Every operator we talk to has the same AI problem: too many subscriptions, no order of operations, and no idea which tool actually moves the needle in their business. The fix we keep coming back to is not another automation build pitched cold. It is a paid AI assessment: a short diagnostic engagement that earns trust before we touch a single workflow.
Stop selling implementation first
Most agencies open with a pitch. They show up with a deck full of automations and ask a business owner to commit budget to a build before that owner has any idea what is actually broken in their operation. That order is backwards. A busy operator running a 10 to 50 person company does not know which five hours of their week are recoverable. They just know they are drowning in email, follow-up, and repetitive admin work.
We run our AI assessment as the front door instead. It is a paid, structured conversation that surfaces the actual bottlenecks in a business before we propose anything. Clients pay for it because a diagnostic with a real deliverable feels different from a sales call, and because we credit the full amount toward the build if they move forward. That single mechanic changes the entire trust dynamic: the client is not gambling on a stranger's promise, they are buying a report they get to keep either way.
The four-phase system behind the assessment
We treat the assessment like a product with a repeatable production line, not a one-off consulting hour. Four phases, in order, every time.
Phase one: a recorded discovery conversation, not a pitch
The only job in phase one is to ask questions and stay quiet. We are not prescribing anything on this call. We ask what a typical day looks like, where work piles up, what the owner has already tried and abandoned, and what process they would delete first if they could wave a wand. The answer is almost always something mundane: email triage, following up on leads, chasing invoices, scheduling. The call gets recorded through an AI note taker so nothing gets lost. Tools like Fathom, Otter, and Fireflies all do this job well; the specific vendor matters less than making sure a clean transcript exists afterward.
Phase two: hand the transcript to Claude, not a template
This is the phase most agencies skip, and it is the one that makes the assessment defensible. We feed the raw call transcript to Claude and ask it to pull out every distinct pain point mentioned, then research off-the-shelf tools or system designs that address each one. A model reading a full transcript catches patterns a human running back-to-back calls will miss. A scheduling complaint mentioned in passing connects to a staffing complaint raised later in the same call, and a good analysis pass flags that as one underlying bottleneck rather than two separate ones.
The prompt does not need to be elaborate. Attach the transcript, state the business context, and ask for a ranked list of pain points mapped to concrete tools or systems that fix them. The output is a draft, not a deliverable.
Phase three: the quality pass no one gets to skip
Raw model output goes to a human before it goes to a client. A four-person landscaping business does not need an enterprise CRM just because the model matched "no follow-up process" to a well-known enterprise tool. Someone on our team reviews every recommendation against the size, budget, and technical comfort of the actual business and swaps anything oversized or underpowered. This is also where we decide which pain points get a prescribed tool and which get a proposal for something we build outright. A recurring theme across five or more clients usually means the answer is a system, not a subscription.
Phase four: the report and the upsell path
The client gets a short report: the bottlenecks we heard back to them in their own language, and a ranked list of fixes. Some fixes are a tool they can turn on themselves. Others are systems we build for them, which is where the assessment converts into real engagement value. In our world that means an automation build, an AI Operations Agent that works the pipeline while the team sleeps, or a voice agent picking up the calls that used to go to voicemail. The assessment is the diagnosis; the build is the treatment.
Why a fixed AI stack becomes a liability
Part of the case for running this as an ongoing service, not a one-time report, is that the tool landscape underneath it keeps moving. Pricing tiers change, models get renamed, and the "obvious" recommendation from six months ago can be the wrong one today. Anyone recommending AI tools needs to check current terms before promising a stack, not rely on memory. We keep our own recommendations current against primary sources like Anthropic's pricing page rather than screenshots or secondhand claims, because a client who gets burned by a stale recommendation does not come back for the upsell.
This is also why the assessment beats a static automation audit template. A template gets stale the moment a vendor changes a plan. A conversation plus a live research pass stays current because the research happens fresh, every time, against whatever the landscape looks like that week.
| Phase | What happens | Who owns it |
|---|---|---|
| Discovery call | Structured questions, zero pitching, call recorded | Human, AI note taker |
| Transcript analysis | Pain points extracted, tools and systems researched | Claude |
| Quality pass | Recommendations sized to the actual business | Human |
| Report and upsell | Ranked fixes delivered, build path proposed | Human plus system |
What we are building this quarter
We are treating the assessment as the front door to every new client relationship, not a side offer. Concretely, that means three things on our own roadmap:
- Tightening the phase-two prompt so the research pass checks pricing and feature claims against a live source before it reaches the quality-review step, instead of trusting a model's training data on tool pricing.
- Building a standard upsell menu attached to every report, so a client who wants five things fixed sees a clear path from the fix to the build. It is the same logic behind our product lineup.
- Pointing every assessment client at our revenue leak heatmap before the call even happens, so the discovery conversation starts from a working hypothesis instead of a blank page.
None of this replaces the human judgment in phase three. The model finds the pattern; a person still has to decide whether a four-person business needs an enterprise tool or a fifty-dollar-a-month one. That judgment call is the actual service we sell.
The economics that make this a front door, not a favor
Our assessment is priced at €999, and the full amount is credited to the build if the client moves forward, so the diagnostic never feels like a sunk cost to them. The businesses that get the most value are the ones losing 10 or more hours a week to admin work that a properly scoped system can absorb. When we do move into a build, the systems we hand over run around the clock. The client owns 100% of the resulting code and files, so there is no lock-in argument to overcome later. That combination, a low-friction paid diagnostic plus full ownership on the back end, is why this model works as a front door.
We have documented how this plays out across real engagements in our case studies, and we write about the systems side of this work regularly on the blog. If you are an operator staring at a pile of AI subscriptions with no order of operations, the fix is not another tool. It is a diagnostic that tells you which five things actually matter, followed by someone who will build the ones that do.