daily; a written report weekly; a demonstration on real records at each gate
- Access needed
- Repository, the systems of record, real records, and your security team's time for the review
- Evaluation
- A golden set by week four; every candidate scored against it, as in Evaluation & Observability
- Go live
- Real users and real traffic, once every criterion on the sheet is ticked by your side
- Handover
- Runbook, decision record and a rehearsed rollback, in the last two weeks
- If it slips
- For reasons on our side, we stay until the criteria are met at no further fee; for access or review on yours, the clock pauses and the gap is written down
FDE / 06Proof
3 ROWS
Case studies
Illustrative engagements written end to end, chapter by chapter.
- 02RestaurantsPhone ordering by voice, in the caller's language49% of order calls completed with no person; peak calls answered in 20s 68% → 95%
- 05ManufacturingVisual inspection of machined castingsCustomer-reported defects 41 → 11 ppm; false holds 3.6% → 1.3%
- 14InsuranceMotor claims first notice of loss and triageComplete file with a handler 3.8 → 1.1 days; no claim denied by the system
FDE / 07Questions
6 QUESTIONS
Asked about a rescue.
Do you rebuild the pilot or keep it?
We keep what holds up and replace what does not. The gap report in week one says which is which, line by line, before any code is changed. Most pilots keep their core and lose their sample data, their secrets in code and their lack of tests.
What if our pilot was built by another vendor?
That is common. We need the code, the prompts and any documentation; we do not need the vendor. Intellectual property in the pilot has to be yours for the work to proceed, and we check that at scoping.
Who writes the go-live criteria?
We draft them in week one from the gap report; the owner of the outcome on your side signs them in week two. They are the sheet on this page, with the thresholds filled in for your workflow.
How do you know it is right?
It is scored against a golden set of real cases with known answers, agreed with the people who do the work. The score is measured before every release and is the argument for going live. Nothing goes in front of users on the strength of a demonstration.
Can it be deployed inside our own perimeter?
Yes. Where the pilot must run in your cloud account, on your premises or without a network, the environment work is scoped with Regulated & Secure Deployment and runs inside the same weeks.
What happens after go-live?
Your team runs it, with the runbook and the decision record. Our engineer stays on call for an agreed period. Managed Operations exists if you would rather we keep running it.
FDE / ENDStart
Eight to twelve weeks,and the pilot is in production.
Book a scoping call and bring the pilot; you leave knowing what stands between it and real users.