When to hire an AI consultant instead of somebody who builds

The question a consultant is for

A consultant is the right purchase when you cannot yet name the problem. If you can name it, you need somebody to fix it, and paying for a diagnosis is a way of postponing the decision you have already made.

There are recognisable symptoms of genuinely not knowing. Two departments describe the same delay and blame each other, and both descriptions are plausible. Revenue is fine but everybody is exhausted, and nobody can say which hours are going where. You are about to commit real money to a system and the only case for it is that a competitor has one. Somebody has proposed automating a process that three people describe differently.

Each of those is a measurement problem wearing the costume of a technology problem. Buying a build before it is resolved means automating whichever version of the process the loudest person described, which is how a business ends up with an expensive machine for doing the wrong thing quickly.

• Two teams describe the same bottleneck differently and both accounts hold together
• Nobody can say how many hours a week go into a task everybody agrees is painful
• A significant purchase is being justified mainly by what a competitor appears to have
• The process about to be automated has no single agreed description

The conflict of interest, said out loud

A firm that sells builds and also sells assessments has a reason to find a build in every assessment. This is not usually dishonesty. It is what people see when their attention has been trained by the work they do, and it operates whether or not anyone intends it.

Which is the honest case for a consultant who does not build anything. Their recommendation carries information precisely because their income does not depend on which way it goes. If your situation is that you genuinely do not know, and the sums involved are large, a pure advisor is worth the money and worth choosing over a builder offering the same assessment cheaper.

A firm that does both, and Wobble does, can only manage the conflict in the open. That means the assessment is paid for separately and stands alone, the document is written so a competitor could execute from it, the recommendation includes the option of doing nothing, and buying the assessment never obliges you to buy the build. Judge any firm here by whether it has ever told a paying client not to build. If it has not, the question has never really been asked.

If the assessment cannot be handed to a different supplier and acted on, it was a sales document with a cover page.

What a paid assessment has to hand over

The gap between assessments is enormous, and the difference is not length. A short document that lets you act is worth more than a long one that lets you nod.

There is a simple test for the whole thing. Could you hand this document to a different supplier, one who has never met you, and would they be able to quote and build from it without starting again? If yes, you bought an asset. If no, you bought a presentation and the knowledge is still inside somebody else's head, which is exactly where it was when you started.

A second test, which fewer assessments pass: does it say what not to automate? An assessment that recommends automating everything examined has not exercised judgement, it has produced an inventory.

• A map of how work actually moves today, built from watching people work rather than from interviews alone
• The bottlenecks, with an indication of what each one costs in time or lost work
• A shortlist of what to automate, sequenced by payback rather than by technical interest
• An explicit list of what should not be automated, and why
• A build or buy call on each item, since some of them are an existing product you already pay for
• A cost range for each item, with the assumptions stated so you can challenge them
• The measurement baseline, so whoever builds it can be held to something

What a diagnosis usually finds in a Pakistani business

The findings repeat across businesses of very different sizes, which is itself useful, because it means most of this is a solved problem rather than a special case.

The real customer record is usually in WhatsApp. Not in the CRM somebody bought, in the chat history on a phone, often on a personal phone belonging to a salesperson who may leave. Customers send voice notes instead of typing, screenshots of a bank transfer instead of a reference number, and switch between English and Urdu inside a single message. Any system designed for clean typed English text meets reality on day one and loses.

The critical process almost always lives in one person's head. It works, it has worked for years, and it has never been written down. Documenting it is the highest value week in most assessments and the one clients are most reluctant to pay for, because it produces a document rather than software.

And there is generally no baseline at all. Nobody knows how many enquiries arrive in a week, how long a first reply takes, or what share never gets one. Businesses here describe losing orders because an enquiry sat unanswered for hours, and the reason it keeps happening is that nothing counts it, so it is nobody's problem in particular.

How to tell a consultant from a slide deck

The clearest signal is early and easy to check. Did they ask to sit and watch somebody do the job? A consultant who assesses a business without observing the work is producing an opinion about what managers believe happens, and managers are systematically wrong about the exceptions, which is where the cost is.

The second signal is whether the recommendations are sequenced and costed. A list of opportunities with no order and no numbers is a menu, and you already had a menu. Sequencing forces judgement, which is what you paid for.

The third is whether anything in the document is uncomfortable. Real diagnosis produces at least one finding somebody in the room did not want, usually about a process an individual owns. A report where every finding is flattering was written to be approved rather than to be true.

Also worth asking directly: what would make you tell us not to do this? Somebody who has genuinely delivered has a threshold and will state it. Somebody who has not will say it depends.

When an assessment is procrastination

If you already know what is broken, an assessment is procrastination with an invoice. A business that knows its enquiries wait four hours for a reply does not need a study of why. It needs somebody to fix the reply.

If the business is small enough that the owner personally touches every process, the assessment finding will be that the owner is the bottleneck, and that is available for free right now. Spend the money on the first automation instead and learn from operating it.

If nobody will be free to act on the findings for the next six months, wait. Assessments go stale. Tooling moves, prices move, staff leave, and a document written today is a weaker guide by the time somebody opens it in the new year. Commission it when there is capacity to act, not when there is budget to spend.

Who you would actually be speaking to, and what leaves with you

An assessment is a conversation with a person, and it is fair to ask which one. Haad, the co-founder who owns growth and client solutions, does most of the diagnosis, and Moiz Khan decides whether a thing is buildable and how. Between them the routes into this company run through medicine, chartered accountancy, media and businesses that had to make payroll, which is a different background from a consultancy and shows up in what gets recommended.

Every intake gets a reply within two working hours, and the readiness call is free while the diagnosis that follows is not. Wobble is based in Karachi and works across 25 engagements in six countries, so the reference points are installations published on the work page with their numbers rather than a framework.

What you keep is the whole reason to buy advice separately from delivery. The written diagnosis, the bottleneck list in priority order, the integration report saying what your tools will and will not permit, and the build roadmap are yours whether or not you go on to build with Wobble. You own the document. Take it to three suppliers and get quotes that can be compared.

An assessment does not replace a decision. It stops at the point somebody in your business has to commit budget and staff time, and that waits for a person who can actually authorise it. If nobody in the room can, the assessment is early. Doing the diagnosis in-house is realistic for a business with an operations manager willing to be honest about their own department, and the honesty is the hard half rather than the method.

Common questions

What does an AI consultant actually do?

Works out where a business genuinely loses time or money, decides which of those a machine should handle and which it should not, and sequences the work by payback. The output is a decision document rather than software. The value is in judgement about what to skip, which is the part a build team is least incentivised to produce.

Do we need an assessment, or can we go straight to building?

Go straight to building when you can already name the problem, the process is written down or observable, and the first workflow is small enough to learn from. Get an assessment when several people describe the same process differently, when nobody can quantify the cost of the pain, or when a large commitment is being justified mainly by what competitors appear to be doing.

Should the consultant also be the one who builds it?

It is convenient and it carries a conflict, because a firm that builds has a reason to find a build. If the sums are large and you genuinely do not know what is wrong, a consultant who does not build gives you a cleaner recommendation. If you use a firm that does both, insist the assessment is paid separately, stands alone, and is written so another supplier could execute it.

What should an AI assessment document contain?

A map of how work moves today, the bottlenecks with an indication of what each costs, a shortlist sequenced by payback, an explicit list of what should not be automated, a build or buy call per item, a cost range with the assumptions shown, and the measurement baseline. The test is whether a supplier who has never met you could quote from it.

How long does an AI assessment take?

It depends mostly on how much of your process is already written down and how quickly people can be observed doing the work. The part that sets the pace is access to the people who do the job, not analysis time. If you want it faster, prepare a list of your tools, admin logins, and two weeks of simple counting before it starts.

Is a free consultation the same as a paid assessment?

No, and treating them as the same is how buyers get disappointed. A free call establishes whether there is a plausible fit and where to look. It is a qualification conversation and it should be honest about being one. A paid assessment involves observing the work, mapping it, and producing a document somebody could build from.