AI Consulting
AI consulting is the practice of bringing machine learning and language models into a company's operations. Demand grew rapidly once boards began asking for an AI strategy and vendors began selling one. The standard engagement picks a model provider, wraps a chatbot or copilot around an internal process, and runs a pilot. The question that decides whether any of it pays is often skipped: what decision is this system supposed to improve, and does the data exist to improve it? Our design rule is that AI layers on top of deterministic software and machine learning you own rather than substituting for them; the full argument, including what substitution costs in token spend and reliability, is in our model lock-in entry.
How it actually works
The off-the-shelf version starts from the model and works backward. A vendor is chosen, an interface is stood up, usage is billed per token, and success is measured in adoption: how many people used the tool, how many queries it answered. The underlying data, the CRM records, the analytics, the customer feedback, is taken as it sits.
The test that decides whether an engagement pays runs before any vendor is chosen: name the decision the system is supposed to improve, price what improving it is worth per year, and compare that figure to the build cost plus the token run-rate at projected volume. A proposal that cannot pass this test on paper does not become profitable in production. A language model pointed at flawed data returns flawed answers faster and with more confidence, and a large share of what gets pitched as AI is work a lookup or a deterministic rule handles at a fraction of the cost. A model earns its place upstream: extracting features from raw and qualitative data, finding structure a team could not see, and feeding predictions into a decision someone makes.
Engagements that skip that question follow a recognizable pattern. The token bill scales with usage whether or not results follow. The build hard-wires one vendor's model into the workflow, so switching later means rebuilding. And the pilot demonstrates that the tool works without establishing that it changed a number the business tracks. The scale of that pattern is measured: in 2025, MIT's NANDA initiative published findings that 95% of generative AI pilots delivered zero measurable impact on the P&L, despite $30 to $40 billion in enterprise investment. We cover the mechanism in [AI and bad data](/insights/why-strategic-initiatives-fail#ai-and-bad-data).
In practice
For Digital Realty and Equinix, we built the systems architecture behind taking new software to market for enterprise buyers. The work covered one pipeline from data sources to decisions, designed so the intelligence layer served the go-to-market.
Where we come in
This work sits inside our Revenue Systems Architecture practice, built for businesses from new startups to the Fortune 15, including Digital Realty and Equinix. We demonstrate ROI before each build, everything is scoped up front, and the systems are portable: you are never locked to a single model, platform, or vendor. Where a deterministic rule outperforms a model call, we ship the rule and reserve the models for the work that genuinely needs them.
Start a Revenue Health Pre-Assessment →See it in action
Related terms
- Machine learning (ML)
- Models that learn patterns from data to classify, rank, or predict. In revenue work, it is the layer that turns raw and qualitative data into features and forecasts.
- Large language model (LLM)
- A model trained on text that can read, summarize, and generate language. Priced per token, which means its cost scales with usage rather than with outcomes.
- Feature engineering
- Turning raw data into the specific inputs a model needs. It is where most of the real work in an ML project happens, and where most pitched projects have done none.
- Context engineering
- Structuring what a language model is given, the data, instructions, and constraints, so its output is grounded in your business instead of in its training data.
- Deterministic rule
- Logic that produces the same output for the same input, a lookup, a threshold, a regex. When a rule can do the job, it beats a model call on cost, speed, and auditability.
- Vendor lock-in
- A build that only works with one provider's model or platform. It converts a technology choice into a permanent cost, because leaving means rebuilding.