AI consulting services

AI consulting services: how to compare firms, and what the work should cost

Updated 7 September 2026 11 min read

Short answer

AI consulting services should deliver three things: an assessment that costs your manual processes in hours and money, a sequenced plan of what to change first, and an implementation that leaves working software behind. A proposal containing only the first two is a research report with an invoice attached. Prices vary by roughly twenty times across the market for the same outcome — a large consultancy will quote six figures for the assessment alone, a specialist studio a few thousand for assessment plus a working build. Ask any firm which of the three deliverables is in scope, who personally does the work, and what you own when the engagement ends.

A published evaluation checklist you can use on us and on every competing proposal
Every price listed, next to the market rate for the same deliverable
Engagements end with a workflow in production, not a slide deck
You own the output: source, credentials, documentation, and the process map

Request a scoped proposal Compare the cost yourself

Engagements run with companies of 5–300 people in services, ecommerce, logistics, finance operations, and professional services.

How to evaluate an AI consulting company — including this one

Published as a checklist because most buyers are comparing three proposals with no shared vocabulary between them. Run these seven questions past every firm on your list. Any of them that we would fail, we would rather you knew before signing.

01Which of the three deliverables is in scope: assessment, plan, or implementation? A fixed fee against "AI strategy" with no shipped output is the single most common way this money gets wasted.
02Who personally does the work? Ask for the name of the person who will build it, not the team that will be assigned. If the people in the room are not the people delivering, price in the handover cost.
03What do you own at the end? Source, credentials, and documentation should be yours by default. Workflows that live only in the vendor's account are a retention mechanism.
04Is the engagement priced per deliverable or as an open-ended retainer? Retainers with no named output produce months where nothing ships and nobody can point at what was bought.
05Does the firm take software commissions? Reseller margin distorts recommendations, and you cannot see it in the invoice. Ask directly and get the answer in writing.
06Will they tell you not to automate something? A firm whose revenue depends entirely on builds will find builds. Ask what they have recently advised a client against.
07How is success measured, and against what baseline? If nobody records hours, error rates, and turnaround time before the change, no one can prove the result afterwards — and the next project will not get funded.

Where AI consulting engagements usually go wrong

Rarely on the technology. Almost always on how the engagement was shaped before anyone opened a laptop.

The deliverable was a document

A readiness assessment, a maturity model, a prioritised opportunity matrix — all genuinely useful, and all worthless if the engagement ends there. The gap between a recommendation and a working workflow is where most AI budgets die, and it is usually a second procurement cycle away.

The scope was the whole company

Programmes framed as company-wide transformation have no natural first deliverable, so they produce workstreams instead of results. One process, automated and measured, tells you more about whether this works for your company than any assessment of all of them.

The recommendation followed the vendor relationship

Firms with reseller agreements have a financial interest in a particular answer, and it does not show up in the proposal. This is why the licence question belongs on your checklist and why our own answer is that we take no commissions from any platform.

What our engagement actually contains

Three priced lines rather than one bundled retainer, so you can buy the first and stop. Most clients start at the assessment; some arrive knowing what they want built and skip it.

01
Assessment — $490
Structured intake per department, mapping what actually happens rather than what the process document says. Every recurring process is costed in hours and currency, which is what converts "we should do something with AI" into a ranked list defensible in a budget meeting.
You get Process map, cost-of-manual-work table, ranked shortlist with build estimates, and an explicit not-worth-doing list.
02
Sequenced plan
Delivered with the assessment rather than sold separately. The order matters more than the contents: what to build first, what depends on something else, what to postpone until volumes justify it, and what to leave manual permanently.
You get A sequence with payback periods, dependencies, and a written not-now list.
03
Implementation — from $1,900 per workflow
One process at a time, built end to end: triggers, AI steps for classification, extraction or drafting, human approval gates, exception routing, logging. It runs in parallel with the manual process until both agree on live data.
You get A live workflow in your own accounts, its SOP and runbook, and a measured before-and-after.
04
Support — $690 a month, optional
Monitoring, incident response, and one new or reworked workflow a month. This is how the estate grows without a fresh procurement cycle each time, and it is cancellable with 30 days notice precisely so it cannot become the point of the relationship.
You get Alerting to a named owner, one workflow per month, and a quarterly review of measured savings.

What year one costs, line by line

A model rather than a case study — substitute your own numbers. The point is not that we are cheaper; it is that the comparison is only possible when both sides publish per-deliverable prices, which is why we publish ours.

A 60-person services company automating three processes in the first year: support email triage, supplier invoice entry, and weekly management reporting. Same three outcomes, costed through us and through a mid-market consultancy quoting typical market rates.

Assessment — ours: $490
Three workflow builds at $1,900–2,800 — ours: $6,600
Less the assessment fee, credited against the first build: −$490
Twelve months of support at $690 — ours: $8,280
Internal stakeholder time, 12 h at $45 loaded: $540
Year one total — ours: $15,420
Assessment at market rate: $2,500–8,000
Three builds at market rate ($6,000–15,000 each): $18,000–45,000
Twelve months of support at market rate: $36,000–96,000
Year one total at market rate: $56,500–149,000

The three delivered outcomes are the same; the spread is the sales layer, the bespoke discovery, and the retainer floor. Note the credit line: the $490 assessment is deducted from the first build, so it appears in the total once and then comes back out. Two honest caveats. First, our build figures are floors — five systems and scanned documents push a single workflow well above $2,800. Second, a large consultancy is genuinely the right choice for a regulated multi-country programme with a board-level mandate, and no amount of price comparison changes that. Compare the deliverable list before comparing the totals.

What is in scope, and what we decline

The second list is the more useful one when you are comparing proposals, because it is the list most firms leave out.

In scope

Process discovery, mapping, and cost-of-manual-work analysis
A sequenced implementation plan with payback periods and dependencies
Workflow design: triggers, AI steps, approval gates, exception routing, logging
Implementation on n8n, Make, Zapier, or direct API integrations
Document and email processing: extraction, classification, routing, drafting
Internal SOPs and runbooks so the work survives staff turnover
Baseline measurement and a before-and-after review after launch
Vendor-neutral tooling advice, including when the answer is software you already pay for

Not in scope

Company-wide transformation programmes with no shipped workflow in the first quarter
AI readiness assessments sold as a standalone six-figure deliverable
Replacing your CRM, ERP, helpdesk, or accounting system
Reselling licences with a margin attached — we take no vendor commissions
Custom model training that needs a data science team you do not have
Board-level regulated transformation programmes across multiple jurisdictions — a large consultancy is the correct choice and we will say so
Automating a process that is broken or undocumented — we flag or fix it first

Our prices, next to the market rate for the same deliverable

Published because a quote that only appears after two discovery calls has already cost you a week, and because comparing firms is impossible when only one side shows numbers. We sit deliberately at the bottom of this market, which invites a fair question — answered below the table.

Assessment

$490

Typical market rate: $2,500–8,000

You are choosing between proposals and need a costed picture of your own processes first.

Up to three departments in scope
Process map and cost-of-manual-work table
Ranked shortlist with build estimates and payback periods
Sequenced plan with an explicit not-now list
Credited in full against your first build

Implementation

from $1,900

Typical market rate: $6,000–15,000

The process is chosen and you want it live, measured, and documented.

One workflow, end to end, in your own accounts
Exception routing and human approval gates
SOP and runbook for whoever maintains it
Parallel run against the manual process before cutover
Two weeks of post-launch tuning

Support

$690 / month

Typical market rate: $3,000–8,000 / month

Workflows are live and you want them monitored, maintained, and extended.

Monitoring and incident response
One new or reworked workflow per month
Integration maintenance as vendors change their APIs
Quarterly review of measured savings
Cancel with 30 days notice

Build prices are floors rather than averages. System count and input quality move a quote further than anything else: two systems with clean structured data lands near $1,900, five systems with scanned documents and inconsistent supplier formats lands materially higher. You receive a fixed figure in writing after the intake call, before work begins, with the $490 assessment deducted in full. If your situation is one of the cases in the exclusion list above, we will tell you which type of firm to approach instead.

Why our number is a fraction of a consultancy quote

Below-market pricing should make a buyer suspicious, so here is where the difference actually comes from. None of the four reasons is "more junior people".

No sales or delivery layer

You talk to the person who builds the workflow, on the first call and every call after. No partner, no account manager, no pre-sales engineer, and nobody whose utilisation target has to be recovered from your invoice.

The assessment is productised

Same intake, same cost model, same deliverable every time. Consultancies bill discovery as bespoke work because for them it is genuinely bespoke; here the template already exists, so you are not funding its invention.

The integration work repeats

Your process is specific. The plumbing under it — mailbox to helpdesk, document to accounting system — rarely is. The tenth build of a pattern takes a fraction of the first, and we price the tenth.

No licence margin and no commissions

Revenue comes only from delivered work, which is also why an assessment is willing to conclude that a process should stay manual. There is no software deal riding on the recommendation.

Four types of firm, and when each is the right call

The market presents these as competitors for the same work. They are not — they are structurally different products, and three of the four are the correct answer for someone.

Criterion	Large consultancy	Full-service dev agency	Specialist automation studio	Independent freelancer
Typical year-one cost	$150k–1m+	$60k–250k	$10k–40k	$8k–50k
Time to first working workflow	3–9 months	2–4 months	2–6 weeks	2–8 weeks
Who actually does the work	A staffed team, usually more junior than the pitch	Assigned developers, often rotating	The person you spoke to	The person you spoke to
What you own at the end	Documents; software depends on the contract	Code, sometimes hosted by them	Source, credentials, and runbook	Depends entirely on the individual
Right choice when	Regulated, multi-country, board-mandated programme	You need a product built, not a back office fixed	A handful of expensive processes crossing several systems	One well-specified workflow and you can manage the risk
Main risk	Six figures spent before anything runs	Automation treated as a software project	Small team — continuity depends on the runbook	Single point of failure with no backup

What to do next, depending on where you are

Read down to the line that matches your situation. Three of these six answers do not involve hiring us.

You are comparing two or three proposals and cannot tell them apart
Run the seven-question checklist above against each one, then ask all of them to price the assessment separately from the build. The ones that refuse are telling you something useful.
You can name the process and it costs more than roughly ten hours a month
Assessment, then one build. Shortest payback path, and the assessment fee comes off the build price.
You have a board mandate, multiple countries, and a regulator to answer to
Engage a large consultancy. That structure exists for exactly this, and a specialist studio is the wrong shape for it — we will say so on the first call.
Departments are each buying their own AI tooling and it is fragmenting
You need a roadmap before builds. That is a different engagement, described on our AI strategy consultant page.
The automation is one trigger and one action
Do not hire any firm. A Zapier or Make subscription and an afternoon covers it; our platform comparison explains which to pick.
You want to know your own numbers before talking to anyone
Run the free process audit tool and the ROI calculator. Both give you the inputs every firm will ask for, which makes their proposals directly comparable.

A one-week assessment you can run before contacting any firm

This produces the inputs every consulting proposal is built on. Doing it yourself first means you evaluate proposals against your own numbers rather than the firm's framing — and if nothing clears the bar, you have saved the fee entirely.

Day 1Ask every team to list each task they perform at least weekly that involves moving information between two systems.
Day 2For the five most frequent, count actual occurrences last month from the systems themselves — search the mailbox, count the records — rather than asking people to estimate.
Day 3Time each end to end, honestly, including the cost of interrupting other work to do it.
Day 4Multiply instances by minutes by loaded hourly cost. Rank by monthly cost, not by how irritating the task feels.
Day 5For the top two, check whether someone can write the rule down. If yes it is automatable; if no, the useful build is an approval queue with a suggestion.
Day 6Confirm each system has an API, a scheduled export, or a dedicated mailbox. None of the three is a hard blocker and every firm will eventually hit it.
Day 7Write a one-page brief: the top three processes, their monthly cost, and the systems involved. Send the same brief to every firm you are evaluating.

Sending an identical brief to each firm is the single most effective thing you can do in this process — it turns four incomparable proposals into four comparable ones. If your top process costs under about $500 a month, the honest conclusion is that no firm should be hired this quarter.

Looking for something more specific?

Frequently asked questions

What do AI consulting services include?

Three deliverables, which should be priced separately: an assessment that costs your current manual processes, a sequenced plan of what to change and in what order, and an implementation that leaves working software behind. Ask any firm to confirm in writing which of the three their fee covers. Proposals that bundle all three into one open-ended retainer are the hardest to compare and the easiest to overspend on.

How much do AI consulting services cost?

The market spreads roughly twenty times for the same outcome. A large consultancy will quote $150,000 upwards for a first-year programme, a full-service agency $60,000-250,000, and a specialist studio $10,000-40,000. Ours are published: $490 for the assessment, credited in full against the first build; from $1,900 per workflow; $690 a month for optional support. Compare deliverable lists before comparing totals, because the numbers describe different products.

How much do AI consulting services cost for a small business?

For a company under about 50 people, a realistic first-year budget is $3,000-12,000 covering an assessment and two or three workflow builds, with support optional. Anything quoted in six figures at that size is scoped for a different kind of company. Below roughly five people there is usually not enough repetition to pay back any build, and the honest answer is a subscription tool.

What should be in an AI consulting proposal?

Named deliverables with individual prices, the name of the person doing the work, what you own at the end, the baseline against which success will be measured, and an explicit statement about software commissions. If the proposal is a fee against a phase name — "discovery", "strategy", "enablement" — with no output described, ask for it to be rewritten before comparing it to anything else.

AI consulting company or freelancer — which is safer?

A freelancer is cheaper and often better at one well-specified workflow, but is a single point of failure with no continuity if they move on. A firm costs more and buys you redundancy and process. The risk is manageable either way if you insist on the same two things: documentation written for someone else to inherit, and workflows living in your own accounts rather than the provider's.

How do I know whether an AI consulting firm is any good?

Ask what they recently advised a client not to build. A firm whose revenue depends entirely on implementation will always find implementation, and the ability to give a specific example of talking a client out of something is the closest thing to a reliable signal in this market. Then check whether the people in the pitch are the people delivering.

Do we need an AI strategy before implementing anything?

Usually not. A company with three obviously expensive manual processes should automate one and measure it, which teaches more than an assessment of everything would. A roadmap earns its cost in a specific situation: several departments buying AI tooling independently, where the risk is fragmentation and duplicated spend rather than inaction.

Will an AI consulting engagement require replacing our systems?

No, and treat it as a warning sign if replacement comes up early. Automation runs as a layer on top of the CRM, helpdesk, and accounting software you already have. Replacing a core system is a separate decision with its own business case and risk profile, and it is not a prerequisite for removing manual work.

How long does an AI consulting engagement take?

Our assessment takes about two weeks, and a first workflow typically goes live two to four weeks after that. Larger firms run three to nine months before anything is in production, which is not necessarily wrong — it reflects a different scope — but it is the number to ask for explicitly. The usual delay in any engagement is access to systems, not build time.

What happens if the assessment says we should not automate?

It says so in writing, and that is a real outcome rather than a failed engagement. Some processes are cheaper to fix by deleting a step, changing a form, or renegotiating with a supplier. Recommending that costs us the build fee and saves you considerably more, which is why the assessment is sold and delivered separately from implementation.

Who owns the work when the engagement ends?

With us, you do: source, credentials, process map, and a runbook written for whoever inherits it. This is worth confirming with every firm you evaluate, because workflows hosted in a vendor's own account are a retention mechanism rather than a technical necessity, and the difference only becomes visible when you try to leave.

Do you take commissions from software vendors?

No. We resell nothing and hold no partner agreements, so there is no margin riding on which platform gets recommended. We suggest asking every firm on your list the same question and asking for the answer in writing — reseller economics are the least visible influence on a technology recommendation.

Next step

Get a proposal you can compare

Send the processes that bother you most. We come back with what they likely cost per month, which ones clear the payback bar, and a fixed price per deliverable — in a format you can hold next to any other firm's proposal.