AI Agents and Assistants
What is an AI agent? A plain-English guide for business owners
What an AI agent is, how it works, where it helps a small business, the risks, and a 7-point check to see if you are ready for one.
AI Features & Bots
AI Proof-of-Concept Sprint
Test one AI idea on your own data with a working prototype, measured results and a clear go or no-go.
An AI idea can look promising in a demo and still fail on your real documents. AI Proof-of-Concept Sprint finds out before you pay for a full build. We prototype one workflow, test it on your records against cases you agree, and report accuracy, speed and cost per run. You get a go or no-go call and a fixed build quote, with the sprint fee credited if you proceed.
- A working prototype for one workflow, built on your data
- Accuracy, speed and cost per run measured on 30 cases
- Go or no-go call, fixed build quote and fee credit
$4,925
10–20 working days
What it is
AI Proof-of-Concept Sprint is a short, fixed-scope test of one AI use case on your data. We build a working prototype for 1 workflow with a simple web interface, test it on up to 100 of your records against 30 agreed cases, and report accuracy, speed and cost per run.
Why businesses buy it
Full AI builds are a real commitment, and the biggest risk is that the idea does not work on your data. Messy documents, edge cases and unclear rules show up only when you test. This sprint tests early. You define what a correct answer is, and we score the prototype against it on 30 agreed cases. The results report shows real failures as well as successes. If the numbers are good, you get a fixed quote for the build, and the sprint fee is credited if you order within 30 days. If not, you have avoided paying for a build that would not work.
Problems this solves
You do not know whether AI will work on your data
A working prototype for one workflow is tested on up to 100 of your own records. You see how it handles real documents, not a polished demo.
Nobody can say what accuracy or running costs to expect
The results report measures accuracy, speed and cost per run across 30 agreed cases. It includes failure examples, so you see exactly where the prototype struggles.
A full build feels too risky without evidence
You get a go or no-go recommendation and a fixed quote and timeline for the build. The sprint fee is credited against a build ordered within 30 days.
A good fit if
You have one AI idea, such as reading documents or answering questions, and want proof it works first.
You can share up to 100 representative records or documents, anonymized where needed.
One person can decide whether to go ahead, and can say what counts as a correct answer.
Your idea is not covered by a ready-made product in the store, or you doubt your data is good enough.
Not the right fit if
You still need to work out requirements and screens before any code. See Discovery Sprint.
You already know the feature works and want it built into your product. See AI Feature Integration.
You want an internal tool built for your team straight away. See AI Internal Tool Builder.
You need ongoing testing of an AI system already in production. See AI Evaluation System.
How it works
- A written kick-off questionnaire sets the workflow, the data source and what counts as a correct answer, with follow-up questions by email.
- You share up to 100 representative records and read-only access to one data source, and we agree a test set of 30 cases.
- We build a working prototype for the workflow with a simple web interface, using models such as OpenAI or Google Gemini.
- The prototype runs on all 30 cases, and each answer is scored against your definition of correct.
- You receive a results report with accuracy, speed, cost per run and failure examples, plus a go or no-go recommendation.
- We share the working prototype and results report, with a fixed quote and timeline for the full build.
What you provide within 72 hours
Your portal lists these the moment you order. Delivery starts the working day after everything is in and your access is checked.
- Up to 100 representative records or documents, anonymized where needed, within three working days.
- Read-only access to one data source, by invite or a key you control.
- One decision-maker who answers the kick-off questionnaire and decides go or no-go.
- Your definition of a correct answer, which we turn into the 30 test cases together.
The 72-hour rule. Send everything on your checklist, including access, within 72 hours of ordering so your build starts on time.
Timeline
Start
Order placed. Sample records, data access and your decision-maker's details are due within 72 hours.
Week 1
Kick-off questionnaire answered, the workflow agreed and the definition of a correct answer written down.
Week 1
Sample data checked and the 30 test cases agreed with you.
Weeks 1–2
Prototype built for the workflow with a simple web interface.
Week 2
All 30 cases run and scored, the results report written, the prototype shared and the build quote sent.
After delivery
14 days of follow-up questions, with the fee credit open for a build ordered within 30 days.
| Option | Delivery time | Price |
|---|---|---|
| Standard | 10–20 working days | $4,925 |
What you receive at handover
- A working prototype for one workflow with a simple web interface (for testing, not live use)
- A test set of 30 agreed cases
- A results report with accuracy, speed, cost per run and failure examples
- A go or no-go recommendation, and a fixed quote and timeline for the build
- A results report and 14 days of follow-up questions
Expected benefits and how to measure them
- Helps you decide on a build with evidence, not hope
- Can reveal data problems before they become costly
- Gives you a realistic cost per run to plan around
- Accuracy rate
- Read the share of the 30 test cases answered correctly in the results report.
- Cost per run
- Take the measured cost per run from the report and multiply it by your expected monthly volume.
- Time per task
- Compare the prototype's measured speed with how long a person takes on the same task today.
Before you start: Time how long a person takes on the task today and note how often mistakes happen. Record the monthly volume for the task.
These are things to track, not promised results. Outcomes depend on your data, your team and your market.
Running costs
Model usage up to $50 during the sprint is included. If testing uses more than that, the extra is billed at cost. There are no running costs after the sprint, because the prototype is a test, not a live system.
Ask about running costsCustom requirements
Add a note at checkout if you need something specific. If it fits within the Scope Card it is simply included. If it goes beyond it, you get a written Change Quote within two working days, and that work starts only once you accept it, with its time added to your delivery date.
Request a custom versionKeep it running
The prototype is for testing, not for live use, so there is nothing to maintain. If the results are good, the fixed quote covers the production build. The sprint fee is credited against that build when it is ordered within 30 days. Maintenance options are then set out with the build.
How this compares
| Do it yourself | Hire a traditional agency | Upmark Labs | |
|---|---|---|---|
| Cost | Trial accounts, plus weeks of your team's time | A paid discovery phase, often before any code is written | One fixed price, credited against a build ordered within 30 days |
| What you get | A demo on sample data | Slides and a proposal | A working prototype tested on your own records |
| Proof | It looked right on a few examples | Depends on the team | 30 agreed cases scored against your definition of correct |
| Running cost | Unknown until you scale | Estimated in the proposal | Measured cost per run in the results report |
| Decision | Made on gut feel | Often assumes you continue | A go or no-go recommendation and a fixed build quote |
Questions about this product
What kinds of AI use cases suit a proof-of-concept sprint?
Ones where an AI reads, sorts, extracts or answers from your own information. Common examples are pulling fields from documents, classifying requests or answering questions from a knowledge base. The sprint tests one workflow on up to 100 records. If your idea needs many systems connected at once, we may suggest a Discovery Sprint first.
What happens to the sprint fee if we go ahead with the build?
It is credited against the build. If you order the production build within 30 days of the sprint, the full sprint fee comes off the fixed build quote. If you decide not to proceed, or wait longer than 30 days, the sprint stands as a paid test, and you keep the results report.
What if the AI prototype does not work well enough?
Then the recommendation is no-go, and that is a useful result. The report shows where it failed, with examples, so you can see whether better data or a narrower task might help. You avoid paying for a build that would not work. There is no obligation to buy anything after the sprint.
How is our data handled during the proof-of-concept sprint?
Access stays narrow. You share up to 100 records, anonymized where needed, and read-only access to one data source by invite or a key you control. Records are sent to the AI model provider used for the prototype, so anonymize anything sensitive before sharing. You can remove our access when the sprint ends.
Are model usage costs included in the sprint?
Yes, up to $50 of model usage is included. If testing needs more than $50, the extra is billed at cost. Most of the spend comes from running the 30 test cases and tuning the prototype. After the sprint there are no running costs, because the prototype is not a live system.
Can the AI prototype go straight into production?
No. The prototype is built to test whether the idea works, not to run live. Production needs security hardening, proper integrations, monitoring and support, which are part of the build. The fixed quote you receive covers that work, and the 30 agreed test cases give the build a ready-made check.
How is accuracy measured in the proof-of-concept sprint?
Against your own definition of a correct answer. In the kick-off questionnaire we agree what correct means, then build 30 test cases from your records. The prototype runs every case, and each answer is scored. The report gives the accuracy rate, typical speed and cost per run, and shows examples of the cases it got wrong.
Guides on this topic
All guides Prove the AI idea before you build it
Order the AI Proof-of-Concept Sprint for a working prototype tested on your own data, honest results and a fixed quote for the build.
Works well with
AI Features & Bots
AI Slack Bot for Google Workspace
Builds a Slack bot that answers questions and runs actions for your team, connected natively to Google Workspace so it reads and…
AI Features & Bots
AI Slack Bot for Notion
Builds a Slack bot that answers questions and runs actions for your team, connected natively to Notion so it reads and updates…
AI Features & Bots
AI Slack Bot for Jira
Builds a Slack bot that answers questions and runs actions for your team, connected natively to Jira so it reads and updates…