unthinkable
Enterprise software (HR, CRM) · AI Product Engineering

Help people do their actual work, while the software does the admin.

We help you build the product that makes it possible: HR and CRM software where AI answers the questions, fills in the forms, keeps the records and follows up, with each user's own permissions and people deciding what matters.

How much of the working day goes to admin, not the actual work

70%of sales reps' time goes to tasks other than selling, such as admin and meeting preparationSalesforce, State of Sales 2024 (5,500 sales professionals): salesforce.com/news/stories/sales-ai-statistics-2024
1 in 5US payrolls contain errors that HR then has to put right, at about $291 eachEY payroll study, Dec 2022 (via HR Dive): hrdive.com
15payroll corrections HR makes by hand per pay period at the average companyEY payroll study, Dec 2022 (via HR Dive): hrdive.com
What we build

HR and CRM software we build to take the admin off people's day.

Three kinds of enterprise software we build again and again. For each: the admin it takes off people's day, what AI does (and what people still decide), and how we help you build it.

HR, payroll & performance (HRMS)

Employees get answers, leave and pay right without chasing HR, and HR spends its time on people, not re-keying spreadsheets or correcting pay runs.

BUILD AI IN

An assistant answers policy and payslip questions, fills in leave, claims and appraisal forms, and flags pay-run errors before payday, so they are not corrected after. HR approves every pay run and decision.

BUILD WITH AI

Pay, tax and statutory rules run as configurable rules in code; connected to attendance, accounting and banking systems; every change logged.

Custom CRM and Salesforce

Sales and service teams spend their time with customers, not updating records.

BUILD AI IN

Agents log calls and emails, keep account data clean, draft follow-ups and route each case to the right person before its SLA runs out. People decide what is sent.

BUILD WITH AI

Built as a custom CRM, or on your Salesforce data model; sandbox first, with least-privilege access.

AI agents inside B2B SaaS products

Your users hand the product their admin in plain words and it gets done, with their permission, while they get on with their own work.

BUILD AI IN

An in-product agent completes the setup, reports and repeat steps across your screens and APIs with the user's own permissions, and asks before anything that matters.

BUILD WITH AI

Tenant-level metering, an evaluation suite that gates every release, and run tracing, all in your repositories and handed over.

Running HR, payroll or sales for your own company? HR, payroll & CRM →

Your customers' journey

Your customers' journey, and what your product does at each step.

The same journey we follow on the HR, payroll & CRM page, seen from the product: what the customer gets at each step, and the software we build to do it.

Today: a wait before every stepWith AI at work: straight through

2 · Time & leave

Leave applied for in one message, with the balance and any missed punch shown first, so nothing is fixed after payroll.

HR, payroll & performance (HRMS) ↓

3 · Pay day

HR checks a pay run instead of correcting it, and employees get payslip questions answered without raising a ticket.

HR, payroll & performance (HRMS) ↓

4 · Performance

Managers write reviews from the year's evidence, not from memory, and spend the meeting on the conversation.

HR, payroll & performance (HRMS) ↓

5 · Customers & sales

Salespeople and service teams spend their time with customers, while records and follow-ups keep themselves up to date.

Custom CRM and Salesforce ↓

What changes for each role (text version)

Joining

  • What if a new joiner arrived with everything already sorted? Today: Only 12% of employees strongly agree their organisation does a great job of onboarding new employees. With an agent: New joiners see every document they need at once, hear straight away what's missing, and arrive to a working laptop and access. Rule, not model: Which documents each role and country needs, and when a bank account is accepted, are joining rules in code. (Gallup)

Time & leave

  • What if a leave request never turned into a payroll correction? Today: Per 1,000 workers a year, vacation, PTO and sick-time requests cause about 721 payroll errors, and time and attendance about 1,139. With an agent: Employees see their balance and any missing punch before they apply, and managers approve with the facts in front of them. Rule, not model: Balances, holidays, carry-forward and the payroll cut-off follow your leave and attendance rules in code. (EY payroll survey, 2022, as reported by HRD)

Pay day

  • What if every pay run was checked before it was approved, not corrected after? Today: The average organisation makes 15 payroll corrections every pay period. With an agent: HR sees every change since last month and every figure that looks unusual, with the record behind it, before approving the run. Rule, not model: Pay, tax and statutory deductions are calculated by your payroll rules in code; the agent only checks and explains. (EY payroll survey, 2022, as reported by HRD)

Performance

  • What if every review started from the whole year's evidence? Today: Only 14% of employees strongly agree that the performance reviews they receive inspire them to improve. With an agent: Managers start from a summary of goals, check-ins, work and feedback with links to the source, and write the rating themselves. Rule, not model: What a manager can see, and the review cycle's dates and steps, follow your HR roles and performance rules in code. (Gallup, 2017)

Customers & sales

  • What if the CRM updated itself after every call? Today: Sales reps spend 70% of their time on tasks other than selling, such as admin and meeting preparation. With an agent: Calls and emails land on the right account, follow-ups are drafted and every change to a deal is the salesperson's to confirm. Rule, not model: Prices come from your price list, and which fields an agent may change follow your CRM permissions. (Salesforce, State of Sales 2024)

Leaving

  • What if nobody kept access after their last day? Today: 83% of people surveyed said they still had access to the digital assets of their previous employer. With an agent: Every exit runs from one checklist: handover, assets, access removal on the last day and a final settlement prepared for HR to check. Rule, not model: Notice periods, leave payout, gratuity and access removal follow your policies and payroll rules in code. (Beyond Identity, 2022)

Try it with your own case →

Why products stall

Why HR and CRM products that promise less admin stall, and how we help you avoid it.

BUILD AI IN
“The agent will see data it shouldn't.”
Agents act as the signed-in user, through the same role and tenant checks as your UI and API. They cannot see or change what that user cannot, and every call is logged.
“We can't tell if a release made it worse.”
We build test sets from real customer tasks, score every release and block changes that reduce accuracy, the same way unit tests block regressions.
“AI will eat our margins.”
Every agent action records model, tokens and cost by tenant, so you know what each task taken off a user costs, can price the feature, set limits per plan and protect margins.
BUILD WITH AI
“Payroll and tax rules have to be exact.”
Pay, tax and statutory rules run as configurable rules in code, tested against past pay runs, never left to a model's guess, so HR checks a pay run instead of correcting it. People approve every pay run.
“It won't fit Salesforce or our other systems.”
We connect through their APIs and app frameworks, sandbox first, with least-privilege access, and keep your existing data model, so nobody re-enters data in two places.
“Once it's live, we won't know what it did.”
We trace each agent run, alert on failures and unusual cost, and give your support team a view of what the agent did for each customer, so nobody has to piece it together by hand.
You inherit

Building blocks already built.

Already built
Agent runtime that calls your API with the user's permissionsin production
Tenant-level usage metering and cost limitsin production
Evaluation suite and release gatein production
Run tracing and monitoring dashboardin production
Human approval step for sensitive actionsin production
Connectors to common CRM, ERP and ticketing systemsin production
Your enterprise software (HR, CRM) productstarts on these, not from scratch

The team around it

Product designerShapes how people and agents work together
AI engineerAgents, prompts, evaluations
ArchitectRecords, integrations, security
DevelopersBuild it with AI, keep it maintainable
TestersSpecs and tests that prove it works
Domain expertKnows your industry's rules and edge cases
Proof

HR and CRM software we have built, and the admin it took away.

Noida Metro Rail Corporation (NMRC), India1,000+employees moved off paper forms and files onto one HR platform, across 21 metro stations

Payroll, leave, attendance and personnel records ran on paper forms, files and spreadsheets, with no central employee record. A cloud HRMS covering 14 HR processes from recruitment to exit: payroll posted to Tally, a self-service portal for every station, role-based access and an audit trail on every transaction.

Read the case study →
American Embassy School, New Delhi65%+less time to process payroll

US and Indian staff were paid in USD and INR under different tax rules, and the old system couldn't handle both. A payroll system for both currencies with US and Indian tax reports, loan management, bank letters and a no-code panel so HR changes pay rules without developers.

Read the case study →
V2 Retail, India10,000+employees across 300+ stores on one platform, with payroll off spreadsheets

Each store kept its own staff records, attendance had no central view and payroll ran on spreadsheets. A custom HRMS with 15 modules: employee records, geofenced attendance, leave, payroll with statutory deductions, recruitment, incentives and exit, with dashboards for five roles.

Read the case study →
A UAE-based HRMS providerresult to confirm

Every new customer of its white-label HRMS needed manual code changes and a full redeploy before its HR team could start. A modular rebuild with builders for objects, forms, screens and reports, an automated release pipeline, and modules for leave, payroll, attendance and loans. The client later asked us to add AI features.

Read the case study →
A leading Indian online furniture retailer30%higher workflow efficiency once cases were assigned and escalated automatically, 15% fewer order-management errors

After moving to Salesforce, service cases were still assigned by hand, service calls failed and SLA commitments were missed. Automatic case assignment and routing by location, SLA escalations and alerts, fixes to order-management timeouts, and case reporting.

Read the case study →
A US customer-engagement platform>95%accuracy across 1M+ accounts

Its Salesforce database of over a million accounts was full of incomplete, wrong and outdated records. Firmographic and content enrichment with checked company hierarchies, done inside the client's own systems over VPN so the data never left them.

Read the case study →

Results as reported in our client case studies.

Which repeated task would your product take over first? Name the admin your users repeat every week: approvals, records, reports or follow-ups. After a short discovery you get a plan with dates for the first release, measured against the time it takes today.

Tap one: Ask Unthinkable starts your plan from there. Or say it in your own words ↓

How to choose

How to choose who builds enterprise software (hr, crm) products.

The four checks that matter most here, and how we meet them.

Has built this before

We have built hr, payroll & performance, custom crm and salesforce, ai agents inside b2b saas products for NMRC, American Embassy School, V2 Retail, Chennai Petroleum (CPCL).

Often elsewhereOften new to your industry, or know it only from a requirements document.

Understands how your business works

We start from your customers' journey and also help hr, payroll & crm run, so we know the workflows your product must fit.

Often elsewhereConsultants know it but don't build; software companies build what is written.

Agents that act, not a chat window

For example, in hr, payroll & performance: an assistant answers policy and payslip questions, fills in leave, claims and appraisal forms, and flags pay-run errors before payday, so they are not corrected after. HR approves every pay run and decision.

Often elsewhereUsually a chatbot or a dashboard added on top.

Gets past the prototype

The architecture is set before the prototype and AI writes inside it, with the rules that must be exact here (Pay, tax and statutory rules run as configurable rules in code; connected to attendance, accounting and banking systems; every change logged) running as code, so feature twenty ships as fast as feature one.

Often elsewhereA fast prototype that stalls on inconsistent architecture.

The full comparison: us, other software companies, in-house teams and consultants →

Ways to work

Start the way that suits you.

Time & materials

A senior AI team billed for time, within an agreed monthly budget. Even on time and materials the work is tied to your business objective: every plan starts from it, and every report shows the number next to the hours.

One fixed outcome

One agreed outcome at a fixed price, with acceptance criteria, including the measure, written up front.

Dedicated team

A complete senior AI team, monthly, working as part of yours and reviewed each quarter on what shipped and what moved.

Linked to the result

Where the gain can be measured from your own systems: a reduced implementation fee plus a percentage of the expected gain, for a set period.

How we work and price →

Straight answers

Questions HR and CRM software teams ask.

How do I choose a company to build enterprise software (hr, crm) products?

Check six things: that they get you past the prototype to a product that works, without slowing down; that they design the architecture before the code; that they deliver agents that act inside your systems, not just a chat window or a report; that they understand your business and your customers' journey; that they talk about your business problem, not the technology; and that you own everything built. Unthinkable is an AI-first software engineering company with a consultant's understanding of your business, and does all six, and works on both sides of your market (we also help hr, payroll & crm run). Work like this includes Noida Metro Rail Corporation (NMRC), India and American Embassy School, New Delhi.

How do I add AI agents to my SaaS product?

Start with one admin task your users repeat often and can describe in a sentence. We expose the actions it needs from your API as tools, run the agent with the signed-in user's permissions, add approvals for sensitive steps and build a test suite from real examples. The first agent feature comes with a dated plan after discovery, then more tasks reuse the same foundation.

Is SaaS dead now that AI agents can do the work?

No, but its value is moving. Customers still need your data model, workflows, permissions and integrations; agents need them too. What changes is that users expect the product to do the admin for them, not only record it. SaaS companies that put agents on top of their own data and rules are well placed, because that foundation is hard for a generic agent to copy.

How do you stop an AI agent from accessing data it should not see?

The agent never gets more access than the user it acts for. It calls your existing API with that user's token, so the same role and tenant checks apply. Sensitive actions need explicit confirmation, and every call is logged with who, what and why. We test for prompt injection and cross-tenant access before release, and keep those tests in the release gate.

How much does it cost to run AI features in a SaaS product?

It depends on the task, the model and usage, so we measure it rather than guess. We meter every agent action by tenant with model, tokens and cost, then reduce it with caching, smaller models for simple steps and limits per plan. Typical actions cost fractions of a cent to a few cents, which lets you price the feature with a known margin.

How do you test AI features before release?

With an evaluation suite, like unit tests for behaviour. We collect real tasks with expected results, score each release for accuracy, safety and cost, and block releases that make things worse. After launch, failed or corrected runs become new test cases. This is what lets your team change prompts or switch models without guessing what broke.

Can you turn our single-customer software into multi-tenant SaaS?

Yes, usually without stopping the current product. Unthinkable separates each customer's data by tenant, moves configuration out of code, adds plans, billing and usage metering, and migrates customers one at a time, with tests that check no tenant can see another's data. Whether customers share one database or large ones get their own is decided with you, from security needs and running cost.

Which technology stack is best for a SaaS product?

Usually the one your team already runs and can hire for. Unthinkable works in your existing stack and recommends a change only where it blocks what you need, such as multi-tenancy, scale or AI features. For a new product we choose with you, often from common, well-supported options such as React, Node.js or Python and PostgreSQL on AWS, Azure or Google Cloud, so you are never tied to us.

Ask Unthinkable

Which admin task will your product take off its users first?

Tell us about your product and the admin your users repeat most. Our AI assistant, Ask Unthinkable, helps you define your requirements, then connects you with a senior AI engineer.

Send us a note instead

Required

Thanks. A senior AI engineer will reply to you.
u
Ask UnthinkableClarity first, then the right team
Your conversation

An AI agent. This prototype uses prepared answers and sends nothing.