Odokai
Ask Odokai

How can AI help my business?

Ask in your own words. Every answer is written by the people who build Odokai: what it looks like in practice, what it costs, and whether a call is worth your time.

Common questions

Every answer

Browse by topic.

The questions business owners ask us most, answered plainly. Open one to see how it works in practice and whether it is worth a call.

Getting started

How can AI help my business?

Start with one job, not a strategy. Pick something your team does every week that is repetitive and document-heavy, and that someone dreads: renewals, month-end commentary, invoice checks, the Friday client update. Automate that properly, then use what it sets up for the next one.

In practice

  • Write down the three jobs that take the most hours each week. The best first candidate usually has clear rules and a person who checks the result anyway.
  • On a free 45-minute call we build an agent for one of them, on your own documents, while you watch.
  • If it is worth doing, the first workflow goes into production in one to five weeks for a fixed fee.
  • The second workflow reuses the connections, rules and approvals from the first, so it takes days.

Worth a call if

  • You can name a job that takes someone more than a few hours a week.
  • The job works from documents, emails or spreadsheets you already have.
  • Someone already checks the output by hand.

Probably not for you if

You want a general AI assistant for everyone and no particular job in mind. One of the big assistants is a simpler buy.

See ten jobs agents runHow Odokai works

Which tasks can AI automate in my business?

Good first jobs are repetitive, rule-based and close to documents: reading, checking, assembling and drafting. The result should be easy for a person to check, and the job should come round every week or month, so the time saved adds up.

In practice

  • Strong candidates: renewal packs, management accounts commentary, invoice checks against orders, a first read of contracts, ticket triage, lead research, the weekly client update.
  • Weaker candidates: one-off projects, jobs that depend on a conversation nobody wrote down, and decisions with no clear rules.
  • A useful test: could you write the instructions for a capable new starter on one page? If so, an agent can usually follow them.
  • We score candidates on hours saved, risk if it goes wrong, and how easy the result is to check.

Worth a call if

  • You have a few candidates and cannot decide which to start with.
  • A job is done the same way every time but still takes hours.

Ten jobs, step by step

What happens on the discovery call?

It takes about 45 minutes and costs nothing. You bring one job and a few real documents. We build an agent that does the job while you watch, and you leave knowing whether it is worth automating. You keep the agent we build.

In practice

  • Before the call: pick the job and gather three to five typical examples of the documents involved.
  • On the call: we talk through how the job is done today, build the agent on your examples, and run it.
  • After the call: if it is worth doing, you get the acceptance criteria and a fixed fee for the first workflow in writing.
  • Nothing is committed until you sign off the criteria. If it is not worth doing, we will say so.

Worth a call if

  • You want to see an agent work on your own material before spending anything.
  • You need something concrete to show a partner, director or board.

Book a discovery call

Can we run a pilot first?

We skip the proof of concept. The discovery call shows an agent working on your documents for free, and the first paid build goes straight into production, with edge cases handled, your systems connected and your team trained.

In practice

  • A proof of concept proves the easy 80 percent. The work is in the other 20: exceptions, odd documents, permissions and sign-off.
  • The first build is scoped against acceptance criteria you sign off, so success is defined before we start.
  • If you want to test more widely, the second and third workflows cost no build fee and your team can build them.

Worth a call if

  • You have run AI pilots that never reached production.
  • You need a measurable result rather than a demo.
How long does it take to implement AI?

The discovery call takes 45 minutes. The first workflow is live in production in one to five weeks, depending on the job and the systems involved. The second usually takes days, because the platform, connections and approval rules are already in place.

In practice

  • One week: a job that works from documents and email, with one approval step.
  • Three to five weeks: several systems to connect, more exceptions to handle, or a careful compliance sign-off.
  • Most of the elapsed time goes on access to your systems and agreeing the rules, not on building the agent.
  • On the shared platform you can start the same day. Your own instance takes a few days to set up.

Worth a call if

  • You have a deadline, such as a busy season or an audit, and want to know what is realistic.
Is AI worth it for a small business?

Odokai is built for firms of a few dozen to a few hundred people, where a handful of repetitive jobs take a real share of the week. If you are much smaller, check the sums first: the platform starts at £500 a month ex VAT.

In practice

  • The sum to run: hours a week on the job, times the loaded hourly cost, times 52. Compare that with the platform fee and the one-off build.
  • Small firms usually start on the shared platform: no infrastructure to manage, and live the same day.
  • You do not need an IT team. We build the first workflow with you and train whoever will run it.

Worth a call if

  • One or two jobs take more than a day a week between them.
  • You are growing and would rather not hire for repetitive work.

Probably not for you if

Your whole team is two or three people and no single job takes more than a couple of hours a week.

Cost and return

How much does AI cost for a small business?

There are three parts. The discovery call is free. The first workflow is a fixed fee, quoted in writing after the call. Then there is a monthly platform fee: from £500 a month ex VAT on the shared platform, or £1,000 a month for your own instance.

In practice

  • The build fee is fixed against acceptance criteria you sign off before we start. It does not change once we begin.
  • The monthly fee includes model usage up to a fair-use volume we agree with you, model upgrades, infrastructure and support.
  • Later workflows carry no build fee if your team builds them. If you want us to, each is a small piece of work.
  • In your own AWS account you hold your own model contracts and pay the providers directly.

Worth a call if

  • You want a fixed number in writing before committing.
  • You need to compare a monthly fee with per-seat licences or a hire.

Pricing in full

What return can we expect from AI?

The return comes from hours your team gets back on work that recurs. Count the hours a job takes each month, cost them at loaded rates, and compare with the fees. Where a job saves two hours a case and runs a few hundred cases a year, it pays back quickly.

In practice

  • Example: 40 invoices a week at 10 minutes each is about 350 hours a year. At £35 an hour loaded, that is roughly £12,000 on one job.
  • The first workflow carries the build fee. Each later one adds hours saved without another build fee, so the return improves.
  • Count the quieter gains too: work done before anyone arrives, fewer missed renewals, faster replies.
  • We work the numbers through with you on the discovery call, using your volumes rather than averages.

Worth a call if

  • You need a business case for a director or a board.
  • You know roughly how many hours a job takes each month.
How do AI usage costs work?

Model usage is included in the monthly fee up to a volume we agree with you, and you set a hard cap. Costs stay low because each step runs on the model that suits it: classifiers do the sorting and checking, and the expensive models only do the hard reasoning.

In practice

  • In an example month, 38 percent of steps ran on classifiers, 33 percent on fast models, 21 percent on mid-tier models and 8 percent on the top tier.
  • On three everyday jobs, matching the model to the step cut the cost per million tokens by 60, 83 and 95 percent.
  • If you go past the allowance, the overage rate is already in your contract. A warning threshold tells you first.

Worth a call if

  • You have been surprised by an AI usage bill before.
  • Your volumes are high, such as thousands of documents a month.

The cost table on the home page

Why not pay per user for an AI assistant?

Per-seat assistants charge for every person, whether or not their work changes. Odokai charges a fixed monthly fee for the platform, because the value comes from jobs that run whether or not anyone is at a desk, not from how many people log in.

In practice

  • A seat helps one person write or search faster. A workflow removes a job from the week.
  • Many firms keep a general assistant for writing and use Odokai for the recurring jobs.
  • Our comparison of the main assistants sets out where each is the better buy, including where we are not.

Worth a call if

  • You already pay for assistant seats and want to know what they are not doing.

Four platforms compared

Jobs it can take on

How can AI help with admin and paperwork?

Admin is where most teams feel the difference first. An agent can sort a shared inbox, pull details out of attachments, update the right records, chase what is missing and draft the replies, then hand anything that matters to a person to approve.

In practice

  • A typical set-up: every hour, read the new emails in a shared inbox, sort them by type, extract the details, and file or draft a reply.
  • Routine replies can go automatically. Anything involving money, a complaint or a client decision waits for approval.
  • Each email it touched, and what it did, is in the record.

Worth a call if

  • A shared inbox needs someone to work through it every day.
  • People rekey details from documents into another system.
Can AI write reports and summaries?

An agent can gather figures and notes from your systems, write the report in your format and have it waiting before anyone asks: the Monday brief, the board pack summary, the weekly update. Every figure links back to where it came from, so checking is quick.

In practice

  • The format comes from a skill: a short document describing how your team writes this report, with an example.
  • It runs on a schedule, such as 07:00 every Monday, and the report arrives in a conversation or a channel.
  • It is told never to state a figure the source does not contain, and to mark anything it cannot explain.

Worth a call if

  • Someone spends half a day each week assembling the same report.
  • The figures come from two or more systems.
Can AI analyse our spreadsheets?

Ask a question of your pipeline, ledger or last quarter's pack in plain English, and the answer comes back as figures, a chart and a table drawn from your own data, with the next steps. It works from exports, spreadsheets, databases and connected systems.

In practice

  • Example: "How are we tracking against Q3 target by region?" returns closed revenue, a chart against target, a table by region and which region needs attention.
  • Databases can be connected read-only, so the agent can query them without changing anything.
  • A question you ask every week can become a scheduled report or a dashboard app.

Worth a call if

  • People export the same data into a spreadsheet every week to answer the same questions.
Can AI check documents against our policy?

An agent can read every document in a batch against your policy or checklist, score each one, list what needs a person, and propose the wording to fix it. Your experts review the flags instead of reading everything from scratch.

In practice

  • Example: 214 supplier contracts checked against a new data policy in 11 minutes: 187 compliant, 27 needing changes, 6 sent to legal, each finding linked to its clause.
  • The rules live in a skill your team writes and can change, so the check follows your standard, not a generic one.
  • It is a first read. A person makes every decision that matters.

Worth a call if

  • A policy change means re-reading a large set of documents.
  • Reviews are done to a checklist that a capable junior could follow.
Can AI write proposals and tenders?

An agent can turn call notes and a tender's requirements into a first draft in your format, borrowing from past proposals and leaving open questions marked as assumptions. Your best person reviews and finishes it rather than starting from a blank page.

In practice

  • It can track every open tender in one place: deadline, value, how many requirements each draft meets, and what needs you.
  • It is told never to send anything. A person always sends the bid.
  • Past proposals become knowledge it can search and cite.

Worth a call if

  • Your team writes several proposals or tender responses a month.
  • The first draft takes hours and mostly reuses earlier work.
How can AI help with customer service?

An agent can triage every incoming ticket or enquiry, set its type and urgency, and draft a reply in your tone, so your team checks and sends rather than starting from scratch. Urgent or sensitive cases go straight to a person.

In practice

  • A typical run: every hour from 08:00 to 18:00, read the new tickets, classify them, draft replies, log them.
  • Simple, low-risk answers can be sent automatically once you trust them. Everything else waits for approval.
  • A chat assistant on your own website can answer from your published help content, styled to match your site.

Worth a call if

  • You handle more than a few dozen enquiries a day.
  • Most replies follow a handful of patterns.
How can AI help my sales team?

Each morning an agent can research yesterday's new leads, score them against your ideal customer and write talking points for the best, so your team calls the right people first with something useful to say.

In practice

  • It reads your CRM, researches each company on the web, scores it from 1 to 100, and explains the score.
  • Talking points come from the company's own news and your positioning, written in your words.
  • It can also keep a one-page weekly brief on what each competitor changed, with sources.

Worth a call if

  • Leads arrive faster than your team can research them.
  • Your best salespeople spend their mornings on preparation.
How can AI help with finance and invoices?

An agent can read every incoming invoice, pull out the lines, tax and totals, check the sums, match each one to its purchase order and hold back anything that does not fit. Finance approves the payment run instead of keying everything in.

In practice

  • A typical run: every weekday at 09:00, collect the new invoices from the accounts inbox and check them one by one.
  • Purchase orders and supplier records are read-only to the agent. It cannot pay anything.
  • Mismatches arrive with the reason: wrong total, missing order, duplicate, unexpected supplier.

Worth a call if

  • Your team processes more than a few dozen invoices a week.
  • Mismatches and duplicates are found late, or not at all.
How can AI help marketing?

Agents can produce first drafts of landing pages, campaign copy, product imagery and social assets to your brand guidelines, so your team picks and edits rather than briefing an agency for every change. Nothing is published until a person approves it.

In practice

  • Your brand guidelines and tone become a skill every agent follows.
  • A landing page can be built and previewed beside the conversation, then published to a link.
  • You can see how often anything you share is opened, without cookies or a tracker.

Worth a call if

  • Small creative requests queue for weeks or go to an agency.
How can AI help operations?

An agent can check the numbers that matter through the day, compare them with their usual range, and tell the right channel what moved and why, so problems are seen in hours rather than at the month-end review.

In practice

  • A typical run: hourly from 07:00 to 19:00, query the operations database read-only and post only when something moves.
  • A quiet hour produces no message, so people do not learn to ignore it.
  • It can wait for a condition, such as a delivery arriving or a figure crossing a line, and then carry on.

Worth a call if

  • Problems are usually noticed days after the data showed them.
Can AI do research for us?

An agent can answer a research question with a sourced, structured brief, or track what each competitor changed this week and rate how much it matters. Every claim comes with its source, so your team checks the reasoning rather than redoing the search.

In practice

  • A two-day background question typically becomes about an hour, most of it spent by a person reading the result.
  • Weekly tracking runs on a schedule and writes one page, with the threat of each change rated low, medium or high.

Worth a call if

  • Research is done repeatedly on the same kinds of question.
How can AI help HR?

Load your handbooks, policies and procedures into a knowledge base, and anyone can ask a question and get an answer with the source quoted, without interrupting someone senior. An agent can also assemble onboarding packs and chase the steps for each new starter.

In practice

  • Answers cite the exact document and section they came from.
  • Access follows your permissions: people only get answers from documents they are allowed to see.
  • When a policy changes, update the document and every answer changes with it.

Worth a call if

  • The same internal questions reach the same few people every week.

By sector

How can AI help an accountancy firm?

Practices use agents to draft each client's monthly management accounts commentary from the figures and last month's words, ready for an accountant to review and sign off, and to take on the routine checks around it. Client files stay in your environment.

In practice

  • A typical run: on the 5th of each month at 07:00, one step per client, with anything the figures cannot explain marked for a person.
  • It is told never to state a figure the source does not contain.
  • Your house style for commentary becomes a skill, so the drafts read like your team wrote them.

Worth a call if

  • You produce monthly commentary for more than a handful of clients.
  • Your busiest season is coming and you would rather not hire temporary staff.
How can AI help a law firm?

Law firms use agents for a first read of every contract against the firm's playbook: the risky clauses listed and scored, the key dates and figures pulled out, and a summary of the top concerns. A lawyer decides what to negotiate.

In practice

  • The playbook is a skill your lawyers write and keep up to date.
  • Client files stay in your environment, and personal data can be redacted before any model sees it.
  • Every output is labelled as a first read, not legal advice, and every step is recorded.

Worth a call if

  • Your lawyers review similar contracts against the same playbook.
  • Clients push back on time spent on routine review.
How can AI help an insurance broker?

Brokers use agents to have each week's renewal packs ready before anyone arrives: the policy details pulled together, the changes spotted, and the renewals that need a conversation flagged. A broker reads them and makes the calls.

In practice

  • A typical run: every Monday at 07:00, reading the policy database read-only.
  • It is told never to contact a client. People make every call.
  • The same pattern works for claims triage: sort, extract the details, and route to the right handler.

Worth a call if

  • Renewals are prepared by hand from several systems.
  • Renewals have been missed or reached late.
How can AI help a consultancy?

Professional services firms use agents to turn a scoping call into a first proposal in the firm's format, and to draft the Friday update for every client from what moved in the week. Senior people spend their time on advice rather than assembly.

In practice

  • Proposal drafts reuse past work and leave anything unresolved as a marked assumption.
  • Client updates follow a fixed shape: delivered, in progress, next week, and what we need from you. A quiet week reads as a quiet week.
  • Nothing is sent without a person.

Worth a call if

  • Your consultants write the same kinds of document for every client.
  • Weekly client reporting eats Friday afternoons.
Can regulated firms use AI?

Odokai has been live for over a year and is already used for regulated work, where outputs go to external parties and are checked. The controls a regulated firm needs are part of the product: approvals, redaction, data rules by classification, and a full audit trail.

In practice

  • Run it in your own AWS account if your data cannot leave your environment.
  • Keep certain classes of data on approved models only, or block them from AI entirely.
  • Audit records can be exported for your compliance team or an auditor.
  • We are open about certification: no SOC 2 or ISO 27001 yet, with an ISO 27001 plan in progress.

Worth a call if

  • Compliance has blocked general AI tools and you need a route that satisfies them.

Probably not for you if

Your oversight committee requires a vendor with SOC 2 or ISO 27001 today.

Odokai for compliance teams

Our sector isn't listed, can AI still help?

The job matters more than the sector. Agents do the same kinds of work everywhere: reading documents, checking them against rules, pulling figures together and drafting. If your team spends hours a week on that, the same approach applies to a charity, a recruiter, a property firm or a manufacturer.

In practice

  • Recruitment: screen CVs against a role's must-haves and draft the shortlist note for a consultant to check.
  • Property: pull dates and terms out of leases and keep a tracker of renewals and break clauses.
  • Manufacturing and logistics: check supplier documents against orders and report exceptions through the day.
  • Charities: draft grant applications and impact reports from past bids and programme data.

Worth a call if

  • You can name one job your team repeats every week.

How it works

What is an AI agent?

An agent is software that does a job end to end: it reads, works things out, acts in your systems and reports back, within the permissions you give it. In Odokai each agent has a name, instructions, the tools and data it may use, and rules for what it must ask first.

In practice

  • Instructions say how the job is done. Skills hold your team's know-how, such as a house style or a checklist.
  • Each agent shows what it does on its own, what it asks you first and what it never does.
  • Agents can work together: one gathers, another drafts, a third checks.

Worth a call if

  • You want to see one work on your own material.

The eight building blocks

How is this different from a chatbot?

An assistant helps one person in one conversation. Odokai runs agents that do whole jobs on a schedule, on your own data, with approvals and an audit trail, in our cloud or yours, for a fixed monthly fee. Many firms use both.

In practice

  • Assistants are strongest for writing, searching and thinking aloud.
  • Odokai is strongest where a job recurs, touches several systems and needs a person's sign-off.
  • What your agents learn belongs to the business, not to one person's account, so it stays when people leave.

Worth a call if

  • You have assistant licences but the recurring work has not changed.

Four platforms compared

Can AI do things without approval?

You decide what an agent may do on its own, what it must ask first and what it may never do. Anything that leaves the business, moves money or changes a record can wait for a person's yes, with the context needed to decide.

In practice

  • Approvals appear on the Home screen and can be given from a phone. The run carries on by itself once approved.
  • High-risk actions can require two people to approve.
  • Approvals can have a deadline, so nothing waits silently.
  • Agents cannot approve their own requests.

Worth a call if

  • Your main worry is AI doing something nobody agreed to.
Can we trust what AI produces?

Agents work from your documents and approved sources, and cite where each answer came from. They are told not to state anything their sources do not support, and to mark what they cannot explain. Anything that matters is checked by a person before it goes anywhere.

In practice

  • Every output shows its working: the documents read, the steps taken, the checks made.
  • You can test an agent before relying on it: write down what it should and should not do, and run the checks whenever it changes.
  • Checks are easiest where the right answer is clear, so we start with jobs like that.

Worth a call if

  • You need to show someone how a result was reached.
Can AI use our own documents?

Your contracts, policies, procedures and past work go into knowledge bases agents can search. Answers quote the document and section they came from. The models are not trained on your data; they read it when a task needs it.

In practice

  • Knowledge belongs to the organisation, not to one person, so it stays when people leave.
  • People only get answers from documents they are allowed to see.
  • Documents can stay where they are and be connected, or be copied into a workspace.

Worth a call if

  • People spend time finding answers that are already written down somewhere.
Can AI work with our existing systems?

Odokai has over 600 connectors, including the common email, calendar, document, CRM, accounting, project and support tools. Agents work where your data already lives, with the permissions you set. For anything else there is an API, webhooks and MCP.

In practice

  • Connections can be read-only, so an agent can look but not change anything.
  • A system with no connector can usually be reached through its API or a database connection.
  • We confirm your exact set-up on the discovery call.

Worth a call if

  • Your work spans several systems that do not talk to each other.
Can AI run jobs automatically on a schedule?

A workflow is a set of steps an agent follows: it can branch, loop over every item, pause for a person, and carry on once approved. It can start on a schedule, when an email or event arrives, or when someone asks. It runs whether or not anyone is signed in.

In practice

  • You can watch each step live, or come back to the result.
  • Home shows what ran while you were away and what needs you.
  • Long jobs survive our deployments: they pick up where they left off.
  • A test run shows what a workflow would do without writing anything.

Worth a call if

  • A job happens at the same time every day, week or month.
Can AI build internal tools for us?

Describe the tool in a sentence and an agent builds it beside the conversation, on your own data: a tracker, a client portal, a dashboard, a form. Change it the same way you asked for it. Share it with your team or publish it to a link.

In practice

  • Apps run on the same platform, under the same permissions and audit trail as everything else.
  • Each app's data stays in your instance, and nothing a visitor does reaches your live data without your say.
  • Small subscriptions for single-purpose tools can often be replaced by an app built on your own data.

Worth a call if

  • You pay for several small tools that each do one thing.
  • Spreadsheets are being used as databases.
Which AI models does Odokai use?

Your choice. Odokai works with the leading model providers and any provider with a compatible API, and sends each step to the model that suits it. Classifiers handle sorting and checking, so the expensive models only do the hard reasoning.

In practice

  • Upgrades reach your instance as new models ship. You can switch per agent or per task.
  • Certain classes of data can be kept to approved models, or to models running in your environment.
  • On your own instance you can add and remove providers yourself.

Worth a call if

  • You are committed to one provider and want options.
Can we add an AI assistant to our website?

You can put an agent on your own website with one line of code: a launcher or an inline panel, styled to match your site, answering from content you choose. Visitors never see how it works behind the scenes.

In practice

  • Its colours, wording and position are set from the app, and your site can override them.
  • It can hand a conversation to your team when it should.

Worth a call if

  • Your website gets the same questions every day.
Is there an API?

Odokai is the platform layer your engineers would otherwise build: model routing, approvals, audit, connectors and deployment. Your team writes the business logic. It exposes an API, webhooks and an MCP server, and the agents can work with git and run in a sandbox.

In practice

  • Deploy shared, in an AWS account we manage, or in your own account with your own keys.
  • Traces can go to your own OpenTelemetry collector on your own instance.
  • Apps and skills are files you can export and version.

Worth a call if

  • Your engineers have been asked to build an internal AI platform.

Odokai for CTOs

Data, security and compliance

Is AI safe to use with confidential client data?

Your documents, search index, audit trail and workflow logic stay inside your environment. Each call to an AI model can be set up as zero-retention, no-training and in-region. Access is role-based, permissions deny by default, and every action is logged.

In practice

  • Personal data can be redacted before a model sees it. This is on by default.
  • You can hold the contract with the model provider directly, under your own terms.
  • Steps that only sort or check run on small classifiers, so less of your text is sent anywhere.

Worth a call if

  • You handle client data that cannot go into public AI tools.

Security controls compared

Is using AI GDPR compliant?

Odokai gives you what you need to use AI lawfully on personal data: data kept in the UK or your own account, redaction before any model sees it, rules for each class of data, role-based access and an audit trail. You remain the controller; we help you document it.

In practice

  • Data processing terms and the list of sub-processors are agreed in the contract.
  • We can support a data protection impact assessment with how the system handles each data type.
  • If you have EU-facing work, the EU AI Act may apply. Our free governance health check covers it.

Worth a call if

  • Your data protection officer needs answers before AI can be used.

AI governance health check

Can AI run in our own cloud?

Three options: our shared platform in AWS London; a dedicated instance in an AWS account we manage; or a dedicated instance inside your own AWS account, with its own network, database and encryption keys. Everything you build moves with you between them.

In practice

  • Shared starts at £500 a month ex VAT and is live the same day.
  • Dedicated starts at £1,000 a month ex VAT, run by us or by your team.
  • In your own account you choose your own model providers and pay them directly.
  • AWS is the only cloud at present.

Worth a call if

  • Your data must stay in your own environment.

Probably not for you if

You need to run on a cloud other than AWS today.

Plans compared

Is Odokai ISO 27001 certified?

We are a small UK company, live for over a year. We do not hold SOC 2 or ISO 27001 yet; an ISO 27001 plan is in progress. Run inside your own AWS account, the infrastructure sits under your certifications and AWS's.

In practice

  • We complete security questionnaires and walk your team through the architecture.
  • Source code escrow is available, released to you if we stop trading.
  • Your data and everything you build can be exported in documented formats at any time.

Worth a call if

  • Your procurement process can work with a smaller vendor given the right protections.

Probably not for you if

Your board requires a certified vendor of large scale today.

Is there an audit trail?

Every step is logged: what each agent read, what it did, which model it used, and who approved what. Records are classified, encrypted and kept for around seven years, and can be exported as CSV, signed JSON or PDF for an auditor.

In practice

  • Each output links back to the documents it came from.
  • Changes are attributed to the person or the agent that made them, and the person the agent worked for.
  • Administrators can review the whole log; everyone else sees their own work.

Worth a call if

  • An auditor or regulator will ask how a result was produced.
Does it support single sign-on?

People sign in with your existing identity provider through SAML single sign-on, and accounts can be created and removed automatically with SCIM. Permissions are role-based and deny by default, and an agent can only do what the person running it can do.

In practice

  • Multi-factor authentication can be enforced across the organisation.
  • When someone leaves, their access ends with their account; their work stays with the business.

Worth a call if

  • IT needs AI tools to follow the same access rules as everything else.
Who owns the data and the work?

Your data and everything you build are yours: documents, workflows, agents, prompts and knowledge bases. They export in documented formats on request, on any plan, whether or not you renew. Source code escrow can go in the contract if you need it.

In practice

  • Skills and apps are files, so they can be taken with you.
  • Escrow releases our source to you only if we become insolvent, stop trading or discontinue the platform.

Worth a call if

  • Being locked into a vendor is a concern for your board.

Your team

Will AI replace my staff?

It takes on tasks, not roles. Agents do the gathering, drafting and checking; your people keep the judgement, the client relationships and the sign-off. Most firms use the time to take on more work with the team they have, or to stop hiring for repetitive work.

In practice

  • The people who know the job best write the agent's instructions, so their expertise is kept.
  • Jobs that were someone's least favourite part of the week are usually the first to go.

Worth a call if

  • You are growing and want to avoid hiring for repetitive work.
Do we need a technical team to use AI?

No. Agents are set up in plain English: what the job is, what it may use and what it must ask first. We build the first workflow with you and train the people who will run it. By the third workflow most teams build their own.

In practice

  • The best person to own an agent is whoever does the job today, not IT.
  • IT is involved for access to systems and sign-in, usually a few hours.
  • We review how it is going monthly, quarterly and annually.

Worth a call if

  • You have no AI or engineering team and want the work done anyway.
How do I get my team to use AI?

Teams adopt what saves them time on work they already do. So we start with one real job, built around how your people work now, and train them as part of the build. Once they see an agent take a chore off their desk, the next idea usually comes from them.

In practice

  • Start in one team, not across the firm.
  • Show the result in the tools they already use: their inbox, their channel, their Home screen.
  • Make one person the owner of each agent.

Worth a call if

  • Earlier AI tools were bought but not used.
Our staff already use AI tools, is that a problem?

If people are pasting work into personal AI accounts, the risk is already there; you just cannot see it. Odokai gives them one approved place with the leading models, under your rules, with an audit trail, so the use comes into the open.

In practice

  • Personal data is redacted before it reaches a model, and some data can be kept away from AI entirely.
  • Everything is logged against the person who did it.
  • Our free governance health check shows where you stand.

Worth a call if

  • You suspect client data is going into personal AI accounts.

AI governance health check

What support do we get after going live?

Once you are live we review the work on a regular rhythm: monthly for what the agents did and what to fix, quarterly for results against what was agreed and the next jobs, and annually for governance and the year ahead. Support is part of the monthly fee.

In practice

  • Model upgrades reach your instance as they ship, at no extra cost.
  • Senior AI expertise is available beyond the platform, as fractional leadership, embedded delivery or a short advisory piece.

Worth a call if

  • You want a partner after launch, not just software.

Services

Buying and switching

Should we build our own AI platform?

You can build it, and some firms should. Expect to rebuild a model registry, approval chains, policy controls, redaction, observability and deployment tooling before the first business workflow ships: typically seven to eleven months of engineering. Odokai is that layer already built.

In practice

  • Your engineers keep writing the business logic, on a platform you can run in your own account.
  • The workflows move with you between shared, managed and self-hosted.

Worth a call if

  • Your team has been asked to build an internal AI platform and has a backlog already.

Probably not for you if

AI infrastructure is your product, not a tool for running your business.

Odokai for CTOs

Can you just give us advice?

Yes. The same team offers advice on any stack: fractional AI leadership on a retainer, engineers embedded with your team on a day rate, or a short independent advisory piece with a written report you keep whether or not you use Odokai.

In practice

  • Fractional leadership is typically two to four days a month.
  • Advisory pieces usually take one to two weeks: opportunities, vendor choice and governance design.

Worth a call if

  • You need an AI strategy or a vendor decision before choosing tools.

Services

Can we move from our current AI tool?

You rarely need to switch everything. Most firms keep their assistant for writing and add Odokai for the recurring jobs. Where we do replace a tool, we move the work job by job, so nothing stops while it happens.

In practice

  • Prompts and instructions you already rely on can become skills.
  • Connections are set up once and reused by every workflow.

Worth a call if

  • Your current tools cover writing but not the work that recurs.
Who else uses Odokai?

Odokai has been live for over a year, including regulated work where outputs go to external parties and are checked. The founders built dotdigital, which listed on the London Stock Exchange. The best evidence is your own: the discovery call shows an agent on your documents.

In practice

  • In production, the time saved has been over two hours on each case handled.
  • We can arrange a conversation with a customer where it fits your situation.

Worth a call if

  • You need more than a demo to take this to your board.

Company

Not sure it applies to you?

Bring one job to a call.

A free 45-minute discovery call. Bring a job your team repeats and a few real documents. We build an agent for it while you watch, and say whether it is worth automating.

Book a discovery call See ten jobs agents run