Dumb AI Consultants Won't Last Long

01 /The Concept

Operand's stated mission is blunt: "We're going to kill McKinsey." In practice the target isn't one firm but an entire industry.

The startup built an AI consultant for e-commerce brands and retail chains – one that delivers the kind of strategic recommendations consulting firms charge a small fortune for, in a fraction of the time.

Operand's AI can advise on pricing optimization, promotional strategy, advertising budget allocation, and customer engagement mechanics. The team is also actively researching which additional advisory topics matter most to potential clients.

Getting a useful answer starts with asking a useful question – not a trivial task, but the AI doesn't generate generic advice divorced from reality. The old joke applies here: when mice ask a wise owl how to survive hawks, and the owl says "become hedgehogs," the mice ask how. "Don't burden me with tactical details," the owl replies, "I do strategy"

To avoid that trap, clients must first connect their real data: Shopify or Amazon seller accounts, Salesforce, Google Analytics, ad dashboards, spreadsheets – all the sources where actual sales and marketing performance lives. The recommendation engine works against specifics, not generalities.

Even then the answer isn't instant. Preparing a recommendation can take a few hours: the AI analyzes the raw data, builds a business model, stress-tests it across different parameter combinations, then proposes an optimal configuration and the steps to reach it.

Critically, all outputs are reviewed and validated by Operand's own domain experts before they reach the client. This catches AI hallucinations and creates a feedback loop for continuously improving the model's accuracy.

Operand is currently in pilot mode. Early clients report results including hundreds of thousands of dollars saved on ad spend, 3x conversion rate improvements, 4x revenue growth, and 20 hours per week freed from manual reporting. Case study documentation is thin for now.

The startup is currently in Y Combinator, which contributed $500K in funding. The platform announcement was published on the YC site four days ago. Retail is the starting vertical, but Operand has made clear it's just the launch point.

02 /Why It Matters

The business consulting market is large, growing, and well worth disrupting. Global revenue was around $751B in 2014, climbed to $876B by 2020, and crossed $1 trillion in 2024.

On the surface the entry barrier looks low. ChatGPT already handles business questions and produces broadly reasonable answers. So build a wrapper, write some prompts, call it an AI consultant.

That approach produces a mediocre product that resembles a weak consultant – someone who clips relevant-sounding passages from business books and recites them back. Real expertise doesn't work that way. An expert has read the books and studied the cases, but the value isn't in the raw material – it's in the internal model they've built from connecting those inputs in their own way. They take a client's situation and run it through that model. The model is the asset.

Amigo ([covered here](/review/ii-experty-eto-sovsem-ne-ii-sotrudniki)) understood this and built a platform where experts, coaches, and consultants can create AI doubles of themselves – trained not just on what they know but on how they think. Each double answers questions differently, according to its model's logic, and can reason about situations the original expert never explicitly addressed. Investors found the approach compelling: Amigo raised $6.3M in its first round.

Operand's equivalent of a world model is the financial model it builds from a client's data before generating any advice. But even that may need further personalization – financial models can be built on fundamentally different strategic philosophies. Trading analogies apply: an aggressive growth strategy and a conservative one will generate very different recommendations from the same underlying data. Personalizing the AI's strategic lens seems like a logical next frontier for Operand.

A second structural gap goes deeper: good business management isn't cyclical. The consulting-as-quarterly-report rhythm exists because traditional consulting is slow and expensive. If the work is handed to AI, the monitoring and recommendation cycle becomes continuous. That's a different paradigm – not "call in the consultant when something feels wrong" but embedded AI modules that watch specific processes in real time and generate micro-recommendations as conditions evolve.

Quantum Rise ([reviewed previously](/review/na-jetom-uzhe-ne-stydno-zarabatyvat)), founded last year and funded with $15M in its first round, is building exactly that version: a "consulting firm for the new era" that doesn't deliver reports but installs AI products that embody its strategic advice and operate continuously inside client workflows. If the AI can tell you what to do, it should go do it. Live consultants were never able to offer that.

03 /Opportunities

A [recent review](/review/nash-produkt-luchshe-jeto-nedostatochno-ubeditelno) noted the brief euphoric period that followed ChatGPT's release – the belief that a smart wrapper around a base model was enough to build a great AI product. That produced a wave of first-generation AI support bots that can answer questions but can't solve problems. Lorikeet ([covered here](/review/nash-produkt-luchshe-jeto-nedostatochno-ubeditelno)) responded by building a customer service AI designed from scratch around problem resolution, not FAQ retrieval, and led with the positioning: "our AI agents do what other AI agents can't."

Operand's value is in the same vein – its AI consultant does something a ChatGPT wrapper can't. Quantum Rise goes further: it delivers what a human consultant can't. The progression is directional.

Paul Graham put it memorably when he said the only role he sees for AI in his essay-writing process is to draft the essay on the given topic – so he can eliminate from his own writing everything the AI managed to say.

The practical implication: the relevant direction right now is building second-generation AI products that are more capable not because they run newer base models, but because they're architected differently than first-generation ones.

Where is AI already deployed in your area of interest – and what would a second-generation architecture do that first-generation tools demonstrably can't? The follow-on question is whether that architecture can eventually do things no human practitioner currently does, which is the threshold that defines category-creating companies.

For those who'd rather start with a clear target than a blank canvas: the consulting market is enormous, and Operand and Quantum Rise are both still early.