What an AI opportunity assessment actually includes
What two to three weeks of assessment buys, what the document contains, and how to tell a real one from a sales exercise with a report attached.
A real AI opportunity assessment produces four things. A ranked short list of what is worth building, a longer list of what is not and why, a baseline figure for each survivor so the result can be checked later, and an estimate of what each one costs to build and to run. It takes two to three weeks and runs roughly $15,000 to $25,000. An assessment that cannot conclude with a recommendation to build nothing is a sales exercise with a report attached.
The word covers two different products and they are priced similarly, which is the problem.
One is a fortnight of somebody sitting with the people who do the work, timing it, and coming back with a short list and the arithmetic behind it. The other is a workshop, a maturity model and a roadmap, produced largely from a template, which concludes that you should begin a programme with the firm that wrote it.
Both are called an AI opportunity assessment. Here is what the first one contains.
The four outputs
1. The short list, ranked
Three to six things, ordered by what they are worth rather than by how interesting they are. Each one names the job, the people doing it, the hours it consumes, and what it would look like changed.
Ranked matters. A list of eleven opportunities with no order is a decision handed back to you, which is the thing you paid to avoid.
2. The longer list, with reasons
The things not worth building, and why each one failed.
This is the output clients remember and the one most firms do not produce, because every entry on it is revenue declined. It is also the most useful page in the document, because those ideas will come back in six months and you will want the reasoning rather than a vague memory that somebody said no.
Volumes too low. Process changes too often to encode. The real bottleneck is elsewhere. A human has to be accountable for this decision and always will be. Those are the usual reasons, and they are specific to your company rather than general principles.
3. A baseline for each survivor
A number, taken before anything is built, that the work can be checked against afterwards.
Without this nothing that follows can be assessed. With it, the build has a target and the closing document has something to compare against. A baseline taken after work starts is not a baseline, which is why this belongs in the assessment rather than in the project.
Most companies have never measured the processes in question. That is normal, and measuring something nobody ever instrumented is a known piece of work rather than an obstacle.
4. What each one costs, to build and to run
Both numbers. The build estimate is the one everybody asks for and the running cost is the one that decides whether a thing survives its first year.
An AI system has a bill every month. A short list with build costs and no running costs is half an answer.
What the two or three weeks actually consist of
Week one is watching. Sitting with the people doing the work, in the tools they use, while they do it. Not a workshop. The written process and the real process differ in almost every company, and the gap is where the opportunities and the bad estimates both live.
Week two is arithmetic. Timing the jobs, sizing the volumes, costing the builds, and eliminating. Most ideas die here, which is the point.
Week three, where needed, is access. Finding out whether the systems involved can actually be reached, because an opportunity behind a system nobody can get into is not an opportunity yet.
How to tell which one you are buying
Ask what the output is. "A roadmap" is a warning. "A ranked list with a baseline for each item" is the other thing.
Ask whether it can conclude with nothing. If the assessment structurally cannot recommend that you build nothing, the conclusion was written before the work started. Ask the firm when they last delivered that outcome.
Ask who does the watching. If the fortnight is run by a consultant who then hands to a delivery team, the knowledge that made the list worth having does not travel.
Ask whether the fee depends on what follows. An assessment priced as a fixed fee is being sold on its own merits. One discounted against a build that has not been scoped yet is a deposit.
One limit worth stating
Two to three weeks is enough to find the opportunities in a company's existing processes and is not enough to do two things people sometimes expect. It will not tell you whether your product strategy is right, because that is a different question with different evidence. And it will not audit your data estate properly; it will tell you whether the data needed for the short list exists and roughly what state it is in, which is a narrower claim than it sounds and is deliberately narrower.
What it costs
Ours is $15,000 to $25,000 as a fixed fee, published on the pricing page with the rest. What moves it is how many teams have to be sat with and how fast access arrives, not the size of the build it recommends.
If that is more than you want to spend finding out whether any of this works, the smaller version is seven days on one workflow for $2,500, which produces a working system and a before and after figure rather than a list.
Either way the first call is free and sometimes ends with us saying there is nothing here worth building. Thirty minutes.
If you have a budget, a deadline, and no clear answer on which AI project deserves either, that is the conversation to have.
Book the callWritten by
Radwan Altaf
Radwan runs AISynq. Before that he delivered software inside enterprise programmes at DHL, AT&T, DirecTV and Accenture, which is where the habit of measuring a result against its baseline came from. More about the firm.
Read next
- How long an AI project really takesRealistic durations for an assessment, a build and a first production release, and the four things that add months without appearing on any plan.
- How to automate a startup, piece by pieceThe order to automate a small company in, why that order is the opposite of what most founders start with, and the test each piece has to pass first.
- Which processes to automate with AI firstThe candidates are the same at every software company. Four tests that predict which survive production, and the order to attempt them in.
Work with us
Want this done properly against your own systems rather than in general?
How the assessment runsGet the next one
New writing in AI opportunity as it goes up, roughly twice a month. One article per email and nothing else in it.