Conversation watcherour own inbox
53 574messages read
The audit of our own inbox found conversations that went quiet; a watcher now flags about a third of the ones we would have dropped. our own system · Aug 2026
Business systems · AI consulting & audit
AI consulting for a practical first step.
One call, a look at how enquiries, records, bookings and invoices move through your company today, and a short plan: the two or three places where automation may be useful, what each costs from our ladder, and what should stay manual. No obligation to build with us.
Audit · illustrative companyscoring
0of 24 points
where a system helps
where a system helps
reading the company
- Enquiries0/4How a customer reaches you and how long they wait
- Records0/4Where customer data lives and who can find it
- Calendar and bookings0/4What is booked by hand and what falls through
- Invoices and payments0/4What is typed twice and what is chased late
- Website0/4Whether it starts the enquiry or ends it
- Reports0/4What the owner compiles by hand on Friday
higher score = more manual work to removehover a tile to hold it
What you get
A map of how work moves today
Enquiries, records, bookings, invoices, reports: where each lives, who touches it, where it waits. On one page, in your words.
Two or three places to start
The processes with the clearest potential benefit, each with a price from the ladder and a note on what stays manual and why.
A plan you can take anywhere
No lock-in: the plan is yours, priced against public numbers, and you can build it with us, with a ready-made tool, or with anyone else.
How it works
One call, one look, one plan
Hover a step to hold it.
- 01The call
About an hour with the person who knows how the work runs. We ask how a customer reaches you and what happens next.
- 02The look
Selected examples or a demonstration first. Any further read-only access is limited to what the audit actually needs.
- 03The scoring
Each area scored for manual work and waiting time; priority also considers accessible data, implementation cost, review effort and the consequences of errors.
- 04The plan
A prioritised scope, estimated costs, dependencies and what stays manual. The delivery date is agreed for the audit.
- 05Your decision
Build with us, use a ready-made tool, or wait. The plan is yours either way.
Who this is for
Owners who want to choose a practical first improvement
A company of three to thirty
Several people share enquiries, records and handovers. We agree which processes the audit covers instead of assuming the whole company fits one exercise.
An owner who does the reports
Friday evenings spent compiling numbers that already exist in four tools.
A business that tried a tool and dropped it
We examine the tool, data, setup and surrounding process to understand what failed before recommending another implementation.
A company before hiring
Before changing staffing, compare the work automation could support with the judgement, review and customer care that still need a person.
Proof
We audited ourselves first
The method was built on our own studio before it was offered to anyone. What it found, with sources.
Clinic websitesown research
272US clinic sites measured
92.4% show a booking button; 36.7% connect it to a system that can take the booking. The same audit method, applied to an industry. oris.agency research · Aug 2026
This websitebuilt on the plan
1 planthen one page at a time
The site you are reading was planned the same way: tree and priorities first, then pages, each accepted before the next. headpills.com · Sep 2026
No client deployment yet as a standalone consulting engagement. Audits so far were part of build projects; the first standalone one is priced after a call.
Price
Price after a short call. Fixed for the agreed audit scope.
Everything the plan proposes is priced on this site’s ladder, so you can check it before you agree.
Begin with a business decision, not a tool shortlist
An AI audit should help an owner decide where to invest, what to test and what to leave alone. A list of fashionable tools does not answer those questions. We begin with work the business already performs: enquiries, document preparation, reporting, scheduling and coordination between people. The useful opportunities are the ones connected to a measurable problem and an owner who can change the process.
A problem with evidence
Describe the problem through examples. A proposal may be delayed because information is scattered; a lead may be missed because nobody owns the handover; a report may take too long because its data uses inconsistent categories. Each situation suggests different remedies. Adding an AI layer before understanding the cause can make an unreliable process harder to inspect.
A decision the audit should enable
The audit should end with a decision-ready scope. That means a proposed first task, the expected inputs, operating boundaries, dependencies and a way to evaluate it. It should also explain the alternatives and why some ideas were postponed. A useful recommendation remains understandable even if you choose another supplier to implement it.
Compare opportunities on four practical dimensions
- Frequency and effort
- How often does the task occur, how much time does it take and how predictable is the workload? Repeated low-complexity handling can be worth investigating, while a rare judgement-heavy task may not justify a dedicated system. We distinguish measured time from a rough estimate and include the effort of reviewing the output. Claimed savings should not assume that all human involvement disappears.
- Data and access
- Does the information exist in a usable form, and can the necessary tools be accessed legitimately? An archive of conflicting documents may need curation before an assistant can use it. A system may have an API but require a plan or permission the business does not possess. These constraints affect feasibility and cost, so they belong in the assessment rather than in a late implementation surprise.
- Consequence and ownership
- What happens if the output is wrong, and who is responsible for the result? Drafting a summary differs from confirming a payment, giving sensitive advice or changing a customer record. We identify the review points and the people available to handle exceptions. An automation without an operational owner can become another unattended tool instead of a business improvement.
Separate process improvement, automation and AI
An audit should be free to recommend a simpler solution. Sometimes the company needs one shared record, clearer responsibilities or a standard template. Those improvements can make later automation more valuable, but they may also resolve enough of the problem on their own. The recommendation should follow the evidence rather than a commitment to build an AI product.
Improve a predictable process
Predictable tasks can often use ordinary rules: route a form by service, copy an approved field, schedule a reminder or prepare a report from structured records. Their behaviour is easier to test because the intended result is known. We look for opportunities to simplify the process before adding a language model to every step.
Assist an interpretive task
AI becomes more relevant when the task involves varied language or material that needs interpretation. It may help draft a response, summarise a conversation or identify information in a document. These uses still need a defined source, validation and a suitable review process. The audit should state where a model is helpful and where its uncertainty would create unnecessary risk.
Turn the selected opportunity into a pilot
Record the baseline
Capture the present workflow and a representative sample of its volume, handling time and errors. If reliable historical data are unavailable, agree a short observation period. The baseline does not need to be elaborate, but it must be comparable with the pilot. Otherwise every later judgement depends on memory and enthusiasm rather than observable change.
Define the acceptance cases
Choose normal examples and difficult examples, with the expected result for each. Include missing information, contradictory instructions and a failed connection where relevant. Decide which errors are tolerable during a controlled trial and which prevent launch. This converts an abstract goal such as better customer service into behaviour that can be examined.
Set the operating boundary
Limit the first audience, channels and permitted actions. A pilot can prepare drafts for staff before it communicates directly with customers. Name the reviewer, the pause mechanism and the manual fallback. Expansion should follow evidence from the initial scope. Adding several new tools and responsibilities before evaluating the first version makes the result harder to understand.
Read an audit recommendation critically
Costs are separated
The proposal should distinguish implementation cost from provider subscriptions, usage charges, hosting and ongoing support. It should identify assumptions behind the operating estimate. A cheap demonstration may become expensive if it processes long documents or needs constant review. Compare the full operating arrangement over a meaningful period, including the work your staff still performs.
Dependencies are explicit
Check whether the recommendation depends on access to particular accounts, data preparation or participation from your team. Those dependencies affect the schedule and should have owners. A promised launch date without the necessary permissions is not a reliable plan. The same applies to integrations described as automatic before their available operations have been inspected.
Success is observable
Look for a clear decision at the end of the pilot: continue, narrow the scope, improve the source material or stop. Success should relate to the business task, such as fewer missed handovers or less preparation time at acceptable quality. A polished interface or a large number of model calls does not establish that the investment is worthwhile.
Prepare for an audit without exposing everything
Start with selected examples
Bring the person who understands the workflow and a few representative examples of the work. A process description, anonymised records and a demonstration can be enough to begin. Broad access to an entire mailbox is not the default requirement. We identify the information actually needed and agree appropriate access as the scope becomes clearer.
Tell us which tools you use, the approximate workload and the constraints that matter: budget, review capacity, language, hosting or sensitive information. Include previous automation attempts and why they were abandoned. Those details can prevent repeating a failed approach and reveal whether the first investment should be in documentation, data quality or an integration.
The audit is priced after the initial discussion, according to the number of processes, systems and people involved. Its output should be a concise diagnosis and a prioritised, reviewable plan with estimates and uncertainties. It creates no obligation to commission a build. The value is a better decision about where AI and automation can realistically help your business.
Questions
Questions owners ask before an audit
What is an AI audit for a small business, and what do we get?
A short written plan: how your enquiries, records, bookings, invoices and reports move today, the two or three places with the clearest potential benefit, what each would cost from our ladder, and what should stay manual. We start with a call and selected examples, agree any further access and scope the analysis. There is no obligation to commission a build with us.
Do you build custom AI solutions for small business, or only advise?
Both, in that order. The audit names the processes; if you want them built, each is a fixed-price project from the same ladder as every page on this site. If a ready-made tool does the job, the plan says so instead of proposing a build.
Do you offer data analytics consulting for a small business?
In the practical sense: which numbers you should see every week, where they already exist in your tools, and how to get them onto one screen without compiling them by hand. We do not run data-science projects; we connect what you have and make it readable.
Where does AI pay off first in a small company?
The first candidate depends on workload, available data, review effort and the consequence of error. Enquiry preparation, handovers and reporting can be useful starting points, but there is no universal order. The audit compares your processes and may recommend an ordinary rule or a simpler manual improvement instead of AI.
What does the audit need from our side?
A process owner, representative examples and an explanation of the tools involved. We start with the minimum necessary information and agree any further access. The scope and delivery date depend on the number of workflows and the availability of evidence.
How much does AI consulting cost?
The audit is priced after the first call, depending on how many tools and people are involved. What we can say now: it is a fixed price, the price of anything we propose is on this site’s ladder, so you can check it before you agree.
Tell us how the work runs today.
One call, one look, and a plan within a week.