You are sitting with a proposal. The consulting fee says £15,000. The board has approved that number. Six weeks in, the consultant flags that the data needs cleansing, your team is spending half their week in workshops, and you have just bought three new SaaS subscriptions to support the implementation. The fee was right. Your budget was wrong.
I see this pattern in nearly every AI engagement at SME scale. The cause is structural. It comes from how the work gets priced. The visible fee is the smallest cost in the project. The rest of the cost lives in places the proposal does not name.
Why is the consulting fee always the smallest line?
The consulting fee covers the consultant’s time. It does not cover anything else. Your team’s hours, your tooling subscriptions, your data preparation work, your change management, and your regulatory compliance all sit outside the fee, even though they are required for the engagement to deliver. By the time you have added them, the fee is roughly a third of what the work actually costs in year one.
The pattern is older than AI. It showed up in CRM rollouts in the 2010s and ERP implementations a decade before that. AI is just the latest category to charge SMEs the same lesson.
The five hidden categories most proposals do not cover
Five cost categories sit outside the fee, every time, in every AI engagement at SME scale. Tooling subscriptions, internal staff time, data preparation, change management and training, and regulatory compliance. None of them are optional. None of them appear on the consultant’s invoice. By year one each one has its own monthly run rate, and together they will usually exceed the consulting fee.
Tooling subscriptions are the first. Microsoft Copilot, ChatGPT Enterprise, a vector database, an integration platform like n8n or Zapier, and monitoring infrastructure can aggregate to £1,500 to £5,000 a month at SME scale. Most of this becomes apparent in the second or third month, after the consultant has scoped the architecture.
Internal staff time is the second. A typical strategy engagement absorbs 100 to 150 hours of your team’s time across stakeholder interviews, data access coordination, internal validation, IT integration, and change management workshops. At loaded rates of £50 to £95 an hour, that is £5,000 to £15,000 of capacity that never appears on an invoice.
Data preparation is the third. A discovery phase usually surfaces that your data needs cleansing, restructuring, or integration before the AI work can run on it. ETL work, master data management, and quality remediation typically run £8,000 to £30,000, and it is rarely flagged at proposal stage because the consultant cannot see the state of your data until the engagement starts.
Change management and training is the fourth. Workshops, runbooks, internal champions, and post-launch support absorb £5,000 to £15,000 in either fee uplift or internal time. Skip this and the AI capability technically lands but never gets used.
Regulatory compliance is the fifth, and at SME scale in regulated sectors it is rarely included at all. EU AI Act QMS work, ICO DPIA documentation, and sector-specific guidance from the SRA, FCA, or CQC each carry cost. The EU Commission’s own impact assessment puts regulatory overhead at 17% of total AI spending.
The 3.3x multiplier, decomposed
A 2025 analysis of 127 enterprise AI implementations found that true implementation costs ran 3.3 times the initial budget, with around 70% of total cost living in hidden categories. That is the AI-specific number. The CRM and ERP rule of thumb has been 2 to 3 times licensing cost for two decades. AI is following the same pattern with slightly more variance.
The maths sits roughly like this for an SME engaging on a £12,000 strategy plus pilot package. Consulting £12,000. Data preparation £8,000. Tooling £6,000 in year one. Internal staff time £3,000 to £4,000 of opportunity cost. Change management £6,000. Contingency at 20%, £7,000. Total: £42,000 to £43,000 in year one.
The fee was sized correctly. The total surprises the buyer almost every time, because the buyer was reading the proposal as a price tag instead of a starting line.
How to budget for the real number from day one
You can plan for the 2-to-4x rule explicitly, and it changes the conversation with the consultant rather than the price tag. Build the budget around line items, not a single fee. The fee is one of six lines, the others sit alongside it, and a contingency wraps the whole thing. That structure forces every cost to surface at proposal stage, not at week six when the data audit lands.
The consulting fee is one line. Tooling year one and forecasted year two run rate is another. Internal staff time, costed at your actual loaded rate, is a third. Data preparation work, even if estimated wide, is a fourth. Change management and training is a fifth. Regulatory compliance, where it applies, is a sixth. Add a 20% to 30% contingency on top of all of it. That is the budget.
When you take that build to the consultant, two things happen. First, you ask whether their fee assumes the data is clean, or includes data preparation. The answer surfaces a real estimate before sign-off rather than a surprise at week six. Second, you can ask which line items they will help you populate, and which they will not touch. The proposal becomes a conversation about scope rather than fee.
What to ask before signing
Three questions tell you whether the proposal accounts for the real cost of the engagement. Each one targets a specific category that consultants either cover or quietly leave for the buyer to discover later. None of them are gotchas. A good consultant will answer them in writing without flinching, and a less good one will give you the answer that tells you to walk.
Ask what the consultant assumes about your data quality. If the answer is vague or pushed to a later phase, factor in £8,000 to £30,000 of data work and ask the consultant to confirm in writing whether their fee covers any of it.
Ask which tooling subscriptions the architecture they recommend will require, and what they cost monthly. A consultant who cannot estimate the tooling stack within a £500 a month range either has not designed the architecture or is hoping you will not look. Either is a problem.
Ask how much of your team’s time the engagement will need, and from whom. A good answer names the roles, the hours, and the weeks. A less good answer is “minimal” or “as needed”. The buyer who hears the less good answer signs anyway, then watches three senior people lose half their week for two months.
The 2-to-4x rule is a budgeting principle, applied early. The fee is one line. The total is the question worth answering before you sign.
If you would like to talk about how to scope an AI engagement at SME scale without falling into the budget gap, book a conversation.
