Requests arrive in five places
Web form, phone, shared inbox, a chat channel, a message someone received personally. Whether a request gets answered depends on who happened to be looking at which app.
AI Operations
AI operations: get the work out of people’s hands and into a system that has a record.
TheZaraAI builds AI automation for operations teams — intake, qualification, scheduling, reporting, and handoffs, run by agents on the tools you already use. Every action is logged, attributable, and reversible. Prices are published below.
Free · No credit card · You leave with a ranked list of what to automate first
U.S. Army Special Forces veteran
Cyber Electronic Warfare Officer
RSA Conference speaker
One workflow · one morningAudit trail
08:02:14Request received — web formsystem
08:02:51Reply sent, two qualifying questions askedagent
08:09:06Rules checked — territory, urgency, authorityrule
08:09:07Escalated to a person — no decision authorityrule
08:14:22Owner assigned, record written to the CRMagent
08:14:23Line added to the weekly operations reportagent
Illustrative interface with fictional data. Not live customer activity, and not a performance claim. The point is the shape of the record, not the numbers in it.
The problem
The work is being done by hand, and nobody can prove what happened.
Not the strategic work. The daily mechanics that decide whether a lead is answered, a job is scheduled, or a report is right. Below is where the hours actually go in an operations team under $100M in revenue.
The rules live in one person’s head
Territory, urgency, budget, decision authority. Everyone applies them slightly differently, and nobody writes down why a request was passed over.
Booking takes four messages
Offer times, wait, lose the slot, offer again. Meanwhile the calendar has capacity that nobody is looking at, and the customer has called a competitor.
Monday morning is spent rebuilding the numbers
Someone exports, pastes, reconciles, and formats. By the time the report is circulated it describes a week that is already over.
Context is retyped at every boundary
Sales to operations, operations to delivery, delivery to invoicing. Each pass loses detail, and the loss only surfaces when a customer complains.
There is no record
When something goes wrong you can reconstruct roughly what happened by asking people. That is not a record. It is a recollection, and it arrives after the damage.
This is not a tooling problem, and adding an AI pilot on top of it does not fix it. 37% of organizations report using AI only at a surface level, with little or no change to underlying processes, and only 21% report a mature model for agent governance (Deloitte, State of AI 2026). Only 9% of organizations have fully deployed an AI use case (Accenture).
The risk is not using AI badly. It is running it without a record.
What gets built
Six deliverables. Each one names what you receive.
Agents run on the stack you already have — email and calendar, file storage, Slack or Teams, Notion, your CRM, your scheduling and ticketing tools. Nobody is asked to move into a new dashboard.
One intake queue
Every inbound route lands in a single queue with the source stamped on it. Missed calls trigger an immediate reply and continue by message.
You receiveA connected intake path per channel, with the routing rules written down and a named owner for each queue.
A qualification rulebook the agent applies
Territory, urgency, budget, and decision authority checked the same way every time. Anything failing a rule stops and goes to a person.
You receiveA written rulebook, the escalation paths, and the explicit list of decisions the system may never make alone.
Scheduling into real capacity
Only genuinely open slots are offered. Assignment follows your rules — route, skill, workload — and the confirmation and calendar hold are written immediately.
You receiveA live calendar integration, the assignment logic in writing, and the confirmation templates your team can edit.
Reporting that assembles itself
The numbers are written as the work happens rather than reconstructed on Monday. Definitions are fixed once so two people cannot produce two answers.
You receiveA scheduled report on the cadence you choose, plus a definitions sheet naming exactly how each figure is counted.
Handoffs that carry their context
The record moves between stages with everything attached. No retyping at a boundary, and no detail lost because the person who knew it was on leave.
You receiveA mapped handoff for each boundary, the fields that travel with it, and the exception path when a field is missing.
An audit trail on every action
What acted, on what basis, at what time, and what it touched. Written before go-live, not added after an incident makes someone ask for it.
You receiveAn audit configuration you could show an auditor, and an access inventory naming what each system can reach.
How it works
Five stages. A decision point at every one.
The same process the whole practice runs on. No stage begins before you have seen what the last one produced. You can stop at any gate and keep everything built so far.
01
Discover
A free 15-minute call, then a short working session. We walk your intake, your queues, and your Monday reporting, and rank the friction by what it costs you — usually not where people expect.
- You receive
- A written map of current state, friction ranked by cost.
- Risk control
- Read-only access. Nothing is changed, connected, or granted.
- Your call
- Do you agree with the diagnosis, and is the top item worth fixing?
02
Design
We specify the workflow before building it — triggers, qualification rules, exceptions, escalation paths, and the explicit boundary of what the system may never decide alone.
- You receive
- A workflow spec and data-flow diagram any competent engineer could build from.
- Risk control
- Data classes identified up front; sensitive material scoped out or protected.
- Your call
- Does the spec match how you would want a good employee to behave?
03
Build
Implementation against the spec, in test mode — your real tools, your real data shapes, but not your real customers until you say so.
- You receive
- A working system you can drive yourself, plus the test evidence behind it.
- Risk control
- Outbound is gated. Nothing reaches a customer while it is still being proven.
- Your call
- Having watched it run on your data, do you trust it to talk to a customer?
04
Secure
The stage most automation work skips. Access is narrowed to what each step needs, credentials move into managed storage, and the audit trail goes on before go-live. Deeper testing and governance sit in the Secure AI practice.
- You receive
- An access inventory and audit configuration you could show an auditor.
- Risk control
- The blast radius of any single compromised credential is established in writing.
- Your call
- Are you comfortable with what each system could reach if compromised tomorrow?
05
Operate
Go-live, then the unglamorous part: watching it, tuning rules as your season shifts, and keeping it working as vendor APIs move underneath it.
- You receive
- Ongoing operation, or a clean handoff to your team — your choice, stated up front.
- Risk control
- Patching and dependency review continue. An unmaintained automation is a liability.
- Your call
- Do you want us operating this, or do you want the keys?
Outcomes & evidence
What you hold at the end of it.
Automation is easy to demo and hard to trust. These are the artefacts an AI Operations engagement produces — the things that outlive the enthusiasm of week one.
The running agents
The live workflows themselves, on your accounts and your credentials, doing intake, qualification, scheduling, reporting, and handoffs. You can switch any of them off without calling us.
The audit trail
Every automated action attributable, timestamped, and reversible. When something goes sideways you can see what acted, on what basis, and when — before you have to explain it to anyone.
The runbook
How to operate it day to day: what each workflow does, how to change a rule, what the alerts mean, what to do when a vendor API moves, and who to call in what order.
The handover
Credentials, accounts, specs, and diagrams transferred to your team, written so they can operate and modify the system without us. Dependence is not a business model we are interested in.
The workflow map
Every trigger, rule, handoff, exception, and escalation path documented — including what the system deliberately will not do on its own.
The access inventory
A plain list of which system can reach which data, under whose credentials, with least-privilege applied and the remaining gaps named rather than hidden.
We do not publish customer counts, revenue impact, time-saved figures, or security outcomes on this site. Numbers like those are trivial to invent and impossible for you to verify, so treating their absence as a signal is reasonable. Client references are available on request and with client permission, matched to your sector. Ask on the call.
Pricing
Here are the prices.
Published, in text, before you speak to anyone. A fixed-scope build is the right first step if you want proof before a relationship. A monthly engagement is the right step once you have more than one workflow to fix.
Start with a one-time build
Fixed scope, fixed price, yours to keep. The care plan carries the agent after handoff.Starter Agent
$325 one-timeHand your email and your files to a lean assistant. Built in under 30 days.
- Email triaged and answered
- File storage organised
- Core automations configured
- Written handoff + a month of light support
Integrated Agent
$789 one-timeConnect your calendar and the tools your day actually runs on. Roughly six weeks.
- Everything in the Starter Agent
- Calendar — invites drafted and sent
- Up to 5 more integrations (Slack, Notion, CRM…)
- Custom skills & templates
Care Plan
$197 / monthKeeps the agent running as your tools and their APIs change. The default next step after handoff.
- Monitoring on the live workflows
- Updates and fixes as vendors change
- The on-ramp to a monthly engagement
Work with us monthly
The ongoing engagement. Build capacity and operation together — cancel or change tier as your season shifts.Reconnaissance
$1,424 starting / monthGet one workflow off your plate at a time, done properly. The usual entry point.
- One workflow at a time — automated, then maintained
- Bi-weekly strategy calls
- Monthly reporting · email support
Operations
$3,026 starting / monthShip new automations while what is already live keeps getting tuned.
- Several workflows built and optimised in parallel
- Tooling integration & documentation
- Weekly strategy calls · priority response
Command
$8,010 starting / monthRun AI operations with an embedded partner. Security leadership included, hosting covered.
- Dedicated build-and-governance capacity
- Full AI workspace architecture
- AI security audit included
- Dedicated lead · embedded weekly cadence
- Hosting, monitoring & governance included
All prices in USD. Reconnaissance and Operations add $325/month for hosting & security — managed hosting, uptime monitoring, updates, and patching. That is already inside Command. One-time builds cover a fixed scope; anything beyond it is quoted separately and can roll into a monthly engagement.
See the full pricing section and the engagement selector on the homepage — four questions, and you get the engagement shape that fits with its actual price.
The AI Field Manual
Building & Securing AI Agents
Free download · two questions
Download the AI Agent Field Manual.
The practical guide we wrote for people who want to build and secure their own agents. Answer the two questions below and it downloads straight away, with a copy to your inbox. If it saves you hiring us, that is a perfectly good outcome.
Questions
The things operations leaders actually ask.
The one that is high volume, rule-shaped, and already documented in someone’s head. Intake and qualification usually win, because they run every day, they follow rules a person can state out loud, and a delay there costs a deal rather than an hour. We rank candidates on the discovery call and write the ranking down. Some of what you want automated should not be — we will say which, and why.
That is the constraint we build to. Email and calendar, file storage, Slack or Teams, Notion, CRMs, scheduling and ticketing tools, and most systems with a usable API. Nothing here asks your team to move into a new dashboard. If a system you depend on has no API, we say so in the design stage rather than building a fragile workaround and handing you the maintenance bill.
It escalates instead of guessing. Every workflow is specified with the rules it must satisfy and the boundary it may never cross alone. Anything that fails a rule, or falls outside the rules entirely, stops and goes to a named person with the context attached. When something does go wrong, the audit trail tells you what acted, on what basis, and when, so the fix is a rule change rather than an investigation.
Start with a build if you want proof before a relationship. The Starter Agent is $325 one-time and the Integrated Agent is $789 one-time, both fixed scope, both yours to keep. The $197/month care plan carries the agent after handoff. Move to a monthly engagement when you have more than one workflow to fix — Reconnaissance starts at $1,424/month. A build can roll into a retainer later; nothing is rebuilt to make that happen.
You own it. The accounts, the credentials, the workflow spec, and the runbook are yours, written so your team can operate and change the system without us. You then choose: keep it in-house, take the care plan, or move to a monthly engagement. Dependence is not a business model we are interested in, and the handover document is written on the assumption that you might leave.
The Starter Agent is delivered in under 30 days and the Integrated Agent in roughly six weeks. On a monthly engagement the first workflow usually reaches live operation inside the first month, because the build stage runs in test mode against your real tools before anything is allowed to reach a customer. You see working output at every gate, not a status update.
15-minute discovery call
Book fifteen minutes.
A short working call, not a sales sequence. We walk your intake, your queues, and your reporting, and name the specific places manual effort is costing you. Free, and it books straight into the calendar.
Free · No credit card · No obligation to buy anything afterwards
- Where your inbound actually lands, and how fast it gets answered
- Which repetitive work is worth automating first, and which is not
- What your systems can currently reach that you have not thought about
- A recommended engagement shape, or an honest “you do not need us yet”
Before the call
Read what gets built — six deliverables, each with what you receive.
Read the five stages — and the decision you make at each gate.
Read the prices — builds from $325, monthly engagements from $1,424.
Read the Secure AI practice — for knowing what your AI did, who told it to, and what it touched.
Back to TheZaraAI home — the interactive demo, the founder, and the full FAQ.
Read first, decide later
The five workflows every business should automate first — how we rank candidates before anyone spends money.
What an AI agent actually is — the plain definition, without the pitch.
Putting an agent on your email — the workflow most teams start with.