The short version
A day rate punishes you for getting faster. Price discovery as its own product with a written deliverable, price the build against the value of the problem rather than the hours, and charge more for the second engagement in a domain, not less — because by then you are selling judgement rather than time.
Why the day rate caps you
Selling days means your income is hours multiplied by rate, so the only lever is the rate, and the rate is anchored to what a buyer imagines a skilled contractor costs. Worse, it inverts the incentive: the better you get, the fewer days a job takes, and the less you earn for the same outcome.
It also invites the wrong comparison. A buyer comparing three day rates is shopping on the one dimension where you cannot differentiate, and the cheapest number wins by default.
Price discovery as a product
The single most useful change most independents can make. Instead of a free scoping call followed by a speculative proposal, sell a short, fixed-price piece of work with a written deliverable: a sample of their real data, a measured baseline, the exceptions categorised, an integration feasibility check, and a recommendation with a scope and a price.
Three things happen. You are paid for the thinking that you were previously giving away. The buyer gets something valuable even if they stop there. And you never again quote a build blind, which is where independents lose money.
It is also the easiest thing in the world to say yes to. A small fixed fee for a document that includes “don’t build this” as a possible conclusion is a low-risk purchase, and buyers know it.
How do I price the build?
Against the value of the problem, not the hours you expect to spend. Ask, during discovery, what the current process costs in a year — people-hours, errors, delay. That number is the ceiling and it is usually much larger than any hourly calculation would have produced.
Then price a fixed scope against it. Fixed price transfers delivery risk to you, which is why it must follow discovery rather than precede it: once you have measured the exception rate and proven the integration, the risk is one you can actually size. Quote fixed before that and you are gambling.
What should I charge for the second one?
More. This is the part that feels wrong and is the most important.
Your second invoice-reconciliation project takes less time than the first, because you have the patterns, the integration approach and the evaluation harness. The instinct is to pass the saving on. Don’t — the client is not buying your hours, they are buying an outcome you can now deliver with far less risk, faster, and with a case study proving it. That is worth more to them, not less.
Specialising is how independents escape the rate ceiling. Three projects in one domain and you are no longer a generalist selling days; you are the person who has done this before, which is a different product at a different price.
What about ongoing work?
Price it, separately and explicitly, and put it in the proposal. Someone maintains the system when a model version changes, monitors the evaluation set, and handles the exception rate drifting. If you do not price it, you will end up doing it free, or the deployment will quietly degrade and become a failure with your name on it.
A modest monthly retainer covering monitoring and a fixed number of hours is easy to sell alongside a build and is the difference between project income and a practice.
The honest summary
Rate benchmarks tell you what the median person selling days charges. If that is the business you want, look them up. If you would rather not be priced by the hour: sell discovery as a product, price builds against the value of the problem, specialise so your second project costs you less and earns you more, and always price the maintenance.
What if the client asks for my day rate anyway?
Give one, and follow it immediately with the alternative. Refusing to answer reads as evasive and costs you the conversation. What you can do is state the rate, then say you would rather price the outcome, and explain why in one sentence: they get a fixed cost and you carry the delivery risk, which is a better deal for both of you than paying by the hour for something neither of you has scoped.
Most buyers accept that instantly, because it is obviously true. The ones who insist on hourly billing for undefined work are usually the ones who will also dispute the invoice.
How do I raise my prices without losing clients?
On new work, not existing engagements, and by changing what you sell rather than what you charge for the same thing. A practitioner who has done three deployments in one domain is selling something different from one who has done none: less risk, faster delivery, a case study, and a set of judgements the client would otherwise be paying to discover.
Price the new thing at the new number and let existing clients continue at the old one until their next project. Nobody objects to that, and it removes the conversation people dread.
Should I ever work for free or at a discount?
For a first case study in a domain you intend to specialise in, a deliberate discount can be a rational investment — provided you get something specific in return and agree it up front: a named reference, permission to publish the work, and an introduction. Write those into the engagement rather than hoping they materialise from goodwill.
What does not work is discounting to win price-sensitive buyers with no strategic value. They do not become full-price clients later, they take longer to manage, and they set the anchor for what your work is worth.
What should I put in the proposal?
Discovery and build priced separately. An accuracy threshold with how it will be measured. The running cost as well as the build cost. Who owns the prompts and code afterwards. And at least one named risk.
That last item wins more work than it loses. A proposal with no risks reads as a sales document; one that names what could go wrong and how it will be handled reads as someone who has delivered before, which is precisely what the buyer is trying to determine.