Start with data, not algorithms

AI consulting for SMBs must start with data audit and process mapping — not model selection — because most SMB AI failures stem from insufficient data infrastructure, not wrong algorithms. A small business with 50 employees and three years of operational data has fundamentally different AI options than an enterprise with millions of labelled transactions.

The honest first question for any SMB AI engagement is: “Do you have the data to support what you want to build?” In our experience, a substantial share of initial SMB AI enquiries require several months of data collection before any model development can begin — an observed pattern across our engagements, not a benchmarked rate. This is not a sales objection. It is an engineering reality.

What’s realistic for AI consulting at SMB scale?

The table below summarises the use cases where SMB data volumes are typically sufficient, drawn from observed patterns across our consulting work. Timelines and ROI windows are planning heuristics, not benchmark figures.

Use case	Data requirement	Typical timeline	Realistic ROI period
Document classification and routing	500+ labelled examples per category	2–4 months	6–9 months
Demand forecasting	2+ years of sales history, consistently recorded	3–5 months	9–12 months
Customer churn prediction	12+ months of customer behaviour data	3–6 months	6–12 months
Visual quality inspection	1,000+ defect images per defect type	4–8 months	12–18 months
Chatbot and FAQ automation	Existing support logs, 6+ months	2–4 months	3–6 months

The ROI window we see for custom SMB AI work sits in the 6–18 month range as an observed pattern across engagements. Consultants promising faster returns are typically selling pre-built tools, not custom solutions. Pre-built tools can be valuable — they are faster and cheaper — but they solve generic problems, not your specific operational bottleneck.

What’s not realistic yet at SMB scale?

Some AI applications that enterprises deploy successfully are not viable for SMBs because of data-volume constraints:

• Personalised recommendation engines — meaningful personalisation typically needs millions of interactions. With a few hundred customers, well-designed rule-based recommendations outperform ML.
• Natural language understanding from scratch — training custom NLU models requires volumes SMBs do not have. Fine-tuning pre-trained transformers (via Hugging Face, PyTorch, or ONNX runtime targets) is viable; training from scratch is not.
• Predictive maintenance — requires sensor traces from actual equipment failures. Most SMBs have too few failure events to train reliable prediction models.

The pattern here is not “AI doesn’t work for small businesses”. It is that the techniques which dominate enterprise case studies depend on data assets SMBs rarely possess. The viable techniques are different, not worse.

How do you evaluate an AI consultant for an SMB engagement?

The right consultant for an SMB engagement behaves in a specific way before any contract is signed.

They start with process, not technology. They map current processes, identify bottlenecks, and quantify the cost of those bottlenecks. AI is only worth building if the problem is expensive enough to justify the investment.

They audit your data honestly. They examine your actual data — format, completeness, consistency, volume — and tell you whether it supports the intended application. If it does not, they propose a data collection plan with a timeline, not a workaround.

They propose appropriate technology. For many SMB problems the right answer is not a custom ML model. It is a well-configured off-the-shelf tool, a rules-based automation in a workflow engine, or a structured database query. Good consultants recommend the simplest solution that solves the problem.

They define success metrics upfront. Before building anything, the engagement defines what “working” means in measurable terms: response-time improvement, error-rate reduction, cost savings, or revenue impact.

For SMBs evaluating consultants, our guide to what AI consulting actually involves provides the broader framework for distinguishing consultants who deliver outcomes from those who deliver slide decks.

The right starting point

For most SMBs, the highest-value first AI project shares four characteristics:

• It uses data you already collect (no new data infrastructure required).
• It automates a repetitive, time-consuming task — not a complex judgement.
• It has a clear, measurable baseline (current cost or time is known).
• It is low-risk if it fails (not customer-facing, not safety-critical).

That narrows the field considerably, and that is the point. Starting narrow and succeeding builds the internal confidence, data discipline, and operational knowledge to tackle more ambitious projects later. Ask a prospective consulting partner for a risk-structured engagement plan that names the pivot points before model work begins. If they can’t produce one, that’s your answer.

FAQ

What should I look for when evaluating AI consulting firms, and what should I screen out? Look for firms that own the outcome rather than rent you engineers, audit your data before proposing a solution, and define success metrics upfront. Screen out firms that lead with technology choice, quote on hours without naming a measurable outcome, or refuse to tell you when a project is infeasible.

How do boutique AI consultants differ from Big Four consulting firms in scope, methodology, and accountability? Boutique firms typically scope to a specific technical outcome with a small senior team that stays on the engagement end-to-end. Big Four firms tend to scope broader programmes staffed via pyramids, with junior delivery and senior oversight; accountability is to the programme governance layer, not to a specific technical result.

Which evidence genuinely separates capable firms from rebranded ones? Concrete case studies with named technologies, measured baselines, and stated failure modes — not logos. References that can speak to how the firm behaved when something went wrong. Technical depth visible in how the firm discusses data quality, evaluation, and deployment, not just model selection.

How much does an AI consultant cost, and what determines the price band for a serious engagement? Serious custom AI engagements are priced by senior-engineer time and outcome risk, not by SaaS-style subscriptions. Price bands are driven by data readiness, integration surface, and whether the firm carries delivery risk or passes it back to the buyer.

Which contractual structures protect the buyer in AI work? Engagements scoped to your problem with explicit milestone gates and pivot points protect the buyer better than open time-and-materials. Outcome-based structures work when success metrics are measurable upfront; otherwise milestone-gated R&D engagements with outcome ownership are usually the right shape.

How do I evaluate a consulting firm’s ability to hand off to my internal team rather than create dependency? Ask how documentation, model artefacts, and operational runbooks are produced during the engagement, not at the end. A firm planning a clean handover writes for your future team from day one and proposes joint working sessions, not closed delivery.