Good AI consulting delivers three things you can hold: a readiness assessment, a roadmap, and a scoped pilot. Bad AI consulting delivers a deck. That is the whole distinction, and most of the money wasted on AI advisory work gets wasted on the deck side of it. The number that explains why the work is worth paying for at all comes from McKinsey: 88% of organizations now use AI in at least one function, but only 6% are high performers seeing real EBIT impact. Almost everyone is doing AI. Almost no one is getting paid for it. Consulting that closes that gap is worth its fee. Consulting that produces a strategy slide and a handshake is not.
We’re gmware, a custom software development firm headquartered at 5900 Balcones Drive in Austin, TX, with engineering centers in Bangalore and Mohali, India. We build AI into operational software for mid-market companies, and we sell the assessment-then-pilot version of consulting, not the vision-deck version. This piece lays out what AI consulting actually delivers, what each deliverable costs against current market rates, where the big firms genuinely beat us, and how we scope an engagement so week thirteen is a measured result instead of a fresh argument.
The adoption-to-impact gap, in three numbers
What does AI consulting actually deliver
Strip away the brochure language and AI consulting produces a small number of artifacts. Three of them matter. The rest are packaging.
The first is a readiness assessment: a structured audit of whether you can ship an AI project at all. Where does your data live, who owns it, what fraction is current and clean. Which workflows have enough volume and clear enough rules to be worth automating. Whether there is a person with the authority to accept or kill the result. This is the unglamorous deliverable that catches the problems that sink pilots, and it is the one buyers most often skip.
The second is a roadmap: a ranked list of candidate projects, each scored by expected payback, data readiness, and risk, with a recommended first project and an explicit “not yet” list. A roadmap that recommends everything is not a roadmap. The value is in what it tells you to skip.
The third is a scoped pilot: a single workflow, a baseline metric measured before any code, a build against real data, and a kill-or-scale gate at the end. This is the deliverable that turns advice into evidence. Everything before it is preparation for this.
Notice what is not on the list: a strategy deck as the final output, a “center of excellence” org chart, a vendor shortlist with no project attached. Those can be useful inputs. None of them is a deliverable you can take to a CFO and say “this paid back.”
The three deliverables that matter
What each AI consulting deliverable should contain
Here is the version we would hand a buyer who asked “what am I actually paying for.” Use it to read any proposal, ours included.
| Deliverable | What it contains | What it answers | Typical market price |
|---|---|---|---|
| Readiness assessment | Data inventory and quality audit, use-case scoring, owner and budget check, integration map | ”Can we ship an AI project, and where?” | $2K to $8K small business; $5K to $15K mid-market |
| Prioritized roadmap | Ranked projects, payback estimate per project, risk flags, a first project, an explicit “not yet” list | ”Which one do we run first, and what do we skip?” | Usually bundled with the assessment |
| Scoped pilot SOW | One workflow, baseline metric, build plan against real data, kill criteria, integration and QA budget | ”Will this specific thing pay back in 90 days?” | Separate, scoped per workflow |
| Pilot build + measurement | Working software, instrumented usage, a measured result against the baseline, a written verdict | ”Did it move the number, yes or no?” | The largest line item |
The market pricing for the assessment is the part you can check independently. A structured AI readiness assessment runs $2,000 to $8,000 for small businesses, $5,000 to $15,000 for mid-market, and $15,000 to $50,000 and up for enterprises. Anything priced under two thousand dollars is usually a sales call with a worksheet attached. Against the cost of a pilot that fails because nobody audited the data, the assessment is cheap insurance, which is the same case we made in why most AI pilots fail and what the survivors do.
What an AI readiness assessment costs
Why the readiness assessment is the deliverable that earns its money
Most buyers want to skip straight to the pilot. We get it. The assessment feels like overhead, and the pilot feels like progress. The assessment is where the real risk lives.
The single biggest hidden cost in any AI project is the data. In retrieval-augmented builds specifically, data cleaning and preprocessing runs 30% to 50% of the project cost. If you discover that mid-pilot, your 90-day plan becomes a nine-month cleanup nobody budgeted. The assessment surfaces it before the SOW, when it is a line item instead of a crisis. We have run production data systems ourselves, so this is not a theoretical warning. Shield Suite, our own retail-intelligence product, pulls signal across 60,000-plus beverage-alcohol storefronts, and the guardrails around messy, inconsistent source data are most of the work. We learned the data-is-the-project lesson on our own dime first.
The assessment also catches the non-data killers. No owner with P&L authority. A workflow that sounds narrow (“customer service”) but is actually five workflows wearing a trench coat. A success metric nobody has measured, so there is no baseline to beat. These are the failure modes that show up in month three of a pilot, and every one of them is cheaper to find in week one of an assessment.
What an AI roadmap is for, and what it is not
A roadmap is a prioritization tool, not a wish list. The deliverable that earns the name does two things most “AI strategies” refuse to do: it ranks, and it says no.
Ranking means each candidate project gets scored on three axes you can defend to a board. Expected payback. Data readiness. Implementation risk. The project that wins is rarely the most exciting one. It is usually the boring high-volume workflow with clean data and a clear owner, because that is the one that actually ships and pays back, which builds the credibility to fund the exciting one next.
Saying no means an explicit “not yet” list. The roadmap that tells you to do all four of your AI ideas is selling you four budgets. The honest version tells you to run one, defer two until the data is ready, and kill one because the workflow is too rare to matter. We think a roadmap that says “skip these three” is worth more than one that blesses everything, even though it is a harder thing to sell. The reason most companies sit in the 88%-adopted, 6%-rewarded gap is that they spread thin instead of going deep on one workflow. A roadmap exists to stop that.
If you want the longer version of how we decide what to automate first, the what to automate first framework walks through the scoring grid we actually use.
How gmware scopes an AI engagement
Here is the sequence, start to finish, so you know what you are buying before you buy it. No phase starts until the previous one produces its artifact.
How we scope an engagement
Two things make this different from the standard engagement. First, we do the data audit before we quote the pilot, because what we find in the data is what determines the price. A vendor who quotes a fixed pilot number before looking at your data is guessing, and you inherit the guess. Second, we write the kill gate into the engagement, ours as much as yours. Week twelve produces a written verdict: the metric moved past the threshold, or it did not. “Let’s keep it running and see” is how zombie pilots are born, and a zombie pilot burns budget and credibility for the next attempt.
We scope the production path, integration, permissions, monitoring, into the pilot itself rather than pretending it is a later problem. That is the part a demo skips and the part that is most of the real bill. Our AI agents and LLM integration practice runs delivery from Austin with engineering in Bangalore and Mohali, which keeps senior oversight on US hours without US-only burn rates.
When McKinsey, BCG, or Accenture is the better fit
We will lose some of these, and we will tell you when to let us. The big strategy and systems firms are not selling the same thing we are, and for some problems they are the right call.
Accenture booked $5.9 billion in AI deals in fiscal 2025, with AI revenue tripling to $2.7 billion across roughly 6,000 projects. That scale exists for a reason. If your engagement is a board-mandated transformation across multiple business units, a global rollout that needs change management in fifteen countries, or a regulated program where the consulting firm’s name on the report is itself part of the risk mitigation, hire the big firm. They have the bench, the methodology, and the institutional cover that a multi-hundred-million-dollar program requires. That is genuinely their game, and it is not ours.
Here is where we win. When the question is “which one AI project should we run first, and will it actually pay back,” a smaller firm that audits your data before quoting and writes a kill gate into the SOW is a better fit than a seven-figure strategy engagement. The big-firm assessment can run into six figures for enterprises and is overkill when you need one workflow proven, not three business units reorganized. The honest test: if you can name the specific workflow that hurts, you want the assessment-then-pilot version. If you cannot, and the problem is genuinely “what is our AI strategy as a company,” that is when the bigger firm’s framing earns its rate.
How to read an AI consulting proposal
Run any proposal, including one of ours, through these checks before you sign.
- Does it end at a deliverable you can take to a CFO, or at a strategy deck? A deck is an input, not an outcome.
- Does it include a data audit before the pilot price is quoted, or is the pilot a fixed number with no look at your data?
- Does it name one first workflow, or does it bless all of your ideas at once?
- Is there a baseline metric and a written kill criterion, or is success defined as “it seems helpful”?
- Is the production path, integration, permissions, and monitoring, scoped into the pilot budget, or is it a later problem?
If two or more of those fail, the proposal is selling you the 88%-adopted, 6%-rewarded experience with better slides. The fix is to ask for the missing pieces before money changes hands. A consultant who built the assessment-then-pilot way will add them without flinching, because that is how the work is supposed to be shaped.
How gmware does it
We sell the assessment first, on purpose, because it is the deliverable that decides whether the rest is worth doing. The audit comes before the quote. The roadmap says no to some of your ideas. The pilot has a baseline and a kill gate written into the SOW, ours as well as yours. And sometimes the assessment’s answer is “don’t start yet,” because the reporting that would tell you whether AI worked is itself broken, in which case the right first project is a machine-learning and AI build on a clean data foundation, not a pilot pointed at numbers nobody trusts.
Tell us the one workflow you are trying to fix and we will give you a straight answer on whether it is ready, what an assessment would cover, and what a scoped pilot would cost, within 48 hours. If the honest answer is that you need a bigger firm, we will say that too.