This guide is for CTOs, operations leaders and CFOs who are about to sign up for forward deployed engineering (FDE), where a senior engineer works inside your team and your systems. You want to know what the first three months look like and what the engineer will need from you.
The calendar below is how we plan it at Kaprin. Treat it as a plan. Your systems, your approvals and your data quality will move the dates in either direction.
What does the Kaprin rhythm look like?
Four beats: Scope within 24 hours, Embed on day 1, Harden in weeks 2-6, and Handoff by day 45. Most engagements put the first workflow into production within 30-45 days. The longer 90-day view adds a second workflow and the final transfer of ownership.
- Scope within 24 hours. One call, one workflow, one written scope with a baseline metric and an exit definition.
- Embed on day 1. The engineer joins your chat, your tickets and your environments, usually as part of a small pod.
- Harden in weeks 2-6. The thin slice becomes production software, with idempotency, locks, validation, monitoring and audit logging.
- Handoff by day 45. Runbooks, architecture decisions and credentials are in your account, and your engineers have shipped changes alongside us.
Engagements run as fixed-scope sprints of 4-6 weeks with a live demo every week. The full process is on our how we deploy page.
Why does the 90-day view differ from the 45-day view?
Day 45 is when the first workflow is in production and handed off. Day 90 is when a second workflow has been built on the same foundation and your team has carried most of the load. The first number proves the work. The second proves the handover.
We say this plainly because speed is easy to oversell. Getting one workflow live in 30-45 days is realistic for a well-scoped problem. Owning it for years takes a different kind of evidence: your own engineers shipping a change on a Tuesday with nobody from us in the room.
What happens week by week?
The table shows what the engineer does, what you'll see, and what we need from you in each window. Weeks 1-6 cover the first workflow and its handoff. Weeks 7-13 cover the second workflow and the transfer of ownership.
| Window | What the engineer does | What you see | What you must provide |
|---|---|---|---|
| Hours 0-24 | Scopes one workflow, writes down the baseline metric and the exit criteria. | A written scope you can approve or reject. | A 60-minute call with the people who run the workflow today. |
| Week 1 | Gets access, maps the workflow, lists failure cases, builds a first thin slice on real data. | Demo at the end of the week and a written plan for week two. | System access, sample data, a named decision-maker. |
| Weeks 2-3 | Replaces the prototype's weak points: validation, retries, locking, error handling, spend caps if models are involved. | Weekly demos that now include failure cases being handled. | Fast answers on edge-case rules. Who approves what. |
| Weeks 4-5 | Adds monitoring, audit logging and an evaluation suite. Runs the workflow in parallel with the manual process. | Dashboards, plus a side-by-side comparison against your baseline. | Staff time to check outputs against the manual process. |
| Week 6 (by day 45) | Cuts over to production. Writes runbooks and decision records. Pairs with your engineers on a first change. | First workflow live. A handoff pack in your account. | An owner on your side and cloud account permissions for them. |
| Weeks 7-10 | Builds the second workflow on the same foundations, with your engineers writing more of the code. | Faster demos. Fewer surprises. | The next priority, chosen by your decision-maker. |
| Weeks 11-13 | Steps back to review and support. Closes open items and confirms ownership. | A team that can run and change the systems without us. | A sign-off against the exit criteria agreed at the start. |
What should you expect in week one?
By Friday of week one you should see a thin slice of the workflow running on real data, a short list of failure cases we found, and a written plan for week two. Week one is for reducing uncertainty. Nobody should expect a polished product yet.
We wrote a day-by-day version of this week in why AI pilots stall and what an embedded engineer does in week one. The short version: access on day 1, a mapped workflow and an agreed baseline on day 2, a failure inventory on day 3, a thin slice on real data on day 4, and a demo with a plan on day 5.
What does hardening actually involve in weeks 2-6?
Hardening turns a working demo into software that survives real traffic and bad inputs. It covers retries that can't double-process, locks around shared state, schema-validated model outputs, spend limits and audit logs. When it works, the system fails in predictable, recoverable ways.
Our technical stance is deterministic scaffolding around non-deterministic models. If a plain script does the job, we use the script. If a model is needed, we put idempotency keys, validation and circuit breakers around it. The details are in how we harden a prototype for production.
A real client example helps here. In the Slender Morris case study, the work turned PDF purchase orders into entries in a Cin7 ERP and onward to Machship freight, and it runs around the clock. The reported result was 90% faster order entry.
What do you owe the engineer?
Access, a named decision-maker and sample data decide whether an engagement moves at full speed. If you can prepare only one thing before day 1, make it access. We can work around the rest for a few days. We can't work around missing access.
- Access means accounts in the relevant systems (ERP, CRM, payment provider, cloud console), a development or staging environment, and a way into the repository or the chance to start a new one in your account. Ask your security team early, because this is the slowest item.
- A decision-maker is one person who can say yes, no or not yet within a day. A committee that meets monthly adds weeks.
- Sample data should be a few hundred real records, including the messy ones: the rotated PDF, the order with three addresses, the refund that was issued twice. Anonymize them where policy requires it.
- Someone who does the work today and can explain the exceptions nobody wrote down.
- A baseline. The current number, even if it's rough: minutes per order, cost per ticket, hours per report.
- An owner for after, on your side, who will hold the pager, the credentials and the backlog when we leave.
What if the engagement runs behind?
Slow access, unclear ownership and data that looks nothing like the sample are the usual causes. We raise it in the weekly demo, cut scope to protect the date, and move the dropped item to the second workflow. A smaller workflow in production beats a bigger one stuck in review.
We'd rather cut scope than quietly extend the calendar. If delays pile up on your side, we'll say so in writing, and we expect you to tell us when the plan no longer matches reality.
How do you know it is working?
Judge it against the baseline you agreed in week one. If orders took two days and now take minutes, or a report took four hours and now takes ten minutes, that's your evidence. Every demo and dashboard should tie back to that one number.
It's also why our approach starts with a number and picks the technology later. And since handover is a planned phase, read how an FDE engagement is supposed to end before you sign.
Next step
If you'd like to check whether a specific workflow fits this rhythm, tell us about it. We'll reply with a scope, or an honest note that it's a poor fit.
