DCS-PIS · the intelligence layer of your business

Business intelligence that thinks ahead.

One model of your business, read from every system you run — predicting what happens next, deciding what to do, and acting as far as you allow.

Contact salesSee it work
  1. Data
  2. Understand
  3. Predict
  4. Explain
  5. Decide
  6. Act
  7. Learn

Example workspace — not a customer result

One engine for the whole business.

It reads everything, reasons over all of it at once, and acts only as far as you allow.

Understands the whole business

Every system you run, read into one typed model — customers, orders, campaigns, people.

Decides what is worth doing

One ranked action, what it costs, why, and who has to say yes before it happens.

Acts — and measures it

Agents carry it out as far as you allow, and every result is checked against a group left alone.

Copilot

Ask the business anything.

Plain words in, an answer out — every figure in it traced to the record it came from.

Watch the Copilot

Business graph

Every system, one model.

Commerce, customers, accounts, people and conversations, read into one typed graph — no migration, no new system of record.

See the platform

Agents

Agents that ask before they act.

Sixteen agents, each on a governance level. The ones that reach your customers stop and wait for a person.

Meet the agents

See it work

Ask a question.
Watch the work get done.

Two short films of the product at its job: the Copilot answering from the business's own records, and an agent carrying the answer out — as far as you allowed, and no further.

01 · Copilot17s
02 · Agents16s

Then hand the next step to an agent, which waits for your approval.

What the films show

Ask it anything — every figure comes with its source

  1. Ask in plain words. No query language, no dashboard to find.
  2. It reads the workspace's own records before it says anything.
  3. Every figure in the answer is a chip that traces to where it came from.
  4. Until an experiment says otherwise, it tells you what moves together — not what caused what.
  5. Open a figure and see exactly what it was computed from.
  6. Then hand the next step to an agent, which waits for your approval.

An agent carries it out — only as far as you allow

  1. An agent has a goal, a fixed set of tools, and a level it cannot go past.
  2. It reads the group the prediction flagged…
  3. …and drafts the message. Drafting sends nothing.
  4. Contacting customers is Level 3, so it stops here and asks a person.
  5. One approval, and it carries the run out — holding one in ten back, at random.
  6. It books the check that will measure the message, and writes every step down.

The gap

Your business already has the data.
What is missing is the intelligence.

The layer between the data and the decision is, in most companies, still a person, a spreadsheet and a week.

How it works today

Days · by hand
CRMAdsFinanceSupport
Exported to a spreadsheetFour versions of the same customer
Hand-off
A dashboardA picture of what already happened
Interpret
A decision, from memoryAnd then acted on, by hand — if at all
Guess

With DCS-PIS

Every morning · on its own
CRMAdsFinanceSupport
One model of the businessEvery system read into one graph, scored at 06:00
A prediction, with its reasonsWho is likely to leave, and what moved them
A decision — approved, then measuredCarried out at your level, checked against a held-back group
  1. What happened, up to yesterday
    What happens next, scored this morning
  2. A chart, and the work of reading it
    The drivers that moved it, named
  3. Correlation, presented as insight
    Cause — only once an experiment concludes
  4. A conclusion you reach yourself
    One ranked action, costed, with who it reaches
  5. An action you go and take
    An agent that carries it out, at your level
  6. No way to tell whether it worked
    The measured lift, against a held-back group

The platform

An intelligence layer,
not another system of record.

Nothing here asks you to move your data somewhere new or to run your business out of a different tool. It reads what you already have, builds one model of it, and returns decisions to the systems you already use.

Your systems

  • Commerce & paymentsShopify · Razorpay
  • Customers & pipelineimportHubSpot · Salesforce · Zoho CRM
  • AccountsZoho Books
  • PeopleKeka · RazorpayX Payroll
  • ConversationsSlack · your mail server · Exotel
  • OperationsInventory · ERP, over REST
  • Anything with an APIREST · webhooks · CSV import
DDCS · PISScoring

One model of the business

Every connected system, read into one typed graph.

  • Customers
  • Orders & payments
  • Campaigns
  • Events
  • Business memory
  • Your metrics

Models over it

  • Churn
  • Lifetime value
  • Uplift
  • Anomaly
  • Time to event

Re-scored every morning, with each model's calibration recorded and its drift watched.

What it does

  • UnderstandsOne graph; change measured against each customer's own pattern
  • PredictsEvery customer scored each morning, with the drivers
  • ExplainsAssociated or caused — labelled, never blurred
  • DecidesOne ranked action, what it costs, who approves
  • ActsAgents at the level you set, and no further
  • LearnsOutcomes measured against a held-back group

  • One boundary, enforced below the application

    Every tenant-scoped table carries its isolation policy in the same migration that creates it, and the platform refuses to start if its database role could bypass them.

  • The model provider is a setting

    Every model call goes through one gateway, and application code makes no provider-specific call — so changing or failing over is configuration, not a rewrite.

  • What was proposed, approved and done is kept

    Each action carries what was proposed, the level it ran at, who approved it and what happened after. An AI decision nobody can reconstruct is one nobody can defend.

How PIS thinks

Intelligence is not a report.
It's a loop.

Follow one member of one studio through all seven stages — from the records four systems already hold to an outcome the product measured, and declined to overstate. Each stage is a layer of PIS, and each needs the one before it.

AvailableEvery stage below is in the product today.

01Your systems

Data

What does the business already know?

Orders, payments, check-ins, contacts — whatever the systems you run already record, read from where it lives on a schedule or the moment a webhook fires. Each record is matched to a person on email first and the provider's own id second.

  • Connectors
  • Webhooks
  • Identity matching
  • Event catalogue

Your systems

Four systems, one member

Example

  • Shopify14 orders
  • Razorpay12 payments
  • HubSpot1 contact
  • Check-ins · CSV156 visits
Member record

Priya R.

  • Matched on email
  • Then on each provider's own id

Four systems, one person — not four customers who happen to share a name.

Your systems stay the system of record. PIS reads from them — nothing has to be migrated to use it.

02Descriptive

Understand

What is happening?

Every record joins one typed model of the business, so a member, an invoice and a campaign are the same objects to everything that asks about them. Change is measured against each customer's own pattern — not an average, and not a fixed threshold.

  • Business graph
  • Metrics you define
  • Data health
  • Anomaly detection

Business graph

One typed model of the business

Example

Anomaly · her own baseline

Visits a week: 3.0 for a year, 0 for 24 days

03Predictive

Predict

What happens next?

Each morning every customer is scored against their own history, with the model's calibration recorded and its drift watched. A score on its own is a number; what matters is the moment it crosses the line where acting is worth more than waiting.

  • Churn risk
  • Lifetime value
  • Lead conversion
  • Time to event

Scored this morning

Her score crossed the line

Example

Churn · 30 days

Act above here

0.71
90 days agoThis morning · her own history
  1. Priya R.High71%
  2. Arjun M.Elevated64%
  3. Fatima S.Watch58%
  4. Deepak V.Watch52%
04Causal

Explain

Why is it happening?

Every score arrives with the drivers that moved it. Where the model cannot explain itself it says so rather than inventing a reason — and until an experiment concludes, a driver is associated with the outcome, never said to have caused it.

  • Score drivers
  • Revenue attribution
  • Uplift models
  • Evidence labels

Why the score moved

From her baseline to 0.71

Example

0.71

Act above 0.60

  1. Every member starts at0.12
  2. No visit in 24 days, after three a week for a year+0.33
  3. Stopped booking evening classes+0.17
  4. Annual plan renews in 9 days+0.09
◇ Correlational

No experiment behind these yet, so PIS says associated with — never caused.

✓ Causal

Only once an experiment concludes. The last stage is how one does.

05Prescriptive

Decide

What should we do?

The finding becomes one action, ranked against every other action available — for her whole group, because that is where the value is still recoverable. It carries what it costs, who it reaches, how it will be measured and who has to approve it.

  • Ranked actions
  • Cost and reach
  • Simulation
  • Approval route

Decision engine

What to do, for her group of 83

Example

  1. 1Re-engagement messageWhatsApp · ₹83 · 1 in 10 held backRecommended
  2. 2Personal-training offerIn person · costs more per member reached
  3. 3Wait a weekNo cost · some scores recover on their own
Reaches
83 members
Cost
₹83
Measured by
Retention at 30 days
Against
The held-back group
ReviewApproveLevel 3 — waits for the owner
06Governed

Act

Can the system carry it out?

Within the level you set for it, and no further. Contacting a customer is Level 3, so the retention agent drafts and waits for the owner. When it goes, one member in 10 is held back at random — and ten kinds of action are out of every agent's reach, permanently.

  • Agent runtime
  • Governance ladder
  • Approval gate
  • Held-back group

Agent runtime

It waits, then it goes — minus one in ten

Example

  1. L0Observe
  2. L1Suggest
  3. L2Act
  4. L3Act & spendRetention agent
  5. L4Restricted
Approved by the owner
07Measured

Learn

Did it work — and what does that teach?

What the contacted group did, minus what the held-back group did: the only number that can be attributed to the action. Both groups need 30 known outcomes before it is reported, and until then PIS says not yet. Every outcome is kept against the score that predicted it.

  • Holdout comparison
  • Sample floor
  • Calibration
  • Drift

Thirty days later

Measured — and not yet reported

Example

Day 0Renewals · both armsDay 30
Contacted
75 outcomes✓ Clears 30
Held back
8 outcomesNeeds 30

No lift is reported yet.

A difference from 8 people is noise with a decimal point. The number appears when both arms clear the floor.

Every outcome is kept against the score that predicted it — calibration is checked, drift is caught.

And the outcome is data. The loop starts again, knowing one more thing.

Ask the business

A chart is not an answer.

Any tool will show you that revenue fell. The question underneath it is what changed, whether it will continue, what to do, and who has to approve that — and the chart answers none of those.

Ask PIS

Example — not a customer result

Why did revenue decline this month?

Answer
The decline is concentrated in one segment, not across the base — repeat purchase rate in it fell while new-customer revenue held.
Evidence
Orders, customers and campaign records for the period, each figure linked to the query that produced it
Drivers
  • Repeat purchase rate in the segment
  • Discount depth on the segment's last campaign
  • Median days between orders, lengthening
Prediction
If the segment's order interval keeps lengthening at the current rate, the shortfall widens next month.
Recommendation
Treat the segment, not the base. A win-back to the segment costs a fraction of a base-wide campaign and reaches the customers whose behaviour actually changed.
Action
Marketing agent drafts the campaign · waits for approval

Measurement

How we know it worked.

Revenue that arrived after a campaign is easy to show. What would have arrived without it is the only number worth acting on.

The method

A group is left alone on purpose.

90% contacted 10% held back

Outcome · same window

Contacted

Held back

Δ The lift — the only part the campaign can claim.

  1. 01Hold back10% of the audience, at random, sent nothing.
  2. 02Follow bothThe same outcome, over the same window.
  3. 03SubtractWhat is left is what the campaign caused.

Below the sample, no number

Every figure says what kind it is.

Win-back campaign · renewals

Not reported yet Measured
Outcomes resolving30 per arm
Measured
Backed by a held-back group.
Attributed
Moved with the action — not proof it caused it.
Predicted
A forecast. Never counted as revenue.

Agents

From intelligence
to execution.

A recommendation nobody acts on is a report with extra steps. Agents read the same model of the business the rest of the platform does, and carry out what you have allowed them to — no further.

Retention agentL3Running

Re-engagement for the evening group

Approved once, then carried out end to end

Example

  1. Read the evening group83 members, scored this morningdone
  2. Drafted the messageWhatsApp, for the owner to reviewdone
  3. Asked for approvalLevel 3 — it contacts customersdone
  4. Approved by the ownerOnce, for the whole rundone
  5. Sending75 contacted · 8 held back at randomNow
  6. Outcome checkDay 30, against the held-back groupBooked

Agents at work

Example

6 of 16 agents active this morning

  1. Executive wrote the daily brief — attendance fell in one group

    L006:00

  2. Analyst explained the drop: visits, bookings, a renewal date

    L006:04

  3. Sales drafted follow-ups for 4 stalled deals, for review

    L106:10

  4. CRM proposed merging 2 duplicate members

    L106:12

  5. Advertising recommended moving spend off a saturated campaign

    L106:15

  6. Retention is waiting for approval to message 83 members

    L306:20

L0 reads and reports L1 recommends L3 waits for a person

Sixteen agents. One loop. Every one on a leash you set.

Each runs the same six steps around the same core, and ships at a governance level from the ladder below. Most of them cannot change anything at all.

DPIS
  1. 1 Observe
  2. 2 Reason
  3. 3 Plan
  4. 4 Act
  5. 5 Verify
  6. 6 Learn

Executive & analysis

  • ExecutiveThe daily brief: what changed and what it meansL0
  • StrategyProgress against goals, and where to concentrateL0
  • AnalystAnswers “why did this move?”L0
  • ResearchInvestigates accounts and segments from your historyL0

Revenue

  • SalesReviews pipeline, drafts follow-ups for stalled dealsL1
  • RevenueFinds expansion and recovery opportunitiesL1
  • CRMDuplicate detection, missing fields, stale ownershipL1

Customer

  • Customer successWatches account health, proposes retentionL1
  • RetentionActs on churn risk — inside an approval gateL3

Marketing

  • MarketingSegments audiences and drafts campaigns for reviewL1
  • AdvertisingMonitors ad performance, recommends budget changesL1

Operations

  • OperationsFinds bottlenecks, delays and underused resourcesL1
  • AutomationProposes workflows for patterns it observesL1

Finance & people

  • FinanceCash, receivables and abnormal expensesL0
  • HRWorkforce analysis. Never touches employment decisionsL0

Data

  • Data qualityWatches the health of what everything else depends onL0

How far each may go

  • 7ship at L0 — read and report, and change nothing
  • 8ship at L1 — write a recommendation for a person
  • 1ships at L3 — contacts a customer, and waits for you

Of 16, none above the level you set.

Governance

AI that knows when to act —
and when to ask.

Every agent ships at a level on this ladder, and nothing moves up a rung on its own. The top rung is not a setting: it is a list of things no agent can do.

  1. L0

    OBSERVE

    Reads only. Changes nothing.

    Seven agents — Executive, Strategy, Analyst…

  2. L1

    SUGGEST

    Writes a recommendation for a person to act on.

    Eight agents — Sales, Revenue, CRM…

  3. L2

    ACT

    Takes a small, reversible action within its budget.

    No agent by default — granted per action

  4. L3

    ACT & SPEND

    Contacts a customer or spends money. Needs your approval.

    Retention

  5. L4

    RESTRICTED

    Never taken automatically, with or without approval.

    Ten actions — no agent, ever

Waiting for approval

Example

Advertising agent · recommends

Move ₹2,50,000 from Campaign A to Campaign B

Why
Campaign A's audience is saturated — its reach has flattened while its spend has not.
Level
3 · it spends money, so a person decides
Measured by
Conversions, against a held-back group
If rejected
Nothing moves; the recommendation is kept on record

Whatever the answer, it is written to the audit trail — who decided, when, what was proposed, and what happened after.

Trust, built into the platform.

Not settings you remember to turn on — the way it is built.

Every customer's data, walled off

Row-level security in the database itself. The platform will not even start as a role that could step around it.

Plugs into what you already run

Commerce, CRM, accounts, people and conversations — connected, not migrated.

Nothing happens off the record

Every decision, approval and action is written down — and the platform cannot edit or delete what it wrote.

Backed up, and rehearsed

Encrypted before it leaves the host, nightly and continuously — and the restore is practised, not assumed.

Capabilities

Ten capabilities.
One model of your business.

Every one reads the same graph, so a customer the retention agent is worried about is the same customer on the revenue screen and in the experiment that measures it.

Customer intelligence · the screens in the product

Sector intelligence

One engine. Fourteen ways of reading a business.

The engine never changes. A sector pack tells it what a customer is called, what leaving looks like, and which numbers matter — and every screen speaks that business's language.

01 · The same signals, every business

  • Last activity
  • Payments
  • Usage and visits
  • Plan and renewal
  • Support

02 · The Fitness & Wellness pack

customer
Member
leaving
no attendance in window
window
21 days
revenue
Membership Revenue
pipeline
Renewals Due

03 · What your team reads

Members likely to leave

Example

Flagged on no attendance in window · 21-day window

  • Active Members

  • Member Retention

  • Attendance Rate

  • Class Fill Rate

The actual interface

Every number says what kind of number it is.

These are the product’s own components, not pictures of them. The card below is the one an owner approves an action with; the panel beside it is the one that refuses to print a figure it cannot stand behind.

Example approval — not a customer’s data

Retention Campaign

L3 Act & spend

Average attendance across this group has fallen 31% over four weeks, and renewals fall due within a fortnight.

Affects 83

What it does, to whom, at what cost, under which governance level, and how anyone will later tell whether it worked — before the button, not after.

Example figures — not a customer’s results

Revenue recovered

₹4,12,000

95% range 0.041 to 0.098

From 1,840 observations

Incremental

Campaign uplift

Not enough observations yet — 18 of the 30 needed.

No estimate yet

The second panel is the one that matters. Most products would print a zero there.

Example timeline — not a customer’s history

  1. Renewed

  2. Attendance recovered

  3. Delivered

  4. Re-engagement message

    Approved by the owner before sending.

  5. Risk assessed as high

  6. Attendance declining

Who it is for

Built for the people
who make the calls.

Example

Command Center

Customers
862
Predicted to leave
41
Value at risk
₹4.9L
Awaiting approval
1

● Evening-class attendance down 31% in four weeks

CEO

See the whole business, not twelve dashboards.

The platform

Example

New-year offer · 30 days

Contacted
Held back

Measured — the gap is what the campaign caused

CMO

Stop optimising for clicks. Optimise for profit.

How measurement works

Example

Pipeline · stalled 14+ days

  • Northwind renewalDrafted
  • Studio 42 upgradeDrafted
  • Coastline expansionDrafted

Sales agent · L1 — follow-ups for review, never sent

CRO

Turn pipeline data into pipeline action.

The agents at work

The whole platform

One operating system for the business.
Intelligence at its core.

The systems you run, one model of the business, the intelligence that reasons over it, the agents that act on it — and the applications your teams work in. One graph under all of it.

  1. 05 · Where your teams work

    Applications

    • CRM
    • Sales
    • Marketing
    • Ads
    • Customer success
    • Finance
    • HR
    • Operations
    • Copilot
    • Automation
    • Experiments

  2. 04 · What carries the work out

    Sixteen agents, governed

    • Executive
    • Strategy
    • Analyst
    • Research
    • Finance
    • HR
    • Data quality
    • Sales
    • Revenue
    • CRM
    • Customer success
    • Marketing
    • Advertising
    • Operations
    • Automation
    • Retention
    Ten actions stay with people, always

  3. 03 · What reasons

    DCS-PIS — the intelligence layer

    • Understands
    • Predicts
    • Explains
    • Decides
    • Acts
    • Learns

  4. 02 · What it knows

    One business graph

    • Customers
    • Orders & payments
    • Campaigns
    • Events
    • Business memory
    • Your metrics

  5. 01 · What it reads

    The systems you already run

    • Shopify
    • Razorpay
    • HubSpot
    • Salesforce
    • Zoho CRM
    • Zoho Books
    • Keka
    • RazorpayX Payroll
    • Slack
    • Exotel
    • ERP over REST
    • Webhooks
    • CSV import

Talk to us

Tell us what your business is trying to work out. We read every one of these.

From prediction to action

Build a business that sees
what is coming.

DCS-PIS connects your systems, builds one model of the business, predicts what is likely to happen next, and puts agents to work — with measured outcomes and a person in control wherever it matters.

Contact salesSign in

Or write to the team directly at contact-us@datachondria.com.