What the first 90 days after go-live should cover

23 October 2026

The project ends where the value starts

In a plan that ends at go-live, the budget, the steering committee and the supplier's team are all organised around that date. When it arrives, the project team celebrates, the supplier rolls off and IT inherits a live system.

For an AI platform, that plan stops too early. Before go-live, AI agents have only seen test cases. After go-live, they meet the real queue: incomplete forms, unusual requests, a supplier who replies in a new format, a month-end that runs late. That is when you learn what the platform can carry.

So here is our view: go-live is the start of implementation, not the end of it. The first 90 days should be planned, staffed and measured as carefully as the build.

For a CIO or Head of IT, that changes the question to ask before signing. "When do we go live?" matters less than "What happens in the 90 days after?"

Days 1 to 30: does it hold under real work?

The first month is about stability, in business terms as well as technical ones.

Every exception has an owner. When an agent cannot finish a task, a named person sees it quickly and knows it is theirs.
The rules meet reality. What agents may do alone, and what needs a person's approval, was written before go-live. Now it meets cases nobody predicted. Expect to tighten some rules, and to relax none yet.
Problems are logged, not absorbed. Teams under pressure fix things quietly by hand. Each quiet fix shows something the platform still has to learn, and it is lost if nobody records it.

The test at day 30: are exceptions falling, and does everyone know where they go?

Days 31 to 60: is it inside the work?

A platform can be live and still unused. The second month is about adoption, and the evidence is in behaviour, not in a survey.

Watch for the side routes. A spreadsheet that comes back. An approval that moves to WhatsApp. A team that asks a colleague instead of the platform because it is quicker. Each one shows where the platform is not yet inside the work. Treat it as design information, not a training problem.

This is also when knowledge starts to drift. A price changes, a process is updated, a new supplier is added. If the platform's knowledge is not kept current as part of normal operations, its answers age quietly, and trust goes with them.

The test at day 60: is the platform the normal route for the work it was built for?

Days 61 to 90: is it creating the value we agreed?

By the third month there is enough real activity to measure. Use the measure agreed before the build, against the baseline taken then: time returned to a team, errors avoided, revenue protected or software spend consolidated, counted in rand where it is money.

If nobody took a baseline, this is when you find out why it mattered. Without one, early value turns into an argument about impressions.

Day 90 is also the first fair point to review autonomy. Where agents have handled a task reliably, with a clear record, a person can decide to let them do more of it alone. Where they have not, the rules stay as they are. Autonomy is earned against evidence, never granted on a schedule.

The test at day 90: can you show the value in business terms, and decide the next step from evidence?

Who runs it through all of this?

None of this happens by itself. The first 90 days need named people: someone accountable for how the platform performs, an owner for each exception queue, and a business owner who decides when the rules change.

If the supplier's team leaves at go-live, those roles fall to whoever is nearest, on top of everything else. Running the platform is part of implementation, not an optional extra after it.

Five questions to ask before you sign

Who runs the platform the day after go-live, and for how long?
How are exceptions seen, owned and fed back into the platform?
How will we know whether people use it inside their daily work?
Which business measure, and which baseline, will we judge it by at day 90?
Who decides when agents may do more on their own, and on what evidence?

If a supplier's plan ends at go-live, the hardest part of implementation is still unplanned.

Build, implement and run

At Futuvara, we build, implement and run AI platforms where AI agents do defined work under rules your business sets.

If you are planning an AI platform, or living with one that went live and then stalled, we would like to hear about it.

Start an Agentic conversation →

We are what we sell.

We run Futuvara on our own AI platform. Every face below is one of our AI agents.

Scroll the rows sideways to see every agent.

AI agent
Kat
Chief Operating Officer: coordinates the business units
Reads each business unit's reports against the company's goals, finds where one unit is waiting on another, and sends each fix to its owner.
AI agent
Tom
Chief Growth Officer: brand, marketing and sales
Runs Growth as one system, from brand to signed customer, against approved targets. Finds where the sales funnel leaks and sends each fix to whoever owns it.
AI agent
Lin
Chief Delivery Officer: project health and risk
Runs Delivery as one portfolio. Reads every project's status beside demand and budget, finds where the portfolio drifts, and sends each correction to its owner.
AI agent
Tamara
Financial Officer: reports on the company's finances
Reconciles the ledger with the bank, produces the statements and cash report, and flags every variance the controls name. It reports; a person decides.
AI agent
Anja Johansen
Management Officer: runs planning and performance reviews
Drafts the company's performance reviews and quarterly plans, so leadership decides from evidence. Every figure traces to a record.
AI agent
Freya Lindqvist
Leads brand and creative direction
Directs Futuvara's brand, look, pictures and motion, and checks every piece a buyer will see before a person decides.
AI agent
Amara Okafor
Defines target customers and positioning
Recommends who Futuvara sells to and what it says: the ideal customer profile, segments, target accounts and messages.
AI agent
Lucas Brennan
Writes article and web page copy
Writes Futuvara's articles and web pages, briefs the Website agent to build them, and keeps the "Start an Agentic conversation" route working.
AI agent
Dawit Haile
Moves deals from lead to close
Moves deals from first interest to a win Delivery can start on, and keeps every deal record true. A person approves anything a prospect sees.
AI agent
Tarek Hassan
Runs demand generation and campaigns
Shapes demand campaigns and their assets, and recommends what to do with them, so Growth earns leads worth a sales call.
AI agent
Kwame Asante
Marketing operations and measurement
Sets up how Growth measures itself: tracking, customer-record categories, lead routing standards and baselines, each with a source and an owner.
AI agent
Petra Svobodová
Drafts the public website for approval
Builds Futuvara's public website: pages, code, pictures and motion. It writes drafts; a person decides what goes live.
AI agent
Maya
Plans screens and how they work
Plans the screens of a project, from menus and views to an app or a web page, with four specialist agents. Delivers only after a person says yes.
AI agent
Nadia Haddad
Designs menus, labels and site structure
Plans the structure before any screen is drawn: sitemap, menus, labels and content model, with every item in the one place users would guess.
AI agent
Kenji Watanabe
Chooses page layouts, grids and spacing
Chooses the layout for each view: the paradigm, the page skeleton, and the grid and spacing where a project has none.
AI agent
Lucía Romero
Checks what people see first on a screen
Critiques a drafted view's visual hierarchy: what is seen first, where the eye goes next, and what carries too much weight. Every finding comes with its fix.
AI agent
Samuel Lindqvist
Writes specs for new components
Specifies components a project does not have yet: anatomy, variants, every state, behaviour and accessibility. First it checks nothing existing does the job.
AI agent
Ludwig
Chairs meetings and records decisions
Chairs rooms and calls where specialist agents and a person work together. Picks who speaks next, records what is settled, and names what happens now.
AI agent
Jackson
Business Architect: turns goals into workflows
Helps a business unit's officer turn an approved mandate into workflows, schedules, skills, agents and templates that run, each traced to the measure it moves.

AI agent
Deniz Demir
Solutions Architect: turns client goals into designs
Turns a partner's business problem into a solution design a project manager can start on the next day, naming the assumption most likely to break it.
AI agent
Diego Hernandez
Delivery Project Manager: runs the schedule
Compares each Delivery project's plan with what happened, logs the gap and raises it to whoever can act, so a slip is seen in week one.
AI agent
Mariana Quispe
Frames and refines early ideas
Turns a raw partner opportunity into an Opportunity Brief: the problem in the partner's words, the value it could unlock and the idea to test first.
AI agent
Mateo Silva
Shapes early ideas into delivery direction
Turns a framed opportunity into a platform direction: what it is for, the first use-cases that show it working, and why it is worth building.
AI agent
Rafael Oliveira
Shapes agreements and maps milestones
Checks that what Delivery will build and what the partner will sign describe the same thing, and flags billing implications for Finance.
AI agent
Sofia Marchetti
Shapes requirements and records design decisions
Turns a platform direction and its requirements into a design that can be locked: scope boundaries, testable acceptance criteria, milestones and recorded decisions.
AI agent
Njeri Mwangi
Turns build packs into draft backlog items
Turns approved requirements and build packs into backlog items a developer can finish in one sprint. Its items are drafts until someone pushes them.
AI agent
Robin Lindqvist
Tracks sign-offs and billing readiness
Turns a finished, signed-off milestone into a billing package Finance can act on: what was delivered, who signed and when, and what is still open.
AI agent
Dawit Bekele
Checks a build is ready to release
Checks that what Delivery built does what the partner agreed, criterion by criterion against its evidence, before anyone signs, ships or bills.
AI agent
Zofia Kowalski
Writes contact reports from client meetings
Turns the record of a client meeting or call into a draft contact report: requirements, decisions, actions, risks and open questions, each tied to a quote.
AI agent
Isabela Rossi
Prepares the handover of a running platform
Packages an accepted platform for running use: a handover pack that tells the running owner what they own, what is still open and who acts next.
AI agent
Min-jun Choi
Sorts client feedback and drafts the replies
Reads the feedback partners and users send Delivery, removes personal details, sorts each item for whoever decides, and makes sure the sender hears back.
AI agent
Rami Haddad
Watches capacity and routes the workload
Paces Delivery's work so agents never produce faster than people can approve or build. Compares what comes in with what is cleared, and recommends the pace.
AI agent
Kathy
Turns client requests into requirement records
Turns what clients ask for, in emails, meetings and screenshots, into requirement records: one outcome each, who asked, and the evidence. Never how to build it.
AI agent
Priya Raman
Turns requirements into build packs
Turns requirement records into build packs: one brief per shippable slice that a developer or coding agent can build without asking anyone.
AI agent
Sione Taufa
Opens or updates a new client's project
Reads a client kickoff and decides whether to open a new Delivery project home, update the existing one, or hand the choice to a person.
AI agent
Clara
Identifies companies and people, and updates their records
Keeps Futuvara's company and participant records true. Finds the one record each piece of evidence belongs to, and replaces placeholders with cited facts.
AI agent
Quinn
Pulls facts out of incoming information
Turns each incoming document into knowledge Futuvara can trust: organises its facts, cites every one, and files each only in the place the platform assigns it.

AI agent
Dilnoza Karimova
Sends each incoming record the right way
Sorts each incoming record into one route, or none. A route starts real work in another business unit, so a record must earn it.
AI agent
Dieter
Keeps the knowledge base well organised
Keeps the knowledge base true to itself: one home per topic, every link working, and every change followed to the documents that depend on it.
AI agent
Keiko Watanabe
Checks which document owns each topic
Before anything is written to the knowledge base, rules on one question: does an existing document already own this topic?
AI agent
Farhan Hossain
Decides what else changes after an edit
When a document changes, works out what else must change, and takes the cases where more than one answer could be right.
AI agent
Arief Wibisono
Checks claims against their sources
Checks whether each claim is true of the source it cites, and returns one verdict per claim.
AI agent
Catalina Muñoz
Drafts new documents, citing every fact
Writes knowledge-base documents in their standard shape, with a source cited for every factual sentence.
AI agent
Luka Petrović
Turns agents' recent mistakes into lessons
Turns the evidence of another agent's recent work into at most one true lesson. A person's own words decide what the lesson is.
AI agent
Ines Carvalho
Turns requests for new tools into specs
Turns a business unit's need for a new tool or integration into one request the platform's build queue can act on. Searches for what already exists first.
AI agent
Will
Builds and tests workflows
Turns what a person wants, such as "chase invoices two weeks late", into a workflow on the Futuvara platform, then proves it runs.
AI agent
Noa Hernandez
Looks after the shared skills library
Turns how people really work into skills any agent can follow: a name, when to use it, ordered steps and what can go wrong. Tests each one before it is shared.
AI agent
Hugo
Drafts email replies from company knowledge
Drafts email replies in the operator's voice, after reading what the knowledge base holds about the sender. The operator edits and sends them.
AI agent
Minh Tran
Writes the code for build packs
Builds one build pack at a time, on its own branch, until every acceptance criterion is met and proven.
AI agent
Joost van Dijk
Resolves clashes when code is merged
Settles clashes when code is merged, so both changes still do what they were built to do. When it cannot keep both, it stops and says so.
AI agent
Malia Tuilagi
Helps create reusable document templates
Designs reusable document templates: the sections, their order, and guidance telling the next author what goes in each.
AI agent
Eleni Papadopoulou
Builds dashboards in conversation
Builds dashboards so operators can see a number and act on it. Every tile says where its number comes from.
AI agent
Mihai Popescu
Decides when recurring work runs
Decides when the platform's recurring work runs, and turns that rhythm into schedules that start the right workflows.
AI agent
Siriporn Chaiyaporn
Plans and rearranges the working day
Shapes an operator's working day on Google Calendar so the best hours go to focused work. Never moves someone else's meeting to make room.
AI agent
Aiman Rahim
Manages the platform's cloud storage
Looks after the platform's Google Cloud Storage: what is stored where, and the changes people ask for. Every overwrite or delete waits for a yes.

What the first 90 days after go-live should cover

The project ends where the value starts

Days 1 to 30: does it hold under real work?

Days 31 to 60: is it inside the work?

Days 61 to 90: is it creating the value we agreed?

Who runs it through all of this?

Five questions to ask before you sign

Build, implement and run

Keep reading

Why AI pilots stall, and what gets them into production

How an AI agent earns the next step

We are what we sell.