Fr
En
De
It
Es
Uk
What are AI consulting services, and what should they produce?
AI consulting services help a company answer three questions: where AI will actually save time or earn money, which technical approach fits each case, and what it will cost to build and run. Good AI consulting ends with decisions and a plan your team can execute next month, not a maturity model or a slide deck about transformation.
Lytvynov Production approaches AI consulting as engineers, not as strategists. The people who run the assessment are the people who build LLM products, RAG pipelines, AI agents and fine-tuned models for clients and for our own products. That changes what the output looks like: instead of "consider using generative AI in customer support", you get "a retrieval-based assistant over these three document sources, this model tier, this evaluation set, around this cost per thousand conversations, and here is a prototype answering 50 of your real tickets".
What is included in our AI readiness assessment?
The AI readiness assessment is a fixed-scope engagement of two to four weeks. The length depends on how many teams and use cases are in scope. The table shows the standard steps and what each one produces.
| Step | What we do | What you receive |
|---|---|---|
| 1. Kickoff and interviews | 45-minute sessions with the owners of 3 to 8 processes or product areas | A long list of candidate use cases in plain language |
| 2. Data and systems review | Look at where the relevant data lives, its format, quality, access rights and personal data | A data readiness note per use case, including blockers |
| 3. Use case scoring | Score each case on value, feasibility, risk and running cost | A ranked shortlist with the reasoning written down |
| 4. Architecture options | Prompting, retrieval (RAG), fine-tuning, agents with tools, or an off-the-shelf product | A recommended approach and model shortlist for the top cases |
| 5. Prototype of the riskiest case | A small working prototype run against your real inputs | Measured accuracy on a test set, sample outputs, cost per request |
| 6. Roadmap and budget | Phases, dependencies, team shape and budget estimates | A roadmap document and a fixed quote for phase one if you want us to build it |
The prototype step is the part most AI strategy consulting services skip. It is also the step that most often changes the plan, because real inputs expose problems no interview reveals: scanned PDFs, missing fields, inconsistent naming, or answers that depend on a system nobody mentioned.
How do we decide which AI use cases are worth building?
A use case is worth building when it saves meaningful time or brings in revenue, the data needed is available, mistakes are cheap or catchable, and the running cost per task is well below the value of the task. We score every candidate on those four dimensions rather than on how impressive the demo would be.
Common patterns that score well: drafting documents from structured data, classifying and routing incoming requests, extracting fields from invoices or contracts, answering questions over internal documentation, and summarizing long threads for a decision maker. Patterns that often score poorly on first review: fully autonomous decisions with legal or financial consequences, use cases where the needed data does not exist yet, and features where users would not trust the output without redoing the work. For back-office candidates in particular, our guide on which business processes to automate with AI first goes deeper into the scoring.
Generative AI consulting: model, data and build-vs-buy choices
Generative AI consulting is mostly about four decisions that are expensive to reverse later: which model family, how the model gets your knowledge, where the data is processed, and whether to build or buy. We make these decisions with measurements from your inputs, not from vendor benchmarks.
- Model choice. We usually test OpenAI and Anthropic Claude models side by side, plus an open model such as a Qwen or Llama family model when data must stay on your own servers. The test compares accuracy on your examples, latency and cost per request.
- Knowledge access. Retrieval (RAG) is the default for answering from company documents. Fine-tuning is worth it when you need a consistent style or persona, which is what we did with LoRA adapters in the AI Grief Companion project, combined with retrieval for facts.
- Data handling. We map which personal or confidential fields would reach a third-party API and propose masking, regional endpoints or self-hosting where needed.
- Build or buy. If a mature product already solves the use case, such as help desk AI inside the tools you already pay for, we say so. Custom builds make sense when the workflow is specific to your business or the AI is part of your own product.
How is engineer-led AI consulting different from strategy consulting?
Engineer-led AI consulting is narrower and more concrete than classic strategy consulting. It does not cover org design, change management programs or market sizing. It covers what can be built with current models, how reliably, at what cost, and in what order. For most companies with 10 to 200 people, that is the part that has been missing.
| Strategy-led consulting | Engineer-led consulting (our approach) | |
|---|---|---|
| Typical output | Vision, maturity score, opportunity map | Ranked use cases, architecture, prototype results, costed roadmap |
| Evidence | Industry benchmarks and case studies | Your data, measured on a test set |
| Running cost estimate | Rarely included | Per request and per month, for each shortlisted case |
| Who executes afterwards | Often a separate implementation partner | Us, your team or another vendor, using the same documents |
| Typical length | Several months | Two to four weeks |
The link to delivery matters. We build AI products ourselves, from AI Resume Master, our LLM-based resume builder that reached 50,000 monthly active users, to R&D such as an autonomous drone swarm prototype that turns satellite data into tasks a fleet picks up on its own. We also use AI coding agents in our own engineering process and build MCP servers for our internal tools. Recommendations come from that practice.
How much do AI consulting services cost and how long do they take?
A focused AI consulting engagement usually takes two to four weeks: a single use case validation with a prototype can fit in one to two weeks, while an assessment across several teams with a prototype of the top case takes two to four. Effort depends on the number of use cases, the number of teams to interview and whether a prototype is included.
We quote a fixed price for the assessment after a short scoping call. If you then build with us, the documents from the assessment become the scoping phase of the build, so the work is not repeated. AI features for an existing product start from $5,000 with us; our AI app development cost guide shows how larger builds are priced.
When you do not need an AI consultant
You do not need an AI consultant when the use case is already clear and small. If you know you want a chatbot on your help center or an LLM feature inside an existing product, start with a scoping call for that build: see AI integration services or, for a new product, AI product development. You also do not need one to "get started with AI" in general. Buying seats of a general assistant for your team costs little and teaches you where the real needs are. An assessment earns its place when there are several competing ideas, sensitive data, or a budget decision that needs numbers behind it. If you are comparing vendors for that work, our checklist on how to choose an AI development company lists the questions worth asking.
How we run AI consulting engagements
We start with a 30-minute call to understand your business, your data and the ideas already on the table. If an assessment makes sense, we send a fixed-scope proposal with the teams to interview, the use cases in scope and the prototype we plan to build. You get a written report, the prototype code and a roadmap within two to four weeks, in English, with a senior engineer as your single point of contact. Book an AI consulting call.
