Prompt Strategy
n8n, Make.com and Zapier automation with AI steps and a human approval check — fixed price, documented on handover.Prompt engineering services turn ad hoc AI instructions into a documented system of goals, context, constraints and output rules. As a Dhaka-based company, we build that prompt strategy for international teams, test it against real inputs, version it and hand over the prompt library, test sets and documentation in accounts you own.
Each prompt is tested on real inputs from your work against written pass and fail criteria, so a change can be judged instead of argued about. You start with a four-day audit, then a pilot you judge on your own test inputs before the library is built.
Every prompt is judged on a written test set. Send one prompt you already use first: the review takes 30 minutes and costs nothing.
You get a written note on what is missing from it and what fixing it would cost, whether or not you hire us.
4 days 01 Audit · $300, up to 10 prompts
2–3 weeks 02 Pilot · $600 for one workflow
With the pilot 03 Library setup · $899 fixed
Monthly 04 Prompt Desk · $999, up to 10 updates
What prompt engineering services include
Our prompt engineering services deliver six things every time. You get a use case and goal map, a prompt architecture, custom prompts, testing against written criteria, a versioned prompt library, and documentation with handover. Optimisation continues against the same criteria, not by feel. You receive a working prompt system, not a list of tips.
Use case and goal map
A written map of the tasks you want AI to do, what a good output looks like for each, and which ones are worth a prompt at all.
Prompt architecture
How each prompt is put together: system instruction, context you supply, task, constraints and the output format the next step expects.
Custom prompt development
The prompts themselves, written for your data and your wording, with the reusable parts split out as templates your team can fill in.
Test set and evaluation
A set of real inputs with the output you would accept for each, plus written pass and fail criteria, so a prompt change can be judged instead of argued about.
Prompt library with versions
Every prompt stored in one place with a version number, what changed, who changed it and which test run it last passed.
Documentation and handover
Written notes on why each prompt is built the way it is, how to change it safely, and a walkthrough for the people who will own it.
Not included: model fine-tuning or training · AI model and application programming interface (API) usage fees, paid by you to the provider · application or agent development, which is a separate build · guaranteed accuracy, ranking or cost-saving figures · scope above the published caps, quoted before the work starts.
Which use cases do prompt engineering services cover?
Prompt engineering services cover content, support, operations and sales: the repeated tasks your team already hands to an AI model. Some teams now call this wider work context engineering, but it still rests on written, tested instructions. Each use case gets its own prompts, rules and test set.
Content
Briefs, drafts, product descriptions and repurposing, each with a fixed structure and a house voice the prompt enforces. For editing the output afterwards, see AI content editing.
Support
Ticket summaries, draft replies, tone rules and escalation wording, with refusal cases written into the prompt.
Operations
Document extraction, classification, routing rules and summaries, returned as structured output another system can read.
Sales
Research briefs, call summaries and follow-up drafts, built from fields your customer relationship management (CRM) system already holds.
How do prompt engineering services test a prompt before it ships?
Prompt engineering services test each prompt on 20 real inputs before it ships. Every output is scored against written pass and fail criteria, each failure is traced to the prompt, the context or a model limit and fixed, and the new version is logged in the library. The sample shows one support prompt going through that test.
Test run · support reply prompt, 20 real inputs
Version 1.3 18 of 20 pass
0102030405060708091011121314151617181920
Passed the written criteria Failed, then traced and fixed
Input 07 · order number missingTraced to the context · fixed in version 1.4
Input 15 · refund asked twiceTraced to the prompt · escalation rule added in version 1.4
Version 1.4 20 of 20 pass
Test method · five steps
- Collect 20 real inputsFrom your own work, including the awkward ones
- Write the acceptable outputOne for each input, agreed in writing
- Run and score the promptAll 20 inputs, against those criteria
- Trace and fix each failureTo the prompt, the context or a model limit
- Ship the test set with the promptLogged in the library with the version number
Accepted by Abc Xyz against the agreed criteriaVersion 1.4 logged in the prompt library with its test run
When a model version you use changes, the Managed Prompt Desk re-runs the same test set and reports which prompts behave differently.
Illustrative example. Yellow marks where a person decides; every prompt version is logged with what changed, who changed it and the test run it last passed.
How does the prompt build work?
Our prompt engineering services run in four fixed-price steps, each with a published scope cap. They are a four-day audit of up to 10 existing prompts, a two-to-three-week pilot on one workflow, the prompt library setup built alongside it, and an optional Managed Prompt Desk. Each step ends with a written result you can stop on.
01 4 days
Audit We review up to 10 prompts you already use across one use case, record where the output drifts, map the goal and quote the pilot.
02 2–3 weeks
Pilot One workflow, up to 5 production prompts, built and run against a 20-input test set, with two revision rounds and a go or no-go report.
03 With the pilot
Library setup Up to 15 prompt templates across up to 3 use cases, architecture, versioning, test sets and documentation, handed to you.
04 Monthly · optional
Managed Prompt Desk Up to 10 prompt updates a month, re-testing when a model version changes, and a monthly note on what moved.
What does a first engagement cost?
Fixed prices on the cards at the top, live in about four weeks.
Models Your own AI accounts and API keys
Where they live Your drive, Notion or repository
Versions Every change logged with its test run
Reporting Shared channel, monthly test note
Who is prompt strategy for?
Prompt strategy suits any team whose AI output changes from one run to the next. Three buyer types hire us most, and each gets the same system tuned to a different job. Prompt engineering is still a thing: every AI feature still runs on written instructions, and those need testing whenever a model changes.
Marketing and content teams
Every writer prompts differently, so the output does too. For content and copy alone, prompt engineering for marketing is the dedicated service; here, marketing prompts sit inside one library shared with your other teams.
Support and operations teams
A reply that varies by run is a support risk. You get prompts with fixed output structures, defined refusal and escalation rules, and a test set you can re-run before anything ships.
Product and AI teams
Prompts live in your codebase and break on model updates. You get versioned prompt files, a documented test set, and a re-test when a model version you use changes.
Why choose Eicra for prompt engineering?
Prompt engineering services are worth it when the output matters: they design, test and document the instructions your business gives an AI model. We work test-first: prices carry scope caps, prompts are judged on an agreed test set, and personal data is covered by the processor terms of Article 28(3) of the General Data Protection Regulation (GDPR).
Without a tested prompt system
!!!!!
- Output that changes from one run to the next
- Prompts that break when a model version changes
- Open-ended quotes with no scope cap
- Prompts only their author understands
With EICRA
Pilot report · support prompts
WorkflowOnePromptsUp to 5Test set20 real inputsRevisionsTwo roundsVerdictGo or no-go
Illustrative example
- Prompts accepted against a test set you agreed
- Fixed prices with published scope caps
- A re-test when a model version you use changes
- Prompts, test sets and documentation in your accounts
Is it safe to outsource prompt work?
Outsourcing prompt work is safe when the safeguards are agreed in writing first. The main risks are leaked business logic or sample data, lock-in, and prompts nobody internally understands. As a Bangladesh-based company, we open every engagement with a mutual non-disclosure agreement (NDA), named least-privilege accounts and a data processing agreement (DPA). Reviewed By Eicra.com team
The Open Worldwide Application Security Project (OWASP) lists prompt injection first in its Top 10 for large language model (LLM) applications, and its definition reads: “A Prompt Injection Vulnerability occurs when user prompts alter the LLM’s behavior or output in unintended ways.” The AI Risk Management Framework from the National Institute of Standards and Technology (NIST) sorts this work into four functions: govern, map, measure and manage. Prompts and test data stay inside your own systems, and the documentation lets your team take over.
Which agreements are signed, and when?
NDA — mutual, signed before we see a prompt or a sample input.
DPA — the processor terms required by Article 28(3) of the GDPR, for any personal information.
International data transfers — the transfer instrument your jurisdiction requires, such as standard contractual clauses, signed before personal data moves.
Certifications — a System and Organization Controls (SOC) 2 readiness roadmap is available on request; certifications are listed only when held.
What access and rework terms apply?
Access Named least-privilege accounts that you create, and we work from redacted sample inputs wherever possible.
Ownership The prompts, templates, test sets and documentation are yours; we keep no copies.
Service level agreement (SLA) Response and update times for the desk, stated in the signed terms.
Rework Free when a prompt fails the agreed test criteria; new scope is priced first as a change request.
What proof do you get before you pay for prompt engineering services?
Before you pay for prompt engineering services, you get evidence instead of promises. You get measured commitments, a pilot on one of your own workflows that ends in a go or no-go report, and a free review of one prompt. Client case studies with numbers are added here as clients give permission to name them.
20 inputs
Every production prompt is run on a test set of 20 real inputs from your work before it ships.
Model updates
Under the Managed Prompt Desk, your test set is re-run when a model version you use changes.
Free rework
Rework is free when a prompt fails the agreed test criteria; new scope is priced first as a change request.
Case studies: before-and-after prompt samples and client case studies are added here as pilots complete, with each client’s permission.
Free 30-minute prompt review Send one prompt you already use. You get a written note on what is missing from it and what fixing it would cost, whether or not you hire us.
Common questions about prompt strategy
How much do prompt engineering services cost?
Prompt engineering services with us are priced per step, not per hour, and each price is on the price cards at the top. The steps are a four-day prompt audit of up to 10 prompts, a pilot on one workflow, the prompt library setup and an optional monthly Managed Prompt Desk.
How long does a prompt build take?
A first engagement is live in about four weeks: a four-day audit, then a two-to-three-week pilot with the library setup built alongside it. Once the Managed Prompt Desk is running, prompt updates are delivered inside the monthly allowance of up to 10 updates.
What is a prompt strategy exactly?
A prompt strategy is the system around your AI instructions rather than the instructions themselves. It covers which tasks are worth prompting, how each prompt is structured and tested, and who maintains it. For example, a support prompt can set a role, give three sample replies and require a fixed reply format, then pass a 20-input test before it ships.
How is this different from prompt writing?
Prompt writing gives you one phrase that worked once; a prompt strategy gives your whole team a tested system. You get templates anyone can run, a test set that shows whether a change helped, version history and documentation. The difference shows when the author leaves or the model updates.
What is outside the scope?
Prompt work is not model training or application development. We do not fine-tune or host models, build retrieval-augmented generation (RAG) pipelines or write the software that calls your prompts; that is AI agent development and AI integrations work. We do not sign off legal, medical or financial output, and the audit says so when a use case needs engineering instead.
Start with a four-day prompt audit at a fixed price.