AI consulting services that end in a working system
AI consulting is two jobs: deciding where AI is worth building, then building it. InApps audits your systems and data, scores each use case on value and feasibility, and hands back a costed roadmap. Then the same engineers ship it. No handoff to a delivery team you have never met.
Trusted by engineering teams across 15+ countries - from startups to Fortune 500.
Most AI consulting ends at the slide deck
Six things clients told us went wrong the last time. Every one is a reason to ask who builds what gets recommended.
The roadmap arrives as a PDF. The people who wrote it do not ship software.
The pilot demoed well. Nobody measured whether it held up at real volume.
Two discovery phases billed before a line of code existed.
AI was the recommendation. Nobody checked whether the data could support it.
It launched. Then a model version changed and quality dropped with nobody watching.
Every use case scored high. A prioritised list that prioritises nothing.
What AI consulting actually is
Not sure whether AI is the right tool for this?
Assessment, roadmap and build. One team.
AI consulting means working out where AI moves a number you already report, then designing how to get there. Most firms stop at the design. We keep going and build it.
That line is worth checking before you sign anything. A strategy house hands over a document. A dev shop builds whatever you ask for. AI strategy consulting only pays off when whoever writes the plan is accountable for shipping it.
What the engagement hands back
Artefacts, not impressions.
- An audit of the systems and data each use case depends on
- Use cases scored on business value and feasibility
- A phased roadmap with cost and risk per phase
- A working prototype, deployed to staging
- An evaluation report run against your own data
What stays yours
All of it, from the first commit.
- The code, the prompts and the evaluation suite
- Cloud accounts and model API keys, in your name
- Your data, never used to train a shared model
- The roadmap, whether or not you build it with us
- The decision to stop, at the end of any phase
Six places an AI consulting engagement starts
Most clients arrive holding one of these. You do not need to know which before the first call.
AI readiness assessment
Two weeks. We audit the systems and the data, score every candidate use case, and tell you which ones are actually buildable now.
AI strategy and roadmap
A phased plan with cost, risk and a target metric per phase. Sequenced so the cheapest evidence comes first.
Proof of concept
One use case, built and deployed to staging, measured against your data. Cheaper than a second strategy phase and far more decisive.
AI implementation services
The full build, by the engineers who wrote the roadmap. Your repository, your cloud account, your CI.
Data and retrieval foundations
Pipelines, vector storage and permissions, so a model can read your documents without reading past someone's access rights.
Agents and workflow automation
Multi-step work across your tools, with approval gates. A different build and a different budget shape, so it has its own page.
The AI readiness assessment, and what it hands back
Two weeks, fixed fee, four artefacts. It is deliberately the cheapest way to find out whether the rest of the work is worth doing.
Systems and data audit
Where the data lives, who owns it, how current it is, and whether it can be read without breaking someone's permissions. This is the step that decides most use cases, and it is the step most assessments skip.
Use case scoring
Every candidate scored on two axes: the size of the metric it moves, and whether your data can support it today. Scored separately, so a valuable idea with no data behind it does not average out into looking fine.
Costed roadmap
Phases in the order that produces evidence soonest, with build cost and running cost separated. Running cost is the number that breaks AI budgets in year one, so it gets its own line.
Working prototype
One use case, built and deployed to staging inside the engagement, with an evaluation report against your own records. You decide on something you have used, not on a recommendation you have read.
Seven questions to ask any AI consulting firm
Ask all four kinds of vendor the same seven questions. The answers sort them faster than any credentials deck.
| Ask them | InApps | Global consultancy | AI product agency | Independent consultant |
|---|---|---|---|---|
| Who builds what you recommend? | The same engineers, start to production | A separate delivery arm, or your own team | They build, but rarely assess first | Nobody. Advice only |
| When do we see something working? | Week 4 to 5, on staging | After a strategy phase, often month 3+ | Fast, but with no baseline to judge it against | Not part of the scope |
| What does phase one produce? | Audit, scored use cases, costed roadmap, prototype | A roadmap document and a business case | A demo | A report |
| Will you tell us not to build it? | Yes, and it happens in phase one | Rarely. The next phase is the product | Rarely. The build is the product | Often, but with no build to compare against |
| Who watches it after launch? | Us, on an ongoing engagement | Handed to your team at go-live | Support ticket, not a quality baseline | Engagement has ended |
| What is your security posture? | ISO/IEC 27001:2022 certified, audited annually | Strong, and priced accordingly | Varies widely. Ask for the certificate | Individual arrangement |
| How is the work priced? | Fixed-fee assessment, then per phase | Partner-rate day rates | Project fee, build only | Day rate |
Six weeks from first call to a go or no-go
Discovery and assessment
We look at the systems, the data and the metric you want moved.
Roadmap and architecture
The plan, the sequence, and what each phase costs to build and to run.
Build and validate
One use case, on staging, measured against your own records rather than a benchmark.
Decide, then scale
A go or no-go on evidence. If it is a no, that is a result, not a failed engagement.
How we prove it works before you scale it
A demo proves a model can do something once. These four steps are what tell you it will keep doing it in March.
Baseline the current number
- How the work is done today
- Time, cost and error rate now
- Agreed before anything is built
Build an evaluation suite
- Test cases from your own records
- Pass thresholds set with you
- Re-runnable on every change
Model the running cost
- Token and infrastructure spend at real volume
- Cost per successful outcome
- Assumptions written down, not implied
Watch it after launch
- Quality tracked against the baseline
- Evals re-run on every model upgrade
- Alerting when a score drifts
What decides the cost of an AI consulting engagement
We do not publish a rate card for this work, because the honest answer depends on five things. Here they are, in the order they usually matter.
| Cost driver | Cheaper when | Expensive when |
|---|---|---|
| State of your data | One current system, documented, with an API | Four sources that disagree. Reconciling them is a project before the AI work starts |
| How many use cases | One, with a metric everyone already agrees on | A department-wide scan. More audit, more interviews, more scoring |
| Running volume | Hundreds of calls a day, handled by a small model | Millions a month. Token spend passes the build cost inside year one, then repeats |
| Cost of a wrong answer | A user notices, retries, and nothing was lost | Money moved or a customer saw it. Evals and approval steps get much heavier |
| Regulatory load | Internal tool, no personal data, no external reporting | Personal or financial data, audit trails, residency constraints |
What we build on, and why none of it is a lock-in
Model-agnostic on purpose. Providers change their price and their quality on their own schedule. You should be able to move without a rewrite.
Models
OpenAI, Anthropic Claude and Google Gemini. Llama and Mistral on your own hardware where data cannot leave the network. Routing sits behind one interface, so a price rise is a config change and an eval re-run.
Frameworks
LangChain and LangGraph for orchestration. Pinecone, pgvector and Elasticsearch for retrieval. Ragas and custom harnesses for evals. We will also tell you when none of them is needed.
Infrastructure
AWS Bedrock, Azure OpenAI and Google Vertex, in your account rather than ours. Traced end to end. Same CI, review and rollback discipline as the rest of your stack.
Ask the people who stayed
Every quote below is from a verified review. None of them were written by us.
“Clear expectations, rules that actually hold, delivered on time. And the people are genuinely easy to work with, not just professionally, but as humans.”
Worth the call, or worth a no
AI consulting is the wrong shape for plenty of good projects. Here is where it works, and where we will point you elsewhere.
This works well if
- You have a metric you want moved and no strong view yet on whether AI is the way to move it.
- A previous pilot stalled and you want to know whether the idea or the execution was the problem.
- You want the people who write the roadmap to be accountable for shipping it.
- You need a decision on evidence within about six weeks, not a strategy programme.
This is not a fit if
- You already know exactly what you want built. Skip the assessment and go straight to AI agent development or generative AI integration.
- You need a board-level document and no code. A strategy house will serve you better, and we will say so.
- You want the assessment fee waived up front. It is what keeps this from being a free pitch, and it is credited against the build.
- Your data cannot leave your network and there is no budget for on-premise infrastructure. Worth a call, but say it early.
Answered without the hedging
Still deciding?
Send us the metric you want moved. We will tell you whether an assessment is worth your money, including when it is not.
Book a discovery callWhat is AI consulting?
How much do AI consulting services cost?
How do we evaluate AI consulting firms?
What is an AI readiness assessment?
What if the assessment says AI is not the right fix?
Who owns the code, the prompts and the data?
How is this different from an agency wrapping ChatGPT around our product?
Do you build AI agents and LLM features too, or only advise?
Tell us the metric you want AI to move
No pitch and no credentials deck. We come back with a scope for the assessment, what it costs and what it returns. If AI is the wrong tool for the metric, you will hear that on the first call.