AI Product Consulting
Get a first view First view
Practical AI for small and medium-sized businesses

Put AI to work in the parts of your business that actually matter.

We find where AI saves your business time or manual work, determine what should be built, and build it through to production. A vague ambition, a specific opportunity, or a process you already want automated: we pick up where you are.

Get a first view - free, 30 minutes
01 / What practical AI looks like

Systems that remove real operational effort, not AI demos.

20 hours →
2 hours/week
An agent that reconciles supplier invoices against bank transactions for a finance team.
our part Scoped with the finance team, designed the system, built and deployed it. Start to finish, with no handoffs.
outcome 18 hours of repetitive work removed each week.
outcome The team now only sees the cases that genuinely need judgement.
outcome Running in production inside their existing tools, on their keys.
How it runs, per invoice
supplier portals
shared inbox
file storage
01 Collects new documents, knows where to look
02 Reads each one - any layout, any format
03 Matches it to a bank transaction
confident enough?
yes ↓ no ↓
system acts
Posted to the accounting system
person decides
Escalated with the evidence attached
≈ 2 h/week lands in the right-hand column. Everything else clears itself.
2,400+
explainer videos, no film crew

Production work that used to be commissioned one piece at a time now runs as a pipeline, with review kept in human hands.

7 seconds
to find the relevant precedent

A corpus nobody could read end to end became searchable by the actual legal question, with the source passage attached.

02 / Where AI tends to help an existing business

Six kinds of work where the effort is real and the improvement is measurable.

If your team repeatedly reads, checks, copies, decides, routes, reconciles or follows up, there is probably something here.

Finance and reconciliation
Invoices, statements and bank lines matched by hand in a spreadsheet.
Collected and matched; only unclear cases queued.
Inbox and customer operations
Every message read by a person, then re-typed into the CRM or ticket system.
Classified, drafted, routed - judgement calls still reach you.
Documents and data entry
Documents opened one at a time and keyed in, format by format.
Extracted with a confidence score and a trail back to the source.
Reporting and analysis
The same numbers pulled out of four systems every week, by hand.
Assembled on schedule, commentary drafted, figures traceable.
Internal knowledge
Answers that live in old contracts, manuals and two people's heads.
Askable in plain language, with the source passage shown.
Approvals and handoffs
Requests drifting through four inboxes and two systems that never talk.
Policy-checked, routed, chased when they stall, status visible.
03 / The operating principle

The goal is not to maximise AI. It is the simplest reliable system that improves the business.

Ordinary software where rules are stable. AI where the work is messy language or judgement. A person where being wrong is expensive. Drawing those boundaries is the real design work, and sometimes the answer is a rules engine, not a model.

Rules
Deterministic software does the predictable parts

Fetching, moving, calculating, posting. Anything you could write as a table belongs in a table: cheaper, faster and auditable. Your existing systems keep doing what they are already good at.

AI
A model handles the unstructured and the ambiguous

Free-text emails, documents in twelve formats, cases nobody wrote a rule for. This is where AI earns its place: input that varies and a decision that is genuinely a decision.

Human
A person approves where being wrong is expensive

Payments, anything a customer sees, anything the system is not confident about. The threshold needs to be set right: a wrong auto-decision often costs far more than an unnecessary check.

How one piece of work moves through it
rules ai human
01 · rules
Work is collected
Pulled from inboxes, portals or storage on a schedule.
02 · ai
The model reads and decides
Free text, mixed formats, cases nobody wrote a rule for.
03 · rules
Confidence is checked
Above the threshold it acts; below it, it stops and asks.
04 · human
A person approves the rest
Only the unsure and the expensive, with the evidence attached.
05 · result
Written into your systems
CRM, ledger, tracker - wherever the work already lives.
Malte Poppensieker
04 / Why one contact owns the whole path

No handoff between the thinking and the building.

Your contact through the whole project is Malte Poppensieker: he analyses the problem, designs the architecture and manages the build through to production. Your context does not get passed from a consultant to a product manager to an engineer, losing a little meaning at each step.

An SME with a contained operational problem usually needs senior commercial judgement, system definition and hands-on implementation, not a consultancy delivery team funded around it.

Business
Consulting and analysis
Consultant at Capgemini
MBA, University of Cambridge
Definition
Product and system design
Product Manager at Amazon
General Manager at Nerdy
Delivery
AI engineering
MSc Computer Science
Founder and Engineer of an AI learning platform

Depending on the project, experienced developers from an established network do the implementation and bring specific competencies where the work calls for them - named openly, with overall accountability staying with your contact. Where a programme genuinely needs a standing in-house team, we will say so on the first call.

05 / Selected work, by the problem it solved

The same capability applied to very different operational problems.

Systems built for different operational problems: making information usable, producing and reviewing work at scale, removing manual steps, and connecting judgement with action.

Legal

Precedent finder for litigators

A system that scans a huge corpus of completed court cases and surfaces the precedents that bear on one specific legal issue - in seconds, not billable hours.

  • Retrieval at scale
  • Legal corpus
  • Search
Media pipeline

2,400 explainer videos, no film crew

A fully automated pipeline for producing explainer videos end to end. It has been used to create more than 2,400 of them at scale.

  • Generation pipeline
  • Video
  • Automation
Generative

Character animation from one prompt

A fully automated character animation process driven by a single prompt - from written intent to finished animated motion with no manual keyframing.

  • Generative pipeline
  • Animation
  • Prompt-driven
Monitoring

A news scanner with an opinion

A scanner that processes a large number of news sources continuously and surfaces only the specific content its owner actually cares about.

  • Content scanner
  • Monitoring
  • Filtering
Content ops

Human review, at machine speed

A workflow that pre-generates learning content and routes it through structured human review, so quality stays in human hands while volume stops being the bottleneck.

  • Human in the loop
  • Workflow
  • Content ops
Legal

RoleAI Engineer & Product Manager

A retrieval system that scans a large corpus of completed court cases and surfaces the authorities bearing on one specific proposition - in seconds rather than hours.

We worked with a non-technical founding team from requirements through to a working system.

Problem to product

Research you can actually cite

Problem

Finding the authorities that bear on a specific proposition is slow, and the real exposure is not missing a case - it is citing one that has since been doubted or overruled.

Why AI

Keyword search fails here for a structural reason: the same legal principle is expressed in different words across decades of judgments. Matching meaning rather than strings is exactly what embeddings and retrieval are for. This was a case where AI was clearly the right tool, and we said so early.

Product decision

A ranked list of documents would have been easy and useless. Lawyers need to know how much weight an authority carries and what happened to it afterwards. So every result shows its weight on the proposition, its subsequent treatment, and the passage actually relied on. Nothing has to be taken on trust.

What we built

Retrieval across 1,200+ authorities with jurisdiction, court and date filters. Ranked results with treatment history. Passage-level citation. A case reader that puts the ratio, the relied-on passage and the later treatment on one screen.

Architecture

Python, Django and AWS Lambda, with LangChain orchestration over a vector index of the corpus. React front end.

Result

Question to citable position in a single pass. Taken from prototype to working system with a founding team that had no technical background - we owned the requirements, the product design and the code.

Skills and deliverables
  • Retrieval Augmented Generation
  • Python
  • LangChain
  • Django
  • Artificial Intelligence
Media pipeline

RoleAI Engineer & Product Manager

A fully automated pipeline that produces explainer videos end to end - script, narration, animated visuals, render - with no animation team. More than 2,400 videos produced through it.

We designed the architecture and built it.

Problem to product

Volume without a studio

Problem

Producing explainer videos conventionally means a script writer, a voice artist and an animator for every single one. Output is capped at whatever a small team can hand-make in a week, and covering a full curriculum that way was never going to happen.

Why AI

Script writing and narration are generation problems that language and speech models handle well. Sequencing and pacing are orchestration. Those parts genuinely needed AI.

Product decision

The visuals did not. Generative video is impressive and wrong too often - and in a maths explainer, a plausible-looking but incorrect diagram is worse than no video at all. So visuals are rendered deterministically from the underlying mathematics with Manim. AI where it is reliable, code where correctness matters. That split is the reason the pipeline could be trusted at volume.

What we built

Topic in, finished video out: LLM script generation with review, programmatic animation, synthesised narration, render and publish. A job queue, so throughput scales with compute.

Architecture

Python throughout. LLM generation, Manim for animation, ElevenLabs for voice, Celery and n8n for orchestration and retries.

Result

More than 2,400 videos produced. Output limited by compute rather than headcount - and the quality bar held, because the part that had to be exact was never left to a model.

Skills and deliverables
  • Artificial Intelligence
  • Python
  • LLM Prompt Engineering
  • AI Development
  • ElevenLabs
Generative

RoleAI Engineer

A fully automated character animation process driven by one written prompt - from plain-English intent to finished motion, with no manual keyframing.

The hard part was never generating movement; it was keeping a character recognisably itself across takes. Locked reference sheets and seeded, reproducible takes solved that.

Problem to product

Consistency, not novelty

Problem

Character animation is expensive because it is hand-posed. Every second of motion is somebody’s afternoon, which puts it out of reach for anyone producing at volume.

Why AI

Generative video can produce plausible motion from a description, which removes the keyframing. That capability is real, and recent enough to be worth building on.

Product decision

The naive version - prompt in, clip out - is a toy, because the character looks subtly different in every take and nothing can be iterated. Consistency and reproducibility were the product, not motion quality. So each character gets a locked reference sheet, every take is seeded, and artists steer with dials - motion amount, reference adherence, duration - rather than by rewriting prompts and hoping.

What we built

A prompt-to-motion pipeline with per-character reference sheets, a style layer applied at render time, motion dials, and take management with seeds, so any result can be reproduced or rerolled deliberately.

Architecture

Generative video models behind a Python orchestration layer, with a render queue and a studio interface over the top.

Result

A brief like "four springy steps toward camera, antenna wobbling on each landing" becomes finished motion without a keyframe being set - and the same character still looks like itself in the next take.

Skills and deliverables
  • Artificial Intelligence
  • AI Development
  • Generative Design
  • AI Video Generation
Monitoring

RoleAI Engineer & Product Manager

A monitoring system that reads 140+ news sources continuously and surfaces only the content its owner actually cares about.

Watch areas are described in plain English rather than as keywords. The system reads each article, judges relevance against those areas, and explains why an item matters with the source passage attached.

Problem to product

Signal, not volume

Problem

Keyword alerts fail in both directions at once. They miss the article that matters because it used different words, and they bury the reader in articles that merely mention the term. The result is a feed nobody trusts and therefore nobody reads.

Why AI

Relevance to an interest like "regulatory pressure on private credit" is a judgement about meaning, not a string match. Making that judgement over a thousand articles a day is exactly the kind of work that only became possible recently.

Product decision

Two calls decided the product. First, watch areas are described in ordinary language, so the user states an interest instead of guessing which words a journalist will use. Second, nothing is surfaced bare: every item carries a short why-it-matters and the passage it was drawn from. An unexplained AI verdict gets ignored on the first false positive - showing the reasoning is what makes it usable.

What we built

Continuous ingestion across 140+ sources on a schedule. Per-article relevance judgement against the user’s watch areas, sentiment scoring, and a written rationale with the source passage. A review interface with tiering, alerts and coverage tracking.

Architecture

Python, LLM classification and summarisation, vector search over ingested content, scheduled ingestion with deduplication.

Result

Over a thousand articles a day reduced to a handful worth reading, each one auditable back to the sentence that triggered it.

Skills and deliverables
  • Artificial Intelligence
  • Python
  • Retrieval Augmented Generation
  • LLM Prompt Engineering
  • AI Agent Development
Content ops

RoleAI Engineer & Product Manager

A workflow that pre-generates complex learning content and routes it through structured human review, so quality stays in human hands while volume stops being the bottleneck.

Subject experts approve, edit or reject in purpose-built tooling with live LaTeX, geometry and graph preview.

Problem to product

Quality stays human

Problem

Covering a curriculum needs far more questions, explanations and worked examples than subject experts can author by hand. But in education a wrong worked example is worse than a missing one - it teaches the mistake, confidently.

Why AI

Volume. Generation gets a draft of every item in place at a rate no team of authors can match, which turns an authoring problem into a reviewing problem.

Product decision

Generation and publication had to be separate systems, not two settings of one. Everything generated lands in a review queue and only becomes active when a human says so. That is a deliberate cap on throughput, and it is the reason the content can be sold with a guarantee attached. Reviewers also needed real tooling - reading raw model output is slow and error-prone, so the builder renders live LaTeX.

What we built

Bulk generation of explanations, questions and solution steps - including interactive geometry and graph elements, images and complex formulae. A review queue with explicit state. Purpose-built reviewer tooling with live LaTeX preview, question and explanation builders, and a full audit trail.

Architecture

Python and Django, React front end, LLM generation, and a workflow engine with state and audit logging.

Result

Quality stays in human hands and volume stops being the constraint. We defined the workflow with the review team and built the generation layer, the queue and the tooling.

Skills and deliverables
  • Artificial Intelligence
  • Python
  • Django
  • React
  • AI Product Management
Space here for yours. Bring the problem.
Start a project
06 / From question to production

Enough analysis to avoid building the wrong thing. Enough engineering to leave you with a working system.

Four stages. If the earlier ones are already clear in your business, we start further down.

01 Understand the work and the economics

We sit with the people who do the work today and follow it end to end, including the exceptions they handle without thinking about them. What it currently costs in time and error is part of this, not a separate business case exercise.

02 Define the system and its boundaries

What is rules, what is AI, where a person approves, and what the system is deliberately not allowed to do. Architecture and how quality gets measured are settled in writing before production code exists.

03 Prove the difficult part

A working thin slice on representative data: your real documents, your real messages, your awkward cases. If the hard part does not hold up, you find out in week two rather than at the end.

04 Build, deploy and support

Regular demos during the build, then live inside your tools, on your repository and your keys. For 90 days after go-live, anything that breaks is still our problem.

07 / Three ways to start

Start across the business, on one process, or with the build itself.

Most people start in the middle. If the process is already clear in your business, skip ahead — if none of them is, start with a call.

Across the business

AI Opportunity Assessment

For when you don’t yet know which process to start with. Every candidate found and ranked against the others, the economics compared between them, and an explicit list of the ones to drop.

Interviews across the teams that do the work
A ranked shortlist with the economics beside each
A recommended first project, and the ones to drop
Fixed fee · plus VAT
£2,500
Book a call
One process analysed

Free process analysis

Send the one workflow you’d most like to stop doing by hand. We’ll analyse it and you get the findings back in writing: what it costs you today, what a first improvement could cover, implementation pricing and a timeline. Or an honest no.

One process
A video session with the people who run it
Results in writing, in two working days. Yours to keep, whether or not you build with us
Your price today
£0
now — £1,000 from 1 December, plus VAT
Request a free process analysis
Spec, architecture and build

The production system

The whole path from a defined system to something running inside your tools: full spec, architecture, the difficult part proven on your real data, then built and deployed.

System definition and architecture, settled in writing
A thin slice on your real documents before the build
Live on your repository and your keys, 90 days support
Price
Fixed price after scoping
Request more information
08 / Bring the business problem

You do not need a brief, or the right answer, before getting in touch.

Bring the vague ambition, the specific opportunity, or the workflow you already want improved.

Thirty minutes on the actual process and where it hurts, and you leave with a sensible next step. If ordinary software or nothing at all is the better answer, we will say that.

Book the free 30-minute conversation
Free · no deck · no sales sequence
Or email malte [at] aiproductconsultant.com
Will this work with the systems we already run?

That is the normal case. Systems are built around the tools you already use - accounting system, CRM, mail, file storage - through their APIs, and where there is no API, through the interfaces a person would use. Nothing needs replacing to get started.

What about sensitive data and security?

Only what a specific request needs goes to a model, and what goes is logged. No training on your data. Processing can run in your own cloud environment, in a specific region, or entirely without external model providers if that is the requirement - decided during scoping rather than discovered in production.

Where does human review fit?

Wherever a mistake is expensive. The system acts on its own where it is confident and the downside is small, and asks a person where it is not, with the evidence attached so the check takes seconds. That threshold is a design decision we make together, not a default.

What does it cost, and how is it quoted?

It depends on the workflow, which is why there is no number on this page. After the first call and a scoping session you get a fixed price for the specific piece of work, and you decide with that in front of you. Running costs are estimated at the same time, so the operating bill is not a surprise.

How long does something like this take?

A contained operational system is usually weeks, not quarters. The difficult part gets proven on your representative data early, and from there you see it running at regular demos - there is no long stretch where you hear nothing.

Who owns the code and the accounts?

You do. The repository, the keys and the documented architecture are yours from day one. Nothing here is designed to keep you dependent on us.

Who supports it after go-live?

We do, for 90 days after go-live, at no extra cost. After that, either an ongoing arrangement or your own team - the documentation is written for someone else to pick it up.

What is a poor fit for this model?

Enterprise-wide transformation programmes, anything needing a permanent in-house team, and general developer capacity. For a defined operational problem inside an existing business, one contact who runs the analysis, the architecture and the build is usually faster than a large team that has to coordinate. Where a project grows or needs a specific skill, developers from an established network join for exactly that part.

Malte's AI

Three minutes, a few questions, then you have it in writing.

I'm a tool Malte built. I'll ask you a few questions, then tell you honestly whether AI is worth it for you — even if the answer is no. About three minutes.
Tell me in two or three sentences what this is about.