Our approach
Thirty days from first conversation to something running.
The first few days are the ones that matter. That is where we find what is worth automating, cost it, and tell you honestly whether it is worth doing.
01Readiness assessmentWhat data exists, what condition it is in, and what the business is realistically ready to automate. Days one to four.See what you get
66/120
Foundations partly in place
Scored across eight dimensions. Two are strong enough to build on. Three have to be closed before automation will hold.
Illustrative. Your scorecard names the specific gaps and what closing each one requires.
02Use cases identified and rankedWe surface the candidates and rank them on value to you and on how cleanly they can be built.See what you get
Twenty-two candidates. The first twelve shown.
1Order entry from email
2Internal knowledge search
3Supplier invoice capture
4Quote drafting assistant
5Inventory reorder
6Meeting notes and actions
7Report distribution
8Month-end close pack
9Lead capture and routing
10Customer chat
11Competitor price monitoring
12Supplier purchase orders
Cost eliminationValue add
Every candidate placed by cost and risk, so the order of work is obvious before anything is approved.
03Process mappingWe map the processes involved, or work from your documentation where it already exists. Agreed with you within about two weeks.See what you get
| No. | Flow | Type | Trigger | Integrations |
|---|---|---|---|---|
| 01 | Consolidate BoardReads every tab into one deduped view of the week. | Agent tool | MCP tool call from the agent | GS |
| 02 | Capture TaskTurns a request into a row, with the requester recorded. | Agent tool | MCP tool call from a forwarded email | GS |
| 03 | Triage BacklogClassifies unassigned tasks and, on approval, routes them. | Agent tool | MCP tool call from the agent | GS |
| 04 | Nudge OwnerEmails the owner of an overdue task, and logs it. | Agent tool | MCP call, or from the weekly digest | OL |
| 05 | Weekly Coordination DigestThe governance loop. Fires on a schedule, not on the agent's decision. | Scheduled | Schedule, Monday 07:00 ET | GSOL |
Flow numbers are stable. They appear on the blueprint and on every specification sheet.
04Quantification and approvalEach opportunity costed by effort and value. You approve what gets built. Nothing starts before this.See what you get
| Use case | People | Hours / week each | Loaded rate | Annual value | Build | Payback |
|---|---|---|---|---|---|---|
| Order entry from email | 3 | 4.0 | $38 | $20,976 | $9,000 | 5.1 mo |
| Supplier invoice capture | 2 | 5.5 | $42 | $21,252 | $11,000 | 6.2 mo |
| Report distribution | 1 | 3.0 | $55 | $7,590 | $4,000 | 6.3 mo |
| Month-end close pack | 4 | 2.5 | $48 | $22,080 | $14,000 | 7.6 mo |
| Four of twenty-two | 10 | $71,898 | $38,000 | 6.3 mo |
How the annual value is calculated
hours saved per weekpeople affectedloaded hourly rate46 working weeks = annual value
4.0 × 3 × $38 × 46 = $20,976
Loaded rate is salary plus employer cost, not base pay. Forty-six weeks allows for holiday and statutory time. Hours come from the process mapping, not from an estimate, which is why this step follows it rather than leading.
You approve line by line. Anything deferred stays on the list for the next cycle.
Illustrative. Your numbers come from your own hours and rates, gathered during the mapping.
05Build, test, iterateBuilt on our platform, then thirty days of trial with your team before production.See what you get
One of the flows exactly as it is built on the platform, not a diagram of it.
06Hand overWe train the person who will own it, or we take it on ourselves. Decided with you at the end of the build, on what we actually found about your capacity.
A worked example
It starts with a problem, not with an agent.
Every engagement produces a build specification, and you approve that document rather than an idea. This is a real one, for an agentic receptionist we built as a demonstration.
A trade business answers the phone with whoever happens to be free. When nobody is free the call goes to voicemail, and most of those callers ring the next number instead.
Calls go unanswered, booking costs a conversation, follow-up happens when somebody remembers, and the owner finds out how the week went at the end of it. Every one of these is a measurement problem before it is a technology problem. You cannot improve a booking rate you have never seen.
Three channels in, six callable tools out, and one scheduled report it cannot invoke.
Flo answers on the website, the phone and by email, with the same reasoning behind each channel. It decides which tool fits and reports what came back; it never computes the numbers itself. The six tools are workflows rather than guesses, and each returns a predictable result. The daily report is deliberately outside its reach.
Solid lines are the agent calling a tool. Dashed lines are the structured result coming back.
Start here
Start with a readiness assessment
Days one to four. You find out what is worth automating, what it would cost, and whether it is worth doing at all.