Pricing
There is no price list on this page, and that is a deliberate choice rather than a sales tactic. This page explains the three shapes an engagement takes, what genuinely moves the cost up or down, what is always included, and how to get a real number out of us quickly. If you want a figure today, the fastest route is thirty minutes on a call, which costs nothing.
The six things that decide what you pay
Not one of them is the model you use. Every one of them is something about your organisation that we can only assess by looking.
The state of your data
This is the single largest cost driver and it is not close. Clean, documented, accessible data makes a project shorter. Data spread across systems that were never designed to talk to each other makes it longer.
How clearly the problem is defined
A question with a measurable right answer is cheap to build against. A vague aspiration is expensive, because the first weeks go on turning it into something testable.
How wrong the system is allowed to be
A tool that suggests, with a person reviewing, is far cheaper than one that decides unattended. Required accuracy is a price lever most buyers never think to pull.
How many systems it has to touch
Integration is routinely underestimated. Every system the solution must read from or write to adds work, and legacy systems without a usable interface add a lot.
What the rules require
Regulated settings need documentation, bias testing, evaluation records and an audit trail. That is real engineering, not paperwork, and it is priced as such.
Who owns it afterwards
A system handed to an internal team that can run it costs less over time than one we keep running for you. Both are legitimate. They price differently.
The number we care about is not our price
Before quoting anything we ask what solving the problem is worth to you in a figure you already track. Hours recovered, error rate, lost orders, delayed decisions. If the value of solving it is not comfortably larger than the cost of solving it, the correct advice is to leave it alone, and we will give that advice even though it ends the conversation.
Most agency pricing pages do one of two things. They publish three tiers with invented feature lists, which tells you nothing about your project, or they say contact us for a quote and stop, which wastes your time. This page tries to do the useful thing in between: explain exactly how a number gets arrived at, so you can judge whether the process is honest before you spend thirty minutes on a call.
iSpark prices per engagement rather than from a rate card. Assessments are fixed price. Builds are quoted against a written specification produced during the design stage, before any building starts. Managed arrangements are monthly and reviewed quarterly. The first thirty minute conversation is free and carries no obligation, and a meaningful number normally follows within a few days of us seeing the actual data.
Why there is no price list on this page
Because any figure we published before looking at your data would be a guess wearing a suit. We have watched two organisations in the same industry, describing what sounded like the same problem, need projects that differed by a factor of four. The difference was not ambition or scope. It was that one had a documented warehouse and the other had fourteen years of decisions living in spreadsheets on individual laptops.
Published tiers also create a bad incentive. Once a vendor has advertised a package, the work bends to fit the package, and the client pays for a shape of project that suits the seller. We would rather quote the work you actually need, including when that work turns out to be nothing.
What we will commit to in public is the process, which is what the rest of this page describes. You will know what drives the number, what is always included, what you never pay for, and how fast you can get a figure.
If you need a budget figure today
Say so in your first message and describe the problem in a few sentences. We will give you an order of magnitude on the call, clearly labelled as the rough estimate it is, so you can take something to a budget holder. A real quote still waits until we have seen the data, because that is the only point at which a number stops being fiction.
The three shapes an engagement takes
Almost everything we do is one of these three, or a sequence of them. Most clients start with the first and move to the second or third once the problem is genuinely understood. Nobody is asked to commit to a year before they know whether the idea works.
| Shape | What you get | How it is priced | When it fits |
|---|---|---|---|
| Assessment | A structured look at your data, your process and the options, ending in a written recommendation that may be to do nothing | Fixed price, agreed before it starts | You suspect AI applies but cannot yet tell which problem to point it at |
| Build | A defined system delivered to production with tests, documentation, monitoring and a handover session | Quoted against a written specification, produced during design | The problem is understood and the value of solving it is quantified |
| Managed | We run and maintain a system, watch it for drift and failure, and retrain or adjust it on an agreed cadence | Monthly, reviewed quarterly, cancellable on notice | The system matters to the business but you have no internal team to own it |
The assessment is the one people hesitate over, because paying someone to tell you whether to proceed feels like an overhead. It is usually the cheapest money spent in the whole sequence. An assessment that ends in do not build this has saved the client every rupee, pound and dollar of the build that would have followed.
What actually drives the cost
When a quote comes back higher than someone expected, it is almost never the modelling. Modelling is the visible part and rarely the expensive part. These are the things that move a number, roughly in order of how often they are the deciding factor.
Data condition, which decides most of it
Getting data into a usable state is the largest line in most projects. If the information lives in one system, is reasonably complete, and someone can explain what the fields mean, the work is short. If it is spread across an ERP, a CRM, a warehouse and a folder of spreadsheets, with the same customer appearing four times under three spellings, the work is long. Nobody enjoys hearing this. It is still true, and it is why we will not quote before looking.
How precisely the problem is stated
Reduce customer service cost is not a specification. Route incoming emails into six categories with a person reviewing anything the system is unsure about is a specification. The first takes weeks to turn into the second, and those weeks are billable, so arriving with the second saves you money.
The accuracy the situation demands
This is the lever most buyers do not realise they are holding. A system that drafts a response for a person to approve can be useful at a level of accuracy that would be unacceptable in a system that sends the response itself. Moving from assisted to autonomous is often the difference between a modest project and a large one, and the assisted version frequently captures most of the value.
Integration surface
Every system the solution must read from or write to adds engineering. Modern systems with documented interfaces add a little. Legacy systems without one add a lot, occasionally more than the model itself.
Regulatory and evidence requirements
In regulated settings the evidence trail is part of the product, not paperwork bolted on afterwards. Model documentation, bias testing across protected groups, evaluation records and risk classification under regimes such as the EU AI Act are real engineering work. If you are in scope, budget for it deliberately rather than discovering it during an audit.
| Lever | Cheaper | More expensive |
|---|---|---|
| Data | One system, documented, complete | Many systems, undocumented, gaps and duplicates |
| Problem definition | A measurable question with a right answer | A general aspiration needing translation first |
| Autonomy | Assists a person who reviews the output | Acts unattended with no human checkpoint |
| Integration | Few systems, documented interfaces | Many systems, legacy, no usable interface |
| Oversight | Internal use, low consequence of error | Regulated decisions affecting people |
| Ownership | Your team runs it after handover | We run it indefinitely under a managed arrangement |
What you never pay for
Worth stating plainly, because some of these are billed as standard elsewhere and people have learned to expect them.
- The first conversation. Thirty minutes, no charge, no deck, no credentials presentation. If the answer is that we are the wrong supplier, you get that answer free.
- Being told to buy something instead. If a product you can license this week does most of the job for a fraction of the cost, we will say so, and we will not invoice you for the advice that lost us the work.
- A proposal. We do not charge for writing the document that asks you for money.
- Travel. Delivery is fully remote, so there is no travel line buried in a quote and no expenses to reconcile afterwards.
- Your own intellectual property. IP in bespoke work transfers to you on payment, including trained model weights, pipeline code and documentation. There is no licence fee to keep using what you paid us to build.
- Surprises. Builds are quoted against a written specification. If the scope genuinely changes, that is a conversation and a new number, not an invoice you were not expecting.
How to get a real number quickly
The single biggest determinant of how fast we can quote is how much useful detail arrives in your first message. None of this is mandatory, and an enquiry with none of it still gets a proper reply. Each item removes a round trip.
- Describe the problem as a symptom rather than a solution. What goes wrong, who notices, how often.
- Say what it costs you now in a number you already track. This decides whether the project is worth doing at all.
- Name the systems the data lives in, roughly how much there is, and how far back it goes. Messy is expected and is not a reason to hold off.
- Say whether a person can review the output, or whether the system has to act alone. This moves the cost more than almost anything else.
- Name any fixed constraint. Data residency, a cloud you must stay on, a regulator you answer to, a date driven by an audit or a contract renewal.
With that, the free call usually produces an order of magnitude on the spot and a written next step within a couple of days. The contact page lists the same guidance and the form delivers straight to hello@ispark.biz.
How a quote comes together
Five steps. You can stop after any of them, and the first two cost nothing.
Free thirty minute call
You describe the problem, we ask what data sits behind it and what solving it is worth. You may leave with the advice to buy a product instead, at no cost.
Rough order of magnitude
If it looks viable, we give you a figure you can take to a budget holder, labelled clearly as an estimate made before seeing data. Still free.
Fixed price assessment
We look at the actual data rather than a description of it. This is the first paid stage, it is fixed price, and it frequently ends weak ideas before they cost more.
Written specification and firm quote
The approach, the evaluation method, the failure modes we expect, the integration points and a price. Agreed before any building begins.
Build, then decide about running it
Delivered to production with tests, documentation and monitoring. At handover you choose between your team owning it or a monthly managed arrangement.
What is in the price, every time
These are not optional extras or an enterprise tier. A system without them is not finished, so we do not charge separately for finishing it.
An evaluation harness
Written before the system exists, so performance is measured against held out data rather than demonstrated on a curated example. You can rerun it after we leave.
Model and system documentation
What it does, what it was trained on, where it fails, and what to check. The artefact an auditor, a regulator or your successor will ask for.
Monitoring and alerting
Drift detection and data quality checks that produce an actionable alert rather than noise, so you find out before your customers do.
A runbook and a handover session
Written operating instructions and live time with whoever will own it. A system nobody can explain is a system nobody can fix.
Source code and full IP transfer
Pipeline code, trained weights and documentation, transferred on payment. Licensed or open source components are listed in the specification before work starts.
An honest account of the limits
Written down: what the system does not do, where it should not be trusted, and which conditions would degrade it. Stated at handover, not discovered later.
Is this worth spending money on?
A question we would rather you answered before the call than after the invoice.
Worth commissioning if
- The problem costs you something you can put a number against
- There is real data behind the process, even if it is messy or spread across systems
- Someone internally will own the result, or you want us to run it under a managed arrangement
- A person can review the output, at least at first, so the accuracy bar is reachable
- You would rather be told to buy a product than be sold a build
Keep your money if
- No product exists yet and the data to build one has not been collected
- The requirement is a strategy deck for a board rather than a system that runs
- An existing tool does most of the job for a fraction of the cost, which we will tell you
- The underlying problem is a broken process, and AI would only automate the breakage faster
Frequently asked questions
Marked up with FAQPage schema so these answers can surface directly in search results and inside AI assistant responses.
How much does an AI project cost?
It depends far more on the condition of your data than on the technique, which is why this page carries no price list. Two organisations in the same industry describing what sounds like the same problem can need projects that differ by a factor of four, based on whether their data is documented and in one place. iSpark prices per engagement: assessments are fixed price, builds are quoted against a written specification, and managed arrangements are monthly. The first thirty minute conversation is free.
Why does iSpark not publish a price list?
Because any figure published before seeing your data would be a guess. Published tiers also create a bad incentive, where the work bends to fit the package and the client pays for a shape of project that suits the seller. What we commit to publicly is the process: what drives the cost, what is always included, what you never pay for, and how quickly you can get a real number.
Can I get a budget figure before committing to anything?
Yes. Describe the problem in a few sentences and ask for one. On the free thirty minute call we will give you an order of magnitude you can take to a budget holder, clearly labelled as an estimate made before seeing data. A firm quote follows the fixed price assessment, because that is the first point at which a number stops being fiction.
What makes an AI project more expensive?
In rough order: data spread across many undocumented systems, a problem stated as an aspiration rather than a measurable question, a requirement for the system to act unattended rather than assist a person, a large integration surface especially with legacy systems, and regulatory evidence requirements such as bias testing and model documentation. Notably, the choice of model is rarely the deciding factor.
Who owns the code and the model?
You do. Intellectual property in bespoke work transfers to the client on payment, including trained model weights, pipeline code and documentation. There is no ongoing licence fee to keep using what you paid us to build. Where an engagement uses open source or licensed components, those terms are listed in the specification before work begins, so nothing surprises you at handover.
What happens if the assessment says the project should not go ahead?
You get that in writing and you owe nothing further. It happens regularly and it is the assessment doing its job. Paying a fixed price to be told not to proceed is the cheapest money in the sequence, because it saves the entire cost of the build that would have followed. We will also say when the right answer is to buy an existing product instead, even though that ends the engagement.
Before you ask for a quote
Three pages worth reading first, because each one changes what you would ask us for.
AI readiness assessment
Twelve questions, scored instantly in your browser. Tells you honestly whether you are ready to spend money on this yet.
Read more →About iSpark
How engagements run, what we believe about AI work, and the five things we refuse to build at any price.
Read more →AI services
The engineering catalogue. Several of these pages tell you to buy a product rather than commission a build.
Read more →Want a number rather than a page about numbers?
Thirty minutes, no charge, no deck. Describe the problem and we will give you an order of magnitude on the call, or tell you plainly that the project is not worth doing.