How to Build an AI-Enabled Commercial Operation

Most businesses that ask how to become an 'AI-enabled' commercial operation are really asking the wrong first question. The useful question is not which AI tools to buy, but which commercial problems are genuinely worth solving, and whether AI is the right way to solve them at all. A commercial operation built around technology first and problems second usually ends up with a collection of disconnected tools, none of which change how revenue is actually won.

Evans works with commercial leaders on this precisely because it sits between two things Evans does not do: it is not software engineering, and it is not generic AI advice detached from how a sales team, a pipeline or a customer journey actually functions. It is commercial consultancy that treats AI as one more capability available to a sales operation, alongside process, people and structure — never as a substitute for any of them.

This article sets out a practical sequence for building an AI-enabled commercial operation: how to find the real problem, remove bad process before adding technology to it, decide where AI genuinely helps, and connect that into the day-to-day commercial workflow without losing human control over the decisions that matter.

Why do most AI initiatives in commercial operations fail to stick?

The pattern is familiar: a business buys an AI tool because a competitor has one, or because a vendor makes a compelling demonstration, and rolls it out without first agreeing what commercial problem it is meant to solve. Six months later the tool is used inconsistently, the data behind it is unreliable, and nobody can say whether it has actually changed a sales outcome. The technology was rarely the problem. The absence of a clear commercial starting point was.

Evans' approach to this is deliberately sequenced, because skipping steps is exactly where these initiatives go wrong. It starts with the commercial problem, strips out bad process before anything else, only then asks where AI genuinely helps, builds the simplest solution that works, connects it to the real workflow, tests it against actual commercial activity, and keeps a person in control of every decision that carries real risk.

Step one: identify the actual commercial problem

Before any conversation about AI, the question is what is actually limiting revenue — a weak pipeline, poor conversion at a specific stage, slow proposal turnaround, unreliable forecasting, or sales time lost to administration. This sounds obvious, but it is the step most often skipped, because it is easier to talk about tools than to sit with an uncomfortable answer about where the commercial operation is genuinely underperforming.

Step two: remove bad process before adding technology to it

Automating a broken process makes it broken faster, not better. If qualification criteria are inconsistent, if nobody owns follow-up, or if the CRM has been abandoned because it never reflected reality, those problems need fixing on their own terms first. Only once the underlying process is sound does it make sense to ask whether technology, including AI, can make it faster or more consistent.

Step three: identify where AI actually helps

With a sound process in place, the next step is narrow and specific: which tasks within that process are repetitive, time-consuming, and based on information that already exists, rather than judgement that still needs to be made. Prospect research, first-draft proposals, meeting summarisation, pipeline reporting and administrative CRM tasks are typical candidates. Pricing decisions, relationship management and final judgement on deal probability generally are not.

Step four: build the simplest useful solution

The instinct in AI projects is to build something comprehensive. The more reliable approach is to build the smallest version that solves the identified problem, get it working properly with the people who will actually use it, and expand only once it has proven itself. This is the same principle behind Spitfire Pulse, an AI-enabled platform Tom Evans designed while Sales Director at Spitfire Doors, which brought together company knowledge, order verification, installation tracking and marketing assets into one place built around how the sales and operations team actually worked, rather than a theoretical ideal of what a system should do.

Step five: connect it to the commercial workflow

A tool used by one person in isolation rarely compounds into anything meaningful. Value comes from connecting the AI-assisted task into the wider commercial workflow — research feeding into CRM records, CRM records feeding into pipeline reporting, pipeline reporting feeding into how a sales director actually runs forecast conversations. Disconnected point solutions are a common reason businesses end up with several AI tools and no measurable change in commercial performance.

Step six: test it against real commercial activity

A tool should be judged on whether it changes something real — time saved, proposal turnaround, data reliability, conversion at a defined stage — not on how sophisticated it appears in a demonstration. This means running it against live pipeline and real customers for a defined period, with someone accountable for reviewing whether it delivered what it was meant to.

Step seven: keep human control over judgement

AI cannot reliably predict revenue, cannot own a customer relationship, and should not be trusted to make final judgement calls on pricing, deal probability or account priority without human oversight. The businesses that keep this distinction clear — automation for information and administration, people for judgement and relationships — are the ones that sustain AI adoption without eroding the trust their commercial operation depends on.

A diagnostic framework for where to focus first

What implementation steps follow from this?

1. 01Name the specific commercial problem before discussing any tool
2. 02Audit and fix the underlying process, even if this feels like it delays the AI work
3. 03Shortlist a small number of tasks where AI genuinely fits: repetitive, information-based, low relationship risk
4. 04Build or select the simplest tool that addresses one of those tasks properly
5. 05Connect it into the existing commercial workflow rather than running it as a standalone experiment
6. 06Test it against a defined period of real activity and measure against a specific commercial indicator
7. 07Keep a named person accountable for every judgement-based decision the tool touches

What are the common mistakes when building this?

The most common mistake is reversing the sequence — buying the tool first and looking for a problem to justify it afterwards. A close second is trying to automate too much at once, which makes it impossible to isolate what is actually working. A third is failing to keep a person accountable for judgement calls, which erodes trust in the system the first time it gets something important wrong.

What does useful automation look like here, versus bad automation?

Useful automation removes repetitive, information-based work and hands cleaner information to the person making the decision. Bad automation removes the decision itself, replacing a person's judgement with an automated output that looks authoritative but has not been checked against the reality of the relationship or the deal. An AI-enabled commercial operation keeps that distinction deliberately visible, rather than letting convenience blur it over time.

How do you know it is working?

The clearest indicators are specific and commercial: less administrative time per salesperson, faster proposal turnaround, more reliable CRM data, and pipeline reports that a sales director trusts enough to act on directly. If the only evidence of success is that a tool has been adopted, rather than a measurable change in one of these indicators, the sequence has not been followed properly and is worth revisiting from the first step.