Forward-deployed engineer vs traditional consultancy vs AI platform

There are three paths to implement AI in any serious retail operation.

The platform. The consultancy. The implementer.

The question we hear most often, “which is the best?”, is the wrong one. There is no best. There is one that fits your operation and two that do not.

Let me walk through each.

The platform

You buy software. Pricefx for pricing, RELEX for demand forecasting, Revionics for promotion, NCR or Toshiba for self-checkout. The vendor configures, integrates, and owns the uptime.

What it costs: annual subscription, usually per store, per SKU, or per user.

For a mid-size chain of 40 to 80 stores, a demand forecasting platform runs between $200K and $500K per year. Initial implementation sits outside that number.

Who operates it after it goes live: your internal team. The vendor is not there when the line forms on Black Friday morning.

The real advantage is that the learning curve was already paid by 200 other chains. The bugs that would kill a new system have been hunted. Support has business hours.

The trap is less visible.

A platform wants to run the platform’s way. Your produce assortment, your price ladder, your promo cadence, your category structure, all of it will bend to fit the product’s model.

Small adaptations. Acceptable early on.

Less acceptable three years in, when you want to do something the vendor’s roadmap will not carry.

The price climbs when you have nowhere to go. Anyone who has worked the operation knows that renewal is a different conversation than acquisition.

The traditional consultancy

You hire a team of people. BCG, Accenture, IBM, McKinsey, Deloitte. They diagnose, recommend, and in some formats implement with an extended squad.

What it costs: billable hours, plus a heavy overhead of partner travel and weekly senior visits. For a six-month engagement in a 40-store chain, budget somewhere between $750K and $2M. The range is wide because it depends on how much of the implementation sits in their scope versus your team’s.

Who operates it after it goes live: it depends on how the contract was written. If the deliverable is deck and process, your team operates. If it is system, you usually end up with a system integrated with their components that still require their presence.

The real advantage is the panoramic view. They saw this problem at twenty other chains.

The diagnosis tends to be right. The timeline tends to be optimistic.

The trap has two faces.

First: the exit point is vague. You know when the engagement starts.

You do not always know when it ends. Each milestone becomes the justification for the next.

Second: the senior person who sold the proposal is not the senior person who runs the delivery. The real squad is more junior than the slide claimed. You find this out by week three.

The implementer

You hire a small senior team, embedded in your environment, for a single problem, with a fixed timeline and a success criterion agreed before the work starts.

The engineers sit next to your team. They write code on your data. They integrate with your legacy systems.

They leave operations your internal team runs after that.

What it costs: a fixed investment per Implementação, communicated in conversation after a two-week Diagnóstico. For the same 40-store chain and the same demand-forecasting problem, the number lands at a fraction of the other two. The scope is tight and the overhead is zero.

Who operates it after it goes live: your internal team. With a runbook. Without the implementer.

The real advantage is the definition of done. The success criterion is closed before the first line of code.

When it is hit, the engagement ends. When it is not, everyone knows, and the conversation about what to do is honest.

The trap exists.

Implementer breaks down when the client does not have a senior decision-maker on the other side.

If the contact point is an analyst who has to ask for approval on every decision, the engagement stretches. The timeline slips, and the success criterion becomes aspiration.

It works in some shapes. Not in all.

The implementer also does not replace a platform as the system of record. If you need a product running 24x7 across 200 stores forever, at some point you buy a platform. The implementer is what builds what is missing, or connects what exists, or validates whether the platform is worth renewing.

When each path wins

• Platform wins when the problem is generic, the vendor’s roadmap covers what you need, and your internal team has the maturity to operate it.
• Consultancy wins when you need organizational diagnosis, alignment across departments, or the problem is so broad that the real deliverable is influence.
• Implementer wins when there is a critical operational problem, a senior decision-maker on the client side, a firm timeline, and a measurable success criterion.

There are three paths. The first two have catalogs. The third has criteria.

Honestly: of the three, the implementer is the least catalogable and the hardest to evaluate in an RFP.

There is no demo. No feature list. No Gartner quadrant.

What it has is a specific proposal, an honest timeline, and a criterion the two sides decide together before anyone signs.

Think for a second about the next AI decision sitting on your desk. Do you know which of the three paths is yours?

If yes, that is the path. If not, there is work that has to happen before the work.

The question that unlocks this decision is rarely technical. It is operational: who operates the system on Black Friday eighteen months from now?

If the answer is “the vendor”, you are buying a platform. If it is “the consultant”, you never finished buying. If it is “your internal team, with a runbook and criterion written before the code”, you are hiring an implementer.

Send me the shape of the problem. In one hour I will tell you which of the three paths fits your operation, with no bias from my side.

If it is platform, I will point you to the three names that show up in RFPs for this space. If it is consultancy, I will tell you what squad profile to ask for. If it is implementer, the next conversation is the two-week Diagnóstico: ends with criterion, timeline, and cost agreed in the same room, signed before the code.