Both are sold as "AI consulting". They answer different questions, and the cost of choosing wrongly falls in different places.
The distinction in one line
A consultant tells you what to do. A build partner does it. If you cannot yet write down which process is costing you and roughly what fixing it looks like, you are not ready to buy a build.
Which you need
| You need advisory | You need delivery | |
|---|---|---|
| The problem is | vague, or contested internally | named, and agreed |
| You can state the measure to move | not yet | yes |
| The risk you are carrying | spending on the wrong thing | spending on the right thing badly |
| What you buy | a ranked plan and a "not this" list | a working system |
| Ends with | a decision | something running in production |
Most businesses can place themselves in one column within a minute of looking at that table honestly. The trouble is that the columns feel embarrassing in different ways — the left one feels like not knowing your own business, and the right one feels like committing before you are sure.
Signs you are buying discovery you do not need
- You already know which process is the problem, and so does everyone else in the room
- You can name the number you want moved
- The proposal's discovery phase is priced at a meaningful share of the build
- You have had a similar review done in the last year
If three of those are true, ask for a proposal without the discovery phase and see what happens to the price and the confidence.
Signs you are about to build too early
- Two people in the business would describe the problem differently
- Nobody can say what "better" would look like as a number
- The process is being restructured anyway
- The enthusiasm is for a technology rather than for an outcome
The last one is the most common and the hardest to say out loud. Wanting to use AI agents is not the same as having a problem an AI agent solves.
The honest case for doing neither yet
Sometimes the answer is that nothing is stable enough to encode. A business mid-restructure, or one where the process genuinely differs by person because nobody has decided which way is right, is better served settling that first.
Software applied to an undecided process does not resolve the ambiguity. It picks one interpretation and makes it permanent, and you will discover which one months later.
What to ask before you sign either
Ask a consultant what would make them recommend doing nothing. Ask a build partner what they would refuse to build.
Both questions are looking for the same thing: whether the answer was decided before the conversation began. A firm that can describe, specifically, the engagement it turned down is telling you something a portfolio cannot.
Questions
- Can the same firm do both?
- Often, and it saves rebuilding context with a second team. The question to put to them directly is whether the advisory stage can honestly conclude "do not build this" — and what has happened when it did.
- How much of the budget should advisory take?
- There is no correct ratio, but a discovery phase priced near the build deserves an explanation. If the review is that substantial, ask what it produces that changes the build, and whether you could buy the review alone and take it elsewhere.
- What if we get advisory and the answer is uncomfortable?
- That is the outcome you were paying for. An audit that finds the real constraint is a management problem rather than a technology one has saved you a build that would not have worked, which is worth considerably more than it cost.
Want this working on your numbers rather than in an article? We scope it on one call.
Book a Consultation Call