Insights · 10 min read
What AI consulting covers, how engagements work, indicative cost ranges, and how to tell if the strategy phase will pay for itself before you build.
By GGP Editorial
Most AI projects do not fail because the model is weak. They fail because the company skipped the step that comes before any model: deciding what to build, and whether it will move a number that matters. That step is what AI consulting is for.
I write this from the vendor side. GlobeSoft is a software company founded in 2018 with 40-plus engineers and more than 300 delivered projects. We do the consulting and the build under one roof, which colors my view: a strategy that is not tied to something you can actually ship is mostly theater. This guide explains what a serious AI consulting engagement covers, how the work is structured, what it costs, and how to tell whether the assessment phase will pay for itself before you commit to development.
AI consulting is a set of concrete deliverables, not a vibe. If a firm cannot name these, or hands you a deck full of market forecasts instead of a plan, walk away.
An AI readiness assessment reviews your data, your workflows, and your existing systems to find where AI can realistically help and where it cannot. The output is a short list of candidate opportunities, not a 60-page report. A good one will tell you that two of your three ideas are not worth doing, which is exactly the value you are paying for.
Use case identification and prioritization ranks those candidates by expected value, feasibility, and data availability. You end up with a ranked list, with a rough effort and payoff estimate next to each item. This matters because most teams try to do everything at once and ship nothing.
A feasibility study or proof of concept tests the riskiest assumption with a small build. That assumption is usually the data or the model, not the user interface. If the model cannot get past 80 percent accuracy on your actual data, you want to know that in week three, not month six.
Data readiness mapping tells you where your data lives, whether it is clean enough to use, and what has to change before a model can be trained or called. This is where most projects quietly stall. The model is rarely the problem. The data usually is.
Build versus buy versus integrate decides whether you should use an off-the-shelf tool, an API, or a custom model. Most businesses over-build here, and we covered that tradeoff in build vs buy AI software.
An ROI model puts cost against benefit in your own numbers, not industry averages. Governance and compliance covers data privacy, model risk, and any regulatory constraints, which matters most in finance and healthcare.
Most engagements run in three phases, and you should be able to stop after any of them without penalty.
Phase one is discovery and assessment. The consultant interviews your team, reads your data, and maps your processes. This usually takes two to six weeks and ends with the readiness report and the ranked list of use cases.
Phase two is strategy and roadmap. You pick the one or two opportunities worth pursuing, and the consultant produces a build plan: architecture, data pipeline, model approach, timeline, and budget. This is where the work turns from "what could we do" into "what will we build first."
Phase three is proof of concept or first build. A small team builds a working slice to validate the plan. At the end you have something you can extend, or throw away with useful information either way.
The commercial arrangement can be one of four shapes. A fixed-fee assessment is a defined scope for a set price, which suits companies that want a clear deliverable first. Time and materials suits open-ended exploration. A retainer or fractional engagement gives you ongoing advisory time each month, useful when you have internal engineers but no AI leadership. Full delivery rolls the consulting into the build, so the people who wrote the roadmap are the ones building it.
The last option removes the most common failure point, which is the handoff between an advisor and a separate build team.
Costs vary widely by scope, region, and seniority, so treat these as indicative ranges rather than quotes. The numbers below reflect what we see in the market for a China-based senior team with global delivery, which runs below the equivalent engagement from a Big Four firm or a US boutique.
| Engagement | Typical scope | Indicative cost |
|---|---|---|
| Readiness assessment | 2 to 4 weeks, one or two senior people | $15,000 to $40,000 |
| Strategy and roadmap | 4 to 8 weeks, a small team | $30,000 to $80,000 |
| Proof of concept | 4 to 8 weeks, two to four engineers | $25,000 to $70,000 |
| Fractional AI advisory | Monthly retainer | $5,000 to $15,000 per month |
| Consulting plus first build | 3 to 6 months | $60,000 to $200,000 and up |
The single biggest driver of cost is not the AI. It is your data. Clean, well-structured data that lives in one place makes an assessment and a proof of concept cheap. Fragmented data across five systems, with no owner and no shared definitions, can double the effort before a model is even involved.
For a sense of what the build itself costs once the strategy is done, see our breakdown of how much AI development costs in 2026.
There are clear cases where spending on the assessment first is money well spent.
You have a vague goal like "use AI to cut costs." That is not a project, it is a wish. A short assessment turns it into two or three concrete candidates with numbers attached.
You have data but no idea whether it is usable. A readiness review finds out before you spend on a build.
You operate in a regulated industry. In finance and healthcare, skipping the governance conversation at the start costs more later in rework and audit findings.
You have never run an AI project before. An assessment gives you a map and a realistic budget, which is worth more than the fee when it stops you from committing to the wrong thing.
There are also cases where you should skip the long strategy phase and go straight to a build. If you already know the problem, the data is in one place, and the payoff is obvious, a two-week proof of concept will teach you more than a two-month assessment. We say this even though the assessment is revenue: a firm that pushes a big strategy engagement when a proof of concept would do is a firm to avoid.
Ask these questions before you sign anything.
Who actually does the work? A partner sells, but juniors deliver. Ask for the names and backgrounds of the people who will be on your engagement.
Have you shipped something like this? Ask for a case study in your industry, not a list of logos.
Can you build, or do you only advise? Strategy firms hand you a deck and leave. You want a team that can take the roadmap to production, whether that is your engineers or theirs.
How do you charge? A firm that refuses to give a fixed price for a defined assessment is hiding something.
Red flags are easy to spot once you look. Decks full of market statistics instead of your data. A roadmap with no owner and no dates. A refusal to commit to a proof of concept. Those three together tell you the engagement is about billable hours, not about your outcome.
We covered the same evaluation discipline for the build side in how to choose an AI development company.
Here is the honest case for doing both with the same firm.
When consulting and development are separate vendors, the handoff is where things break. The strategy team writes a clean recommendation. The build team wants to ship. A roadmap that looked elegant on a slide runs into a data problem in week one, and neither vendor owns the gap. The result is a pilot that dies in the gap between two invoices.
GlobeSoft runs both, so the person who tells you what to build is the person who has to build it. The roadmap gets written with shipping in mind, because the same engineers will have to live with it. We have 40-plus engineers across Python, Java, Go, and the usual data stack, and we deliver in English and Portuguese across time zones. Our AI work spans generative AI development, AI agent development, AI automation for business, and adding AI to existing business software.
If you already have a specific product in mind and want to skip the strategy talk, start with what AI development services actually include. If you are an enterprise running pilots that never reach production, read enterprise AI development: from pilot to production.
What is the difference between AI consulting and AI development?
Consulting answers what to build, whether it is feasible, and what it will return. Development answers how to build it and then does it. The two overlap in the proof-of-concept phase, which is why firms that do both tend to be more useful than firms that only advise.
How much does AI consulting cost?
An assessment commonly runs $15,000 to $40,000. A full strategy and roadmap runs $30,000 to $80,000. A proof of concept adds $25,000 to $70,000. These are indicative ranges; your scope and data quality move the number more than anything else.
Do I need an AI readiness assessment before building?
Only if your goal is vague, your data is unproven, or you are in a regulated industry. If the problem and the data are both clear, a short proof of concept is usually a better use of the same budget.
How long does an AI consulting engagement take?
A readiness assessment is two to six weeks. Strategy and roadmap is four to eight weeks. A proof of concept is four to eight weeks. You can stop after any phase.
Can the same firm do the consulting and the build?
Yes, and there is a strong argument for it. The handoff between separate vendors is where most AI projects lose momentum. A single team owns the recommendation and the delivery.
Skip the deck. Ask what you will be able to build when the consulting ends, who will build it, and what it costs. A good AI consulting engagement ends with a ranked list of use cases, a build plan with owners and dates, and a working proof of concept for the one that matters most. If it ends with a slide deck and a follow-up meeting, you paid for theater.
Planning an AI initiative and want the assessment and the build handled by one team? Talk to us about your project.
Tell us what you are building and where you are today. We typically reply within 24 hours.
