Palm Technology Group
Internal briefing. Contains a named client's findings. Enter the password to continue.
Palm Technology Group · Internal briefing
And we can do it without a discovery call, a scoping document, or a single billable hour of consulting. This is what the assessment venture is, why it works, and what we would need to find out next.
Assessments built
1 complete
Mako Technology Solutions, end to end, live and linkable.
Findings on that business
12
Three critical, every one traced to a public source.
Human time to produce it
~0 billable hrs
Collection and synthesis are automated. People review and send.
Customers so far
0
Nothing has been sold or priced. That is the open question.
01 - What it is
We point a pipeline at a small business, and it comes back with a report on everything wrong with that company's public digital presence, measured, benchmarked against its actual local competitors, and ranked by what it costs them.
The output is not advice. It is a list of specific, verifiable defects with the evidence attached, plus one strategic recommendation about where the money is. Every claim traces to a public source we can show them.
The commercial model is a free teaser email leading to a paid full report. The teaser gives away two or three real defects, because that is the only way to prove the assessment is genuine rather than a template. The paid report holds the rest, including the revenue-side finding.
The report already exists when the teaser is sent, so buying it is self-serve: the owner pays online and gets access immediately. There is no call, no delivery, and no follow-up work. That is not a detail of the checkout flow, it is the reason the economics work at all.
This is deliberately not a consulting business. No retainer, no discovery, no scoping. If we cannot produce the second assessment for nearly the same effort as the hundredth, the idea does not work and we should not do it.
02 - Proof
Mako Technology Solutions is a dental and medical IT provider in Bradenton. The full assessment is live at mako-assessment.vercel.app. It found twelve things. These three make the case on their own.
Critical
Every person who found Mako through Maps or local search landed on an old Wix site that never mentions dentistry. The current site, built entirely around that specialism, was never seen.
One minute to fix · invisible to the ownerCritical
It displayed their main number and was wired to a different one, used nowhere else in the business. Anyone tapping it on a phone did not reach them.
Two minutes to fix · actively losing callsStrategic
Federal enforcement against healthcare providers has converged on one missing control. Mako's security page describes it accurately, then offers no name, no price, and no way to buy.
The revenue finding · this is what people pay forThe first two are the kind of thing an owner cannot see, because they look at their own website through memory rather than through a stranger's eyes. The third is the kind of thing an agency does not find, because it requires reading the regulatory environment their customers operate in, not just the website.
03 - Why it works
Three things have to be true at once for this to be a business, and on the evidence of the first run all three are.
The problems are real and specific. Not "your site could be faster" but "this button dials a number you do not own". A finding like that is checkable in ten seconds, which is what makes the email get a reply instead of a delete.
The owner cannot find them alone. These defects survive precisely because nobody looks. The owner knows what the page is supposed to say, the person who built it moved on, and nobody in a ten-person company owns "audit our own digital presence" as a job.
The comparison is what makes it true. On its own, Mako's mobile load time looked damning. Measured against six local competitors on the identical test, it came out second best of seven. Without the benchmark we would have sent a confident, wrong conclusion, and they would have binned the document. Competitor benchmarking is not a nice-to-have here; it is what makes a number mean anything.
04 - Why a product
The same work sold as consulting is a decent living and a cap on growth. Sold as a product it is something else. The difference is entirely in what a new customer costs us.
A digital audit, sold as consulting
What we are building
The fixed rubric matters more than it sounds. Because every business is scored the same way, the numbers compare across subjects and over time. That makes a re-assessment six months later a natural second sale, and it makes a portfolio of assessments into something with aggregate value rather than a pile of one-off documents.
Every other version of this idea rents its intelligence. Point a metered API at the problem and each assessment carries a per-token bill, which has to be paid on the ninety-five percent that never sell as well as the five percent that do. Costs rise exactly in step with volume, and the free teaser that makes the whole model work becomes the most expensive thing in it.
We run the models on hardware we already own. The marginal cost of one more assessment is a handful of Google Places lookups and some electricity. Not lower than a competitor paying per token, but a different shape entirely: theirs is a variable cost that scales forever, ours is a fixed asset that has already been bought.
This is possible here specifically because of how the pipeline is built. As Section 05 sets out, the model is never the source of a fact; it summarises and prioritises evidence that deterministic tools already gathered and that a deterministic stage re-verifies afterwards. That is a job a capable local model does well, and it is why a frontier API is a convenience rather than a dependency.
The honest consequence is that the ceiling stops being money and becomes machine time. Seven performance runs plus local generation take real minutes per assessment, so at high volume the constraint is throughput on one machine, not a bill. That is a better problem, and a solvable one, but it is the number to watch as this scales rather than cost.
05 - How it works
This ordering is the whole reason the output survives contact with the client. One false statement about a real named business ends the conversation, so the model is never allowed to be the source of a fact.
1 · Collect
Deterministic tools only. No model involved.
2 · Synthesise
A local model, working only from what stage 1 collected.
3 · Verify
Deterministic again. Gates the deliverable.
Stage 3 is the part competitors will not copy. It is unglamorous and it is where credibility actually comes from. On the last run it caught four defects in our own document, including two wrong numbers printed on the page, and re-verification corrected a claim we had already shipped.
06 - Economics
We have not sold anything, so we do not have a price, a conversion rate, or a cost per run. Rather than invent them, here is the model. Move the inputs and see what the business looks like.
Illustrative modelNone of these inputs are measured. They are the five numbers that decide whether this is a business, and the point of this section is to show which of them actually matters. Set them to whatever you think is realistic.
Each one produces a teaser email. This is the top of the funnel.
Entirely unknown. This is the number we most need to test.
Untested. Anchored against what a local agency charges for an audit.
A person checks the generated report before its teaser goes out. Paid on every assessment, including the ones that never sell. Nothing happens after a sale, so this is the only labour in the business.
Google Places lookups for the subject and its competitors. The models run locally, so there is no per-token cost to scale.
Contribution per hour of our time
$223
This is the number that says product rather than services.
Paid reports per month
5.0
Monthly revenue
$3,750
Monthly API cost
$25
Across every assessment run, sold or not. No per-token cost: the models are ours.
Monthly contribution
$3,725
$44,700 annualised
Our time per month
17 hrs
Three things fall out of moving these dials, and they are the reason this section exists.
Human review is the entire cost base. Move the money dials as far as they go and the shape barely changes; move the review dial and everything moves with it. Nothing happens after a sale, so a person's attention before sending is the only recurring input the business has. How many minutes an assessment needs before it can go to a stranger is the single operational question worth arguing about.
Conversion is worth far more than volume. Doubling the number of assessments doubles the review time along with the revenue. Doubling conversion costs nothing at all. Everything we do next should be aimed at that one number, which is why the test in Section 07 measures it and nothing else.
Cost per assessment does not scale, because we own the models. Running this on a metered API would put a price on every report, sold or not, and the ninety-five percent that never convert would be paid for in cash rather than electricity. On local hardware the marginal cost of one more assessment is a handful of Google Places lookups. That is a structural difference, not a saving, and it is covered in more detail in Section 04.
07 - What next
The proposal
Everything else is guesswork until we know whether a stranger who receives a genuinely accurate list of their own defects writes back. That single number decides whether this is a business or an interesting demo.
A vertical rather than a spread, because the regulatory and competitive research is the expensive part and it is reusable across businesses in the same field. The Mako run already produced a healthcare and dental IT knowledge base we would get for free on the next nineteen.
The cost of finding out is API calls, compute, and the time to review twenty reports before they go.
08 - Risks
The assessment product is built on separating measured facts from modelled ones. The same discipline applied to this business case gives the following.
| Claim | Status | Basis |
|---|---|---|
| The pipeline produces a genuinely accurate, source-backed assessment | Demonstrated | One complete run on a real business, every claim traced and re-verified |
| The defects it finds are commercially meaningful to the owner | Demonstrated | A Google listing sending every local-search visitor to an abandoned site |
| Marginal human effort per assessment is small | Assumed | True for run 1 after the tooling existed, but n = 1 and the tooling was built for this subject |
| Competitor benchmarking generalises to other industries | Assumed | Worked for local IT providers. Untested where competitors are not local or not comparable |
| A stranger will reply to the teaser | Unknown | Nothing has been sent. This is the test in Section 07 |
| Anyone will pay, and at what price | Unknown | No price has been offered to anyone |
| Marginal cost per assessment is cents, not dollars | Assumed | Follows from running the models locally, but no run has been metered end to end. Google Places is the only paid call in the pipeline |
| A local model is good enough for the synthesis stage | Assumed | The design keeps the model away from sourcing facts, which is what makes this plausible. The Mako report was not produced this way, so it is untested |
| Machine throughput at volume | Unknown | Performance runs and local generation take real minutes each. One machine's capacity, not money, is the ceiling |
| It stays accurate without a person checking | Unknown | The self-audit gates format and internal consistency, not whether a finding is fair |
Being wrong about a named business is the failure that ends it. One false statement in a document sent to a stranger costs the relationship and the reputation together. It nearly happened on the first run twice: a stale third-party record almost produced a false compliance alarm, and a simulated load time nearly went out as though it were measured. Both were caught, and the pipeline now has stages specifically because they were not.
Unsolicited email is a regulated activity and a reputational one. Getting the compliance right is table stakes; getting the tone right matters more. The teaser has to read as someone who did real work on your business, not as a mail merge.
The moat is thin at the start. Anyone can point a model at a website. What is harder to copy is the verification discipline, the fixed rubric that makes assessments comparable, and the accumulated per-vertical research. Those get stronger with volume, which is another argument for picking one vertical and going deep.