AI consulting services for software companies
My AI consulting services are for software companies that know they should be doing something with AI and haven’t worked out what. I help you pick the two or three use cases that will pay for themselves, then I build them with your team – measured on ROI, not demos.
I’m an engineering lead first. I don’t hand over a strategy deck and leave. Every engagement starts with a two-week sprint, and by the end of it something real is running against your own data.
The usual starting point
The board has asked about AI. A competitor has shipped a chat panel. Three engineers have each built a prototype on a different provider, none of them has an owner, and nobody can say what any of them cost per user or whether customers want them.
That isn’t a strategy problem so much as a prioritization problem. Generative AI is cheap to demo and expensive to run well. The work is deciding which ideas survive contact with real data, real latency budgets and real unit economics – and killing the rest early.
What AI strategy consulting looks like when I do it
AI strategy consulting is useful only if it ends in a decision and a working system. Here’s what you get:
- A use-case shortlist scored on money. For each candidate: the workflow it changes, who uses it, how often, what it saves or earns, and what it costs per call at realistic volume. Ideas that can’t name a metric don’t make the list.
- A data reality check. Most AI ideas fail on the inputs, not the model. Is the data accessible, permissioned, current and clean enough? If it isn’t, that’s the first project.
- A build-vs-buy call per use case. Plenty of AI features are better bought. An off-the-shelf support tool or coding assistant can beat a custom build for a fraction of the effort. See build vs buy software.
- A working first feature, built inside your codebase with evals, cost tracking and a fallback path, so your team inherits something they can extend.
- A plan your team can own – which provider abstractions to keep, which prompts and datasets to version, and who is accountable for quality once I’m gone.
Where the money usually is
In SaaS, the AI work that pays back fastest is rarely the flashy part. It tends to fall into a few patterns:
- Turning unstructured input into structured data – parsing documents, emails and forms into fields your product already understands. Easy to evaluate, easy to price.
- Cutting time in an existing workflow – drafting replies, summarizing tickets, classifying and routing. See AI for customer support.
- Answering questions over your own content – help docs, policies, account data. That’s RAG development or, for internal use, an AI knowledge base.
- Letting agents operate your product – exposing it through a custom MCP server so customers’ AI tools can use it directly.
Fully autonomous agents are the right answer less often than the hype suggests. When they are, I build them with tight scopes and human checkpoints – see AI agent development.
How an engagement runs
- Intro call. You tell me what’s been tried and what the pressure is. I tell you whether I’m the right person.
- Two-week sprint. I interview the people closest to the workflows, look at the data, score the use cases, and build the most promising one far enough to measure. If you want just this part as a fixed piece of work, that’s an AI readiness assessment.
- Build. If the numbers hold up, I embed for a fixed number of days a week and ship it into production with your engineers – evals in CI, cost and latency dashboards, feature flags.
- Handover. Prompts, eval sets and runbooks live in your repo. Your team can change models or providers without me.
Why an engineer as your AI consultant
Most of what makes an AI feature succeed is ordinary engineering: data pipelines, permissions, caching, observability, graceful failure. A consultant who can’t open a pull request will underestimate all of it.
I’ve spent 20+ years shipping production software, including platforms with millions of users. At Pepkor NexTech I worked across software engineering, solutions architecture and cloud infrastructure while contributing to their AI journey. I also build Craaft, a kanban board where humans and AI agents work the same cards through MCP or a REST API – so the agent-integration problems I advise on are ones I live with in my own product.
When you don’t need me
If you want an engineer to bolt a single chat endpoint onto an app, a good contractor is cheaper. If you need a large team to build a model from scratch, you need an ML research group. And if the honest answer is that AI won’t move your numbers this year, I’ll say so in the first two weeks rather than bill you for a roadmap.
Questions
What does an AI consultant actually do?
A good AI consultant identifies where AI will change a business metric, checks whether your data can support it, decides whether to build or buy, and gets the first use case into production with evaluation and cost controls. I do all of that hands-on, inside your codebase, rather than handing over recommendations.
How do you measure the ROI of an AI feature?
Against the workflow it changes: time saved per task, tickets deflected, conversion lift or error rates, minus the running cost of inference, retrieval and maintenance. I set the baseline before building, so the comparison is honest. My AI ROI guide covers the method.
Do we need a data science team first?
Usually not. Most SaaS AI features use hosted models through an API, so the work is software engineering, data access and evaluation rather than model training. You need someone who owns quality and cost after launch, which can be an existing senior engineer.
Which AI provider should we use?
It depends on the task, your latency and privacy requirements, and price, which changes often – check each provider’s current pricing page. I build behind a thin provider abstraction with an eval suite, so switching models is a measured decision rather than a rewrite.
How long before something is in production?
The first two-week sprint ends with a working build on your data. A narrow, well-scoped feature can reach production users within a few weeks after that; broader agentic work takes longer because the guardrails and evals take longer.
Not sure which AI idea is worth building?
Bring the list. In a two-week sprint we score it on ROI, build the best candidate against your data, and decide together what happens next.
GMT+2 · Full EU overlap · US East mornings